Cancer-associated fibroblasts produce matrix-bound vesicles that influence endothelial cell function

Intercellular communication between different cell types in solid tumors contributes to tumor growth and metastatic dissemination. The secretome of cancer-associated fibroblasts (CAFs) plays major roles in these processes. Using human mammary CAFs, we showed that CAFs with a myofibroblast phenotype released extracellular vesicles that transferred proteins to endothelial cells (ECs) that affected their interaction with immune cells. Mass spectrometry–based proteomics identified proteins transferred from CAFs to ECs, which included plasma membrane receptors. Using THY1 as an example of a transferred plasma membrane–bound protein, we showed that CAF-derived proteins increased the adhesion of a monocyte cell line to ECs. CAFs produced high amounts of matrix-bound EVs, which were the primary vehicles of protein transfer. Hence, our work paves the way for future studies that investigate how CAF-derived matrix-bound EVs influence tumor pathology by regulating the function of neighboring cancer, stromal, and immune cells. Protein transfer through matrix-bound vesicles from CAFs enhances monocyte adhesion to endothelial cells. Editor’s summary Cancer-associated fibroblasts promote tumor growth, in part, by releasing extracellular vesicles, which can carry proteins to cells in the tumor microenvironment. Santi et al. investigated intercellular communication between endothelial cells in blood vessels and cancer-associated fibroblasts isolated from patients with breast cancer. Endothelial cells in vitro and in vivo took up proteins from extracellular vesicles, specifically matrix-bound vesicles, released by cancer-associated fibroblasts. Uptake of the membrane glycoprotein THY1 from cancer-associated fibroblasts increased the adhesion of monocytes to endothelial cells. Cancer-associated fibroblasts that released the most matrix-bound vesicles resembled myofibroblasts. Thus, identifying the proteins released by myofibroblast-like cancer-associated fibroblasts that alter endothelial cell function could yield potential targets for disrupting this intercellular communication. —Wei Wong


INTRODUCTION
Communication between cells is fundamental for the physiological function of tissues (1,2), and alterations can cause diseases and determine their severity (3)(4)(5).In solid tumors, intercellular communication involves cancer cells and neighboring cells of the tumor microenvironment (TME) and modulates tumor growth and metastatic dissemination.The TME is a highly heterogeneous and dynamic compartment that comprises pathological and activated immune and stromal cells, which include cancer-associated fibroblasts (CAFs) and endothelial cells (ECs) (6,7).
CAFs are highly secretory cells and represent the bulk of the stroma of solid tumors with a desmoplastic reaction, such as breast cancer (8,9), and are thus a considerable source of chemical signals that can affect the behavior of cancer, immune, and stromal cells.For these reasons, CAFs have been defined as "architects of cancer pathogenesis" (10) or as "architects of stroma remodeling" (6).The repertoire of chemical signals produced by CAFs includes growth factors; cytokines; noncoding RNAs; components of the extracellular matrix (ECM) and ECM remodeling enzymes, which regulate invasion, proliferation, and chemoresistance of cancer cells, as well as blood vessel formation and the recruitment and function of immune cells (6,(10)(11)(12)(13)(14). CAFs carry out these different functions by acquiring distinct but interchangeable states (15).Myofibroblastlike CAFs (myCAFs) and inflammatory CAFs (iCAFs) are the two main subtypes that have been described in tumors, including breast cancer (15,16).myCAFs are responsible for ECM production and remodeling and have immunosuppressive functions, whereas iCAFs have an immunomodulatory role (15,17).In addition to these mechanisms of paracrine cross-talk, CAFs transfer various nutrients (18)(19)(20), proteins, lipids (21,22), and even entire mitochondria (23,24) to cancer cells, which use these CAF-derived resources to support their own growth and motility.
The intercellular transfer of cell surface and intracellular proteins has been extensively documented between immune cells.The physiological role and functional consequences of this phenomenon are still unclear but may help to regulate the immune response (25)(26)(27)(28)(29).So far, few papers have examined the ability of pathologically activated fibroblasts to transfer their own proteins to cancer cells.These papers showed that the transfer of proteins from CAFs to cancer cells occurs through large extracellular vesicles (EVs) that CAFs release in the conditioned medium (CM) and that it supports cancer cell proliferation (21) and migration (22).There remain several open questions about the protein transfer ability of CAFs.Do other stromal cells also receive CAF-derived proteins?If so, what is the biological relevance of this intercellular protein transfer?
EVs are lipid bilayer-enclosed particles that mediate cell-cell communication by transferring proteins, lipids, and nucleic acids between cells.In accordance with the MISEV (minimal information for studies of extracellular vesicles) guidelines, EVs are classified on the basis of their size as small (diameter, <200 nm) and medium/ large (diameter, >220 nm) (30).Medium/large EVs directly bud from the plasma membrane (ectosomes), whereas small EVs originate from either the endosomal compartment (exosomes) or the plasma membrane (ectosomes) (30,31).EVs that transfer biological material between cells are typically found in cell-derived CM (CM-EVs) (14); however, EVs can be embedded within the ECM of decellularized tissues and of murine NIH-3T3 fibroblast cell cultures (32,33).These matrix-bound vesicles (MBVs) have a similar shape and morphology to CM-EVs but differ in lipid and microRNA content (33).MBVs are biologically active (32,34); however, their protein composition and role in intercellular protein transfer have not yet been reported.
Tumor blood vessels are typically embedded within the tumor stroma; therefore, we have investigated whether CAFs use intercellular protein transfer to influence the function of ECs.Using CAFs isolated from patients with breast cancer as donors and human ECs as recipient cells, we have identified a specific pool of proteins that CAFs transfer to ECs and, using Thy-1 membrane glycoprotein (THY1) as an example, we provide proof of principle that they can be functional in the ECs.Moreover, we found that CAFs deliver proteins principally through MBVs and that CAFs expressing myCAF markers are the main donors of proteins to ECs.

CAFs transfer proteins to ECs
To study whether mammary CAFs transfer proteins to ECs, we used several CAF lines that we have isolated from patients with breast cancer (pCAFs).These pCAFs express the mesenchymal marker vimentin (fig.S1A) (35) but are negative for markers of epithelial, endothelial, and immune cells (fig.S1B).Our laboratory has previously characterized the pCAF2 and pCAF3 lines (35).To study the process of protein transfer between cells and its biological relevance, we used different culturing methods (fig.S1C).
To monitor the transfer of proteins from pCAFs (donor cells) to human umbilical vein ECs (HUVECs, recipient cells), we fluorescently labeled the pCAF proteome with carboxyfluorescein diacetate succinimidyl ester (CFSE), a dye that covalently binds to amino groups.Microscopy analysis showed that HUVECs became fluorescent after being cocultured for 24 hours with CFSE-labeled pCAFs, indicating that pCAFs transferred some of their proteins to HUVECs (Fig. 1, A and B, and fig.S1D).
Using the same CFSE-based labeling method, we quantified the intercellular transfer of proteins by flow cytometry, which confirmed that HUVECs acquire fluorescent signals upon coculture with CAFs (Fig. 1, A, C to F).The quantity of transferred proteins depended on the number of donor cells, and it increased in accordance with the ratio between pCAFs and ECs (Fig. 1, A, C, and D).The shift of the CFSE peak of cocultured HUVECs compared with monoculture showed that the vast majority of the HUVECs received pCAF proteins, indicating that this is a commonly occurring event (Fig. 1, A  and D).Conversely, HUVECs transferred very low amounts of proteins to pCAFs (Fig. 1, A and E) or to other HUVECs (Fig. 1, A and  F).In addition, pCAFs had a much higher protein transfer ability compared with MDA-MB-231 cells, which are aggressive breast cancer cells (Fig. 1, A and F).These results indicate that pCAFs and HUVECs do not mutually exchange proteins and that CAFs are major protein donors.
Once we established that pCAFs transfer proteins to HUVECs in vitro, we sought to assess whether this mechanism also occurred in vivo.For this purpose, we used the C.FVB-tg(Acta2-DsRed)1RK1/J mouse model (36), also known as α-SMA-RFP.This model expressed the red fluorescent protein (RFP) in cells expressing the alpha-smooth muscle actin gene (Acta2, whose product is α-SMA protein).Because α-SMA is a widely used CAF marker (7,15,37), we used the α-SMA-RFP model to monitor the transfer of RFP from Acta2expressing cells to ECs in experimental pulmonary metastases, as a mean of protein transfer from CAFs to ECs.Murine breast cancer cells (4T1) were injected in the tail veins of α-SMA-RFP mice and, after 3 weeks, we dissected tumor-containing lungs (fig.S1E) and analyzed single-cell suspensions by flow cytometry.We used α-SMA-RFP mice that had not been injected with 4T1 cells as control to measure whether RFP could be transferred to the endothelium in the absence of Acta2-expressing CAFs (for example by perivascular cells, such as pericytes, which also express Acta2).Flow cytometry analysis measured a significant increase of RFP + ECs in mice with lung metastases compared with the control (Fig. 1G).To confirm these results, we imaged fixed precision cut lung slices with 4T1 metastases from α-SMA-RFP mice (Fig. 1H) and non-RFP-expressing control mice (fig.S1F).The three-dimensional (3D) reconstruction of tumor sections, which were stained for CD31 to visualize ECs, showed RFP + endothelium in the lung metastases of these mice (Fig. 1, H and I).Overall, our data provide evidence that CAFs communicate with ECs through the transfer of proteins in vitro and in vivo.

CAFs transfer plasma membrane receptors to ECs
To identify proteins that pCAFs transfer to HUVECs, we used a mass spectrometry (MS)-based trans-stable-isotope labeling by amino acids in cell culture (trans-SILAC) proteomic approach (28).First, we labeled the proteome of pCAFs with the heavy isotopolog of arginine and lysine and cocultured them with unlabeled HUVECs for 4 or 24 hours.Then, we sorted the HUVECs and analyzed their proteome by MS (Fig. 2A).We quantified 808 and 1062 heavy-labeled proteins in at least three of five biological replicates at 4 and 24-hour time points, respectively (Fig. 2B and data file S1).Of these, 698 proteins were common to both time points (Fig. 2B).Gene Ontology Cellular Component (GOCC) term analysis of the proteins transferred from CAFs to the HUVECs revealed enrichment in lipid bilayer-enclosed vesicles, endoplasmic reticulum (ER), ER-Golgi intermediate compartment, and macromolecular complexes, including focal adhesions, cell junctions, ribonucleoprotein particles, and proteasome (Fig. 2C).The high number of common proteins and the consistency of the top 10 enriched GO terms between the two time points indicate that there is a continuous transfer of proteins over time from CAFs to ECs in culture.Moreover, the association of these proteins with particular subcellular compartments suggests that mammary CAFs transfer selected protein subsets.
Cancer and immune cells use EVs to transfer functional plasma membrane proteins to ECs (38,39).These types of proteins are highly relevant because they may alter the function of the endothelium, including its interactions with surrounding cells.Therefore, we focused our analysis on plasma membrane receptors and membrane-bound ligands.We found that most of the transferred membrane proteins were involved in immune response, cell locomotion, and cell-cell and cell-matrix adhesion (Fig. 2D) (40)(41)(42)(43)(44), corroborating the idea that CAF-derived proteins may have important implications on the functions of the tumor vasculature.To select for proteins that provided the biggest changes in the HUVEC proteome, we determined the contribution of each transferred protein to the corresponding endogenous protein in the HUVECs and referred to this value as "exogenous fraction." The exogenous fraction ranges between 0 and 1, and the closer the value is to 1, the more the pCAF protein contributes to the endothelial counterpart (Fig. 2E and data file S1).CAF-derived THY1 was the protein with the highest contribution to the HUVEC proteome, with an exogenous fraction of 0.78 and 0.54 after 4 and 24 hours of coculture, respectively (Fig. 2E and data file S1).CD44 antigen (CD44) also contributed highly with an exogenous fraction of 0.46 and 0.35 at 4 and 24 hours, respectively, and then integrin beta-3 (ITGB3), with an exogenous fraction of 0.21 at 24 hours of coculture.The exogenous fraction for all the other receptors and ligands was  lower than 0.15 (Fig. 2E and data file S1).Overall, these results indicate that mammary CAF-derived receptors and ligands can quantitatively modify the proteome of the HUVECs.

CAF-derived THY1 induces functional changes in ECs
To confirm that the THY1 detected in HUVECs was derived from pCAFs, rather than being expressed by HUVECs when cocultured with them, we measured THY1 transcript in HUVECs in monoculture and after 24 hours of coculture with pCAFs.Using pCAFs as the control for THY1 expressing cells, we found that THY1 mRNA amount did not increase significantly in cocultured HUVECs compared to the monoculture (Fig. 3, A and B).In addition, flow cytometry analysis confirmed the transfer of THY1 from pCAFs to HUVECs (Fig. 3, A and C to E).Although THY1 was not present at the surface of HUVECs in monoculture, after 24 hours of coculture with pCAFs, most of the HUVECs positively stained for THY1 (Fig. 3, A, D, and  E).Moreover, pCAFs silenced for THY1 (Fig. 3C) transferred significantly less THY1 to HUVECs (Fig. 3, A, D, and E), whereas the total amount of transferred proteins was not affected (Fig. 3, A and F).THY1 (also known as CD90) is a glycophosphatidylinositolanchored protein that localizes on the extracellular side of the plasma membrane of cells and that binds to cancer cells and leukocytes through plasma membrane receptors (45,46).In inflammatory disease, the recruitment of immune cells requires their physical interaction with the endothelium mediated by adhesion molecules (47,48).THY1 expressed on the endothelium participates in this process by interacting with its binding partners present on the leukocyte surface, such as CD11b [also referred to as integrin alpha-M (ITGAM)] (49-51).To assess the function of pCAF-derived THY1, we measured leukocyte adhesion to HUVECs when cocultured with pCAFs silenced or not for THY1.Specifically, we used the human monocyte cell line THP-1 that expresses several THY1 binding partners (fig.S2A and data file S2).Microscopy analysis of the cocultures showed that significantly fewer monocytes adhered to HUVECs when cocultured with THY1-silenced pCAFs compared with control coculture (siCtrl), supporting the functionality of THY1 on the HUVEC surface (Fig. 3, A and G, and fig.S2, B to E).Hence, CAF-derived THY1 endows HUVECs with additional cell-cell adhesion properties.
We explored whether CAF-derived THY1 is also involved in leukocyte recruitment in breast cancer.4T1 cells and pCAFs expressing shCtrl or shTHY1 (fig.S2F) were orthotopically injected in the mammary fat pads of BALB/c mice, and after 2 weeks, we used immunohistochemical staining to determine the presence and the location of the CD11b + immune infiltrate.We focused on CD11b + cells within and proximate to tumor blood vessels (Fig. 3, H to L), to exclude resident CD11b + populations such as macrophages or dendritic cells.We found that the amount of CD11b + staining within the tissue in close proximity to veins was higher in tumors containing shCtrl pCAFs compared with those containing shTHY1 pCAFs (Fig. 3H).In contrast, tumors containing shCtrl pCAFs showed a lower amount of CD11b + staining within the blood vessels compared with tumors containing shTHY1 pCAFs (Fig. 3I), suggesting that leukocytes are less able to extravasate in tumors with shTHY1 pCAFs.The tumor weight was similar between the two conditions (fig.S2G); this result is in line with other studies showing that THY1 is a marker of tumorpromoting CAFs, rather than an effector of this phenotype (52,53).Overall, these results suggest that the transfer of THY1 from pCAFs to ECs can promote their interaction with CD11b + cells, thus influencing immune cell recruitment to tumor sites.

Different types of CAF-derived EVs contain the proteins transferred to ECs
Next, we investigated how pCAFs transfer their proteins to HUVECs.Our data showed that a high number of transferred proteins belonged to lipid bilayer-enclosed vesicles (Fig. 2C), supporting that EVs can be a major route of intercellular protein transfer.CM-EVs are involved in protein transfer (21,22,48,54,55), but the role of MBVs has not been investigated.We isolated EVs from both the CM and ECM of pCAFs (Fig. 4A).Electron microscopy analysis showed that the two EV types had a similar morphology (Fig. 4B).Nanoparticle tracking analysis showed that the diameter of both types of EVs ranged between 50 and 350 nm (Fig. 4C).However, the amount and size distribution differed between the two EV types.Those in the CM mainly included small particles with a diameter between 50 and 150 nm, whereas MBVs mostly consisted of large EVs, with major peaks at 150 and 200 nm (Fig. 4, C and D, and fig.S3A).
We molecularly characterized pCAF-derived EVs using MS proteomics (data file S3).This analysis confirmed that both types of particles contain common EV markers, such as the tetraspanins CD63, CD81, and CD9 (30) and syntenin-1 (SDCBP) (56), but also highlighted differences, such as the relative abundance of some EV markers and the presence of ADP-ribosylation factor 6 and tumor susceptibility gene 101 protein only in MBVs and CM-EVs, respectively (Fig. 4E and data file S3).Hence, our data have identified distinct traits of CM-EVs and MBVs.
We next compared the proteome of pCAF-derived EVs with the proteome of large-medium and small EVs of three publicly available datasets (fig.S3, B and C) (57)(58)(59).For each dataset, we selected proteins unique to each EV subpopulation and those with significantly different abundance between the two subpopulations.Then, we matched this subset to EV proteins whose abundance was significantly different between CM-EVs and MBVs (data file S3).This analysis showed that proteins typically found in large-medium EVs were generally more abundant in MBVs.In contrast, proteins typically found in small EVs were more abundant in CM-EVs (fig.S3C).This observation was consistent across the three datasets (fig.S3C).Furthermore, proteins identified only in CM-EVs displayed enrichment for endosome-related GOCC terms (fig.S3D), and endosomes are one of the documented intracellular origins of small EVs (fig.S3B).Conversely, unique proteins in MBVs displayed enrichment in GOCC terms associated with plasma membrane, cytosol, ER, and mitochondria (fig.S3D), which are expected in large-medium EVs because of their biogenesis (30).
The majority of pCAF proteins transferred to ECs during coculture were identified in both EV types (Fig. 4F and data files S1 and S3), and their abundance positively correlated to the amount measured in the EVs (Fig. 4G and data files S1 and S3).The majority of the transferred plasma membrane receptors and membrane-bound ligands, including THY1, were more abundant overall in the MBVs (Fig. 4H and data file S3).Overall, these results support that each extracellular compartment contains different subsets of EVs, which carry the proteins transferred from CAFs to HUVECs.

MBVs have a major role in protein transfer to ECs
Next, we measured whether pCAF-derived CM-EVs and MBVs could transfer proteins to ECs.Because these EV types exist in different extracellular sites, we measured protein transfer when CAFs were cocultured in physical contact (direct coculture) or not (indirect coculture) with HUVECs (Fig. 5A).In direct coculture, HUVECs were exposed to both types of EVs, whereas in indirect coculture were exposed to CM-EVs only (Fig. 5A).The amount of transferred proteins in direct cocultures was more than twofold higher compared with indirect coculture (Fig. 5, A and B).We used the same coculture conditions to measure THY1 transfer from pCAFs to HUVECs.As for total proteins, the transfer of THY1 mainly occurred when cells were in direct culture (Fig. 5, A and C).These results suggest that MBVs have increased protein transfer ability compared with CM-EVs.The matrix produced by CAFs influences many cell functions (35,60), leading us to evaluate whether it could also sustain the ability of MBVs to act as vehicles for proteins.We compared the ability of EVs to transfer proteins when they were coated on pCAF-derived matrix compared with when they were coated on gelatin or on the matrix produced by patient-derived normal fibroblasts (pNFs) (fig.S4A), which has different composition and mechanical properties from pCAF-derived matrix (35).We found that compared with CM-EVs, MBVs retained the ability to transfer more proteins, including THY1, when they were coated on gelatin or on fibroblast-derived matrix before HUVECs were plated on top (fig.S4, A to C).The MBV-mediated transfer of THY1 to HUVECs was enhanced by the presence of the matrix compared with gelatin (fig.S4, A and B).However, MBVs transferred the same amounts of proteins whether they were coated on the matrix produced by pNFs or pCAFs (fig.S4, A and C); the same results were observed when the matrices were pre-treated with CM-EVs (fig.S4, A  and C).These data indicate that pNF-and pCAF-derived matrices have common features that promote the EV-mediated protein transfer, but the matrix alone is not able to account for the different efficiency in protein transfer between CM-EVs and MBVs.
To confirm the different role of CM-EVs and MBVs in protein transfer, we also added them directly into the HUVEC culture medium.In line with the previous results, HUVECs received significantly more proteins when treated with MBVs than with CM-EVs, although the difference was less pronounced (Fig. 5, A and D).MBVs still had a higher protein transfer ability compared with CM-EVs even when HUVECs were treated with equal numbers of the two EV types (fig.S4, D and E).Moreover, upon treatment with MBVs, fivefold more HUVECs positively stained for THY1 compared with when treated with CM-EVs, and THY1 amount was twofold higher (Fig. 5,  A and E).
To confirm that pCAF-derived THY1 transferred by MBVs mediates monocyte adhesion to HUVECs, HUVECs were treated with equal numbers of pCAF-derived CM-EVs or MBVs isolated from pCAFs either silenced or not for THY1 (Fig. 5, A and F).Microscopy analysis showed that HUVECs treated with MBVs bound a higher number of monocytes compared with untreated HUVECs or HUVECs that were treated with CM-EVs (Fig. 5F).However, the MBV proadhesive effect was entirely lost when these EVs were isolated from THY1-silenced pCAFs (Fig. 5F).Together, our data provide evidence that MBVs are a major vehicle for protein transfer from mammary CAFs to HUVECs and that they can influence HUVEC function.
α-SMA high TNFRSF12A high CAFs are the major donors of proteins to ECs Human normal fibroblasts (NFs) activated upon treatment with CM of prostate and melanoma cancer cells transfer more proteins compared with untreated fibroblasts (21).Therefore, we compared protein transfer of our mammary CAFs with their matched NFs isolated from the same patient (pNFs), derived from macroscopically healthy tissue adjacent to the tumor.We found that pCAFs transferred more proteins to HUVECs than pNFs (Fig. 6, A and B) and confirmed this result using microvascular ECs (MVECs) (fig.S5, A and B).Because we showed that EVs are involved in protein transfer, we compared the amounts of EVs released by pCAFs and pNFs.Nanoparticle tracking analysis showed that pCAFs deposited significantly more mediumlarge EVs in the ECM than their NF counterparts (Fig. 4, A, C, and D, and fig.S3A).In contrast, pNFs and pCAFs released EVs of similar size and quantity into the CM (Fig. 4, C and D, and fig.S3A).HUVECs treated with CAF-derived MBVs received more proteins than when treated with MBVs produced by pNFs (Fig. 6, A and C).Moreover, CM-EVs and MBVs secreted by pNFs transferred a comparable amount of proteins to HUVECs (Fig. 6, A and D).However, the different protein transfer ability between pNF-and pCAF-derived MBVs was greatly reduced when HUVECs were treated with equal numbers of MBVs (fig.S4, D and E).Together, these results suggest that CAFs transfer more proteins because they produce more MBVs.Despite this, MBVs isolated from pCAFs and pNFs were molecularly and functionally different.Significantly more monocytes adhered to HUVECs treated with MBVs isolated from pCAFs compared with pNFs, even though equal numbers of EVs were used (Fig. 5F).
Our pCAF lines transferred different amounts of proteins to ECs (fig.S5, A and C), raising the question of whether all CAFs can transfer proteins.To address this question, we first measured the correlation between the abundance of common CAF markers in our pCAF lines (data file S4), including ACTA2, prolyl endopeptidase FAP (FAP), ITGB1, dipeptidyl peptidase 4 (DPP4), platelet-derived growth factor receptor alpha and beta (PDGFRA and PDGFRB), caveolin-1 (CAV1) and protein S100-A4 (S100A4, also known as FSP-1), with their protein transfer ability.We found that the amount of proteins transferred by fibroblasts significantly correlated only with ACTA2 protein abundance (Fig. 6E).Microscopy analysis for α-SMA in our pCAF lines confirmed the proteomic data showing that the pCAF1 line, which transferred the most proteins to ECs (fig.S5, A and C), contained more cells with high α-SMA protein amount than pCAF3 and pCAF4 lines (Fig. 7, A and B).Similarly, Western blot analysis showed that α-SMA protein amount was higher in pCAF1 line compared with pCAF3 and pCAF4 lines (Fig. 7C).To assess whether pCAFs expressing high or low protein amount of α-SMA had different protein transfer abilities, we needed to identify cell surface proteins to sort the two living subpopulations for functional assays.To achieve this, we analyzed CAFs sorted according to high and low α-SMA protein amount by MS proteomics (fig.S6A).Principal components analysis of 2080 proteins quantified across the three pCAF lines separated the α-SMA low and α-SMA high subpopulations (fig.S6B and data file S5).Moreover, 67 proteins showed difference in abundance between α-SMA low and α-SMA high subpopulations in at least two of the three pCAF lines (fig.S6C and data file S5), and among those, there were seven cell surface receptors (Fig. 7D).We followed up on the tumor necrosis factor receptor superfamily member 12A (TNFRSF12A, also known as FN14, TweakR, or CD266), because its abundance differed substantially between α-SMA low and α-SMA high CAFs, and it was a good candidate for cell sorting (data file S5).Immunofluorescence staining for α-SMA confirmed that there were more α-SMA high cells in TNFRSF12A high sorted pCAFs than in TNFRSF12A low pCAFs (fig.S6D).On average, TNFRSF12A high pCAFs transferred double the amount of proteins to cocultured HUVECs than TNFRSF12A low pCAFs (Fig. 7, E and F), including THY1 (Fig. 7,  E and G).Moreover, TNFRSF12A low pCAFs had a protein transfer ability similar to that of their NF counterpart (Fig. 7, E and F).Consistent with our findings that identified the MBVs as a major vehicle for protein transfer, HUVECs treated with TNFRSF12A high pCAF-derived MBVs received more proteins than when treated with CM-EVs isolated from the same CAF subpopulation (Fig. 7H).Instead, the protein transfer ability of CM-EVs and MBVs isolated from TNFRSF12A low pCAFs was similar and lower than the amount of proteins transferred by TNFRSF12A high pCAF-derived MBVs (Fig. 7H).The different protein transfer ability between MBVs from TNFRSF12A high and TNFRSF12A low pCAFs did not depend on evident differences in the matrices produced by the two CAF subpopulations, because the amount of fibrillar collagen (CNA35) and fibronectin was similar between the two (Fig. 7I).Hence, α-SMA high mammary CAFs enriched using the transmembrane receptor TNFRSF12A have enhanced ability to transfer proteins to ECs. α-SMA high TNFRSF12A high CAFs express high amounts of myofibroblast markers α-SMA high CAFs are typically those referred to as myCAFs, whereas α-SMA low are typically iCAFs (15,16).Therefore, we investigated the expression of other myCAF and iCAF markers in our CAF subpopulations.We sorted TNFRSF12A high and TNFRSF12A low pCAFs, expanded them in culture, and assessed the expression of CAF markers by reverse transcription quantitative polymerase chain reaction (RT-qPCR).This analysis confirmed that TNFRSF12A high pCAFs expressed higher amounts of ACTA2 and other genes highly expressed in mammary myCAFs (7,16,61), such as those encoding collagen alpha-1 (I) chain (COL1A1) and transgelin (TAGLN), compared with TNFRSF12A low pCAFs (fig.S7, A and B).Conversely, we did not detect significant differences in mRNA amounts of stromal cellderived factor 1 (SDF1, also known as C-X-C motif chemokine 12 or CXCL12) and interleukin-6 (IL6), which are highly expressed in mammary iCAFs (7, 16, 61) (fig.S7, A and B).Similarly, decorin (DCN), which is expressed at similar levels in all CAFs (7, 61), had similar mRNA amounts in our two sorted populations (fig.S7, A and  B).These data suggest that high amounts of the TNFRSF12A receptor are found in CAFs with the myCAF phenotype.Consistent with this observation, in two publicly available single-cell RNA sequencing datasets of CAFs isolated from patients with breast cancer (7, 16), we found that both ACTA2 and TNFRSF12A mRNA amounts were high in the subpopulation defined by the authors as myCAFs (Fig. 8, A and B).In addition, immunofluorescence staining of tumor tissue sections from patients with breast cancer confirmed the presence of TNFRSF12A + and α-SMA + CAFs in the stroma and showed that these cells could be found in close proximity to blood vessels (Fig. 8C and fig.S8).Hence, enhanced CAF-EC communication based on protein transfer is distinctive of those CAFs with a myofibroblast-like phenotype.

DISCUSSION
Using CAFs isolated from patients with breast cancer, we have found that CAFs with a myofibroblastic-like phenotype transfer high amounts of proteins to the surrounding endothelium.The transfer of proteins mainly occurs through MBVs.Using THY1 as an example of transferred protein, our work also shows that transferred proteins can influence the phenotype of the endothelium (Fig. 8D).CAFs in different states secrete distinct subsets and amounts of soluble factors and ECM components, which determine their functions in the tumor (15).Therefore, understanding the heterogeneity of CAFs by associating their states with specific biological functions is fundamental to designing drugs for cancer treatment.CAFs use EVs ( 14) and intercellular transfer of proteins to affect the function of neighboring cells in vitro (21,22).Although these mechanisms have been originally described between CAFs and cancer cells, using various MS-based proteomic approaches we showed that ECs also receive proteins from CAFs in vitro.Moreover, we provide evidence that this process may occur in vivo.What is the fate of these proteins in recipient cells?It has been suggested that the fate of the EV cargo depends on the mechanism of EV internalization (62,63).For example, EV cargo can be directed toward lysosomes for degradation or can escape it (62)(63)(64).Here, we showed that CAF-derived proteins could be functional in recipient cells: ECs can receive plasma membrane proteins from CAFs, most of which are involved in migration and cell-cell or cell-matrix adhesion.Among them, THY1 enhanced the ability of ECs to interact physically with THP-1 monocytes in vitro and supported the recruitment of CD11b + leukocytes in orthotopic 4T1 tumors.The ability of CAFs to affect the extravasation of CD11b + leukocytes has the potential to influence the composition of the immune microenvironment, whose variation plays a crucial role in determining the efficacy of the therapeutic strategies (65).Proteins transferred by cancer cells or CAFs influence the phenotype of recipient cells (21,22,48,54,55).For example, PC3 human prostate cancer cells increase the migration of prostate cancer and benign prostatic hyperplasia cells through the exosomal transfer of integrin αvβ3 (54).Moreover, HeyA8 and TYK-nu human epithelial ovarian cancer cells induce an invasive mesenchymal phenotype of human peritoneal mesothelial cells through the transfer of exosomal CD44 (55), and activated human prostate and dermal fibroblasts support migration of cancer cells by transferring galectin-1 through ectosomes (22).We found that CAFs transferred all these proteins to ECs and that CD44 and ITGB3 were among the receptors with the highest exogenous fraction.This suggests that, in addition to THY1, CD44 and ITGB3 may influence EC functions.
The ECM is an important source of signals that actively regulate tumor progression.Its structure and composition, including ECMassociated proteins such as growth factors, influence many aspects of tumor pathology (66)(67)(68).It is now evident that EVs have an essential role in ECM biology: EVs can be functional components of the ECM (32,34), and CM-EVs control ECM deposition (69) and remodeling (70).Our work provides evidence that CAFs deposit EVs in the matrix and that MBVs play a key role as vehicles for intercellular protein transfer.MBVs and CM-EVs contain EVs of different sizes, and our MS proteomic characterization identified several differences between these two EV types, in accordance with previous work (33).In particular, our findings support the concept that CM-EVs and MBVs may have different intracellular origins, specifically endosomal for CM-EVs and plasma membrane for MBVs, and provide evidence that these two subsets of vesicles have distinct functions.MBVs can deliver a greater amount of proteins to ECs and promote the adhesion of monocytes to ECs compared with CM-EVs.We showed that the ability of MBVs to deliver more proteins to recipient cells was due to their distinct characteristics rather than their extracellular location.This specific function of MBVs could depend on their larger size enabling them to transport a higher amount of proteins or on the presence of surface receptors that make their uptake easier compared with CM-EVs.Our data also suggest that CM-EVs and MBVs are heterogeneous populations that may interact differently with recipient cells.EVs can influence recipient cell function through ligand-receptor interactions without being internalized (63,71), and EV cargo can also be rereleased in the extracellular space (72).The existence of multiple ways of interaction between EVs and recipient cells is also indicated by our proteomic analysis, which identifies EV proteins that are not transferred to ECs, although we cannot exclude that some transferred proteins are below levels detectable by MS or degraded in lysosomes (63).Further study will be required to elucidate the precise mechanism by which MBVs can transfer more protein.
We showed that CAFs secreted more MBVs than their normal-like fibroblast counterpart.This explains why CAFs have a greater capacity to transfer proteins to the endothelium.On this basis, we argue that protein transfer from fibroblasts is a phenomenon predominant in pathological conditions.Our work also indicates that CAF-derived MBVs may play unique roles in altering tumors locally, in addition to being a source of systemic signals, which is a classical function associated with tumor-derived EVs (73).
MyCAFs deposit most of the tumor ECM and contribute to its remodeling (16,68).Our work indicates that there are additional ways through which myCAFs can influence cellular functions.We showed that myCAFs were the major donors of proteins to ECs and that myCAF-derived MBVs had a key role in the process of protein transfer.However, we found no evidence that the matrix produced by my-CAFs supported MBV performance, further confirming that MBVs transferred more proteins because of other distinct properties.These findings suggest that the increased protein transfer from myCAFs to ECs could depend on their ability to deposit different types and/or amounts of MBVs in the matrix.Another interesting aspect that emerged from our work is that MBVs could be a source of nutrients for tumor and stromal cells.Cancer cells (20,74,75) and ECs (76) take up proteins and amino acids from the extracellular milieu and use them for macromolecule synthesis and to modulate redox homeostasis (77).In recipient cells, CAF-derived proteins may undergo proteolytic degradation and supply amino acids that contribute to biosynthetic and bioenergetics processes.Hence, we would speculate that MBVs could be an additional mechanism for local exchange of nutrients within the TME (78).
Last, we have identified TNFRSF12A as potential cell surface marker of mammary CAFs with myofibroblast-like phenotype to be added to the small panel of plasma membrane proteins that can be used to isolate these cells for functional characterization (15).After the exclusion of epithelial cells, immune cells, ECs, and pericytes, TNFRSF12A allows the direct selection of CAFs expressing high levels of myofibroblast markers.This result is in line with another study showing that TNFRSF12A specifically belongs to the myCAF transcriptional profile (16).
In conclusion, our work has identified that myCAFs can transfer functional proteins to ECs through MBVs.Our work paves the way for studies seeking to explore whether and how other transferred plasma membrane proteins can modify EC phenotype and how this affects the function of the tumor vasculature and influences tumor development and progression in vivo.These results will inform on whether targeting the production of MBVs should be further investigated as a strategy to oppose cancer.As an example, the targeting of MBV production could impair THY1 transfer, therefore affecting the composition of the immune microenvironment and the response to therapy.More work should also be done to understand MBV biogenesis and function, a mechanism that so far has been largely overlooked.
A limitation of our study is that we used mice that express RFP under the control of the Acta2 promoter to test the transfer of proteins in vivo, and we cannot exclude that Acta2-expressing cells other than CAFs, such as pericytes surrounding the endothelium, also transfer proteins to ECs (79,80).However, CAFs expand during tumor progression, and we showed that the amount of transferred proteins to ECs increased proportionally with CAF number in vitro.Hence, we think that the majority of RFP found in the tumor endothelium could derive from CAFs.Another limitation is that we cannot exclude that CAFs had transferred RFP mRNA, which has then been translated into protein in the endothelium.Hence, additional studies are needed to further prove that CAFs transfer proteins to the endothelium in vivo, for example using MetRS* mice (81).

MATERIALS AND METHODS
Cell culture pCAFs and pNFs were isolated at the CRUK Scotland Institute from women with breast cancer and immortalized as previously described (35).Samples were obtained through NHS Greater Glasgow and Clyde Bio-repository.Patients agreed with the use of their tissue samples for research.Unless otherwise stated, pCAFs and pNFs used for the experiments are pCAF1 and pNF1.cCAFs and cNFs were provided by A. Orimo (Juntendo University, Tokyo) (35,82).Human MDA-MB-231 and mouse 4T1 breast cancer cells were purchased from American Type Culture Collection (ATCC).Luciferase-expressing 4T1 cells were provided by G. Inman (University of Glasgow and CRUK Scotland Institute, Glasgow).CAFs, NFs, MDA-MB-231 cells, and 4T1 cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 2 mM glutamine, and 1% penicillin/ streptomycin (Life Technologies, Thermo Fisher Scientific).pCAFs and pNFs were cultured on dishes coated with collagen I from rat tail (12 μg/ ml; Gibco, Thermo Fisher Scientific).HUVECs were isolated from donors using previously described methods (67).HMVECs (100-05a) were purchased from Sigma-Aldrich.HUVECs and HMVECs were cultured on 1% gelatin-coated dishes in endothelial cell growth medium-2 (EGM-2) or microvascular EGM-2 (EGM-2 MV, Lonza), respectively.
EV isolation pCAFs or pNFs were plated in serum-free DMEM.After 48 hours, EVs were collected from both the CM and the matrix.The CM was collected; then, the culture plate was washed with phosphate-buffered saline (PBS), and the EVs from the matrix were detached using Accutase (Sigma-Aldrich) and collected in DMEM supplemented with 0.5% FBS, which had been previously ultracentrifuged at 100,000g for 5 hours and filtered with a 0.2-μm filter to reduce the amount of serum EVs.Cells and debris were removed by centrifugation at 300g (4°C, 10 min) and 2000g (4°C, 30 min), and CM-EVs and MBVs were isolated by ultracentrifugation at 100,000g (4°C, 90 min).Pelleted EVs were resuspended in PBS and subjected to another step of ultracentrifugation at 100,000g (4°C, 90 min).EVs were collected in PBS and gently sonicated at 5-μm amplitude using a metal tip (Soniprep 150, MSE) three times for 5 s before using them for protein transfer experiments and nanoparticle tracking analysis and before sample preparation for electron microscopy.

Nanoparticle tracking analysis
CM-EVs and MBVs were isolated as described above from 1.5 × 10 5 pCAFs or pNFs, which were seeded in serum-free medium in 6-cm cell culture dishes.After isolation, the EVs were resuspended in 1 ml of PBS.EV size and concentration were determined using a Nano-Sight LM10 (Malvern Panalytical) and the NTA 3.1 software.Each measurement is the result of three acquisitions of 60 s.The camera level was set to 14 and the detection threshold to 4. The PBS used for EV isolation and collection was filtered with the 0.02-μm filter.

Intercellular protein and THY1 transfer
To measure the intercellular protein transfer in coculture, donor cells were labeled with 10 μM CellTrace CFSE (Life Technologies) in PBS for 20 min at 37°C.After at least 1 hour, donor cells were seeded.Once they were adhered, recipient cells were seeded in coculture with donor cells for 24 hours (also referred to as direct coculture).For indirect coculture, donor cells were plated on a glass coverslip, which was positioned upside down on the culture dish where recipient cells were seeded.The glass coverslip was placed above a polytetrafluoroethylene (PTFE) film ring with a thickness of 0.08 mm (Goodfellow Cambridge Ltd).After 24 hours of coculture, cells were detached with Accutase and resuspended in fluorescence-activated cell sorting (FACS) buffer (25 mM Hepes, 5 mM EDTA, 1% penicillin/streptomycin, 1% FBS in PBS), and the transfer of CFSE labeled proteins was analyzed by an Attune NxT flow cytometer (Thermo Fisher Scientific) and FlowJo software version 10.7.1.To measure THY1 transfer, cells were detached with Accutase after 24 hours of coculture, resuspended in FACS buffer, and incubated with allophycocyanin (APC) anti-THY1 antibody (1/100, [5E10] 328114 BioLegend, RRID:AB_893431) and human TruStain FcX (1/200, BioLegend) for 45 min on ice (100 μl/10 6 cells).4′,6-Diamidino-2-phenylindole (DAPI) (Sigma-Aldrich) was used as a live/dead marker.THY1 transfer was analyzed by an Attune NxT flow cytometer and FlowJo software version 10.7.1.In both protein and THY1 transfer experiments, donor cells were gated as CFSE high and recipient cells were gated as CFSE low .Unless otherwise stated, donor cells and recipient cells were plated at 2:1 ratio.The medium used for the coculture experiments was EGM-2 or EGM-2 MV depending on the EC type.
To measure the EV-mediated transfer of proteins, CM-EVs and MBVs were isolated from 1.5 × 10 5 pCAFs, pNFs, or sorted pCAFs and labeled with 10 μM CellTrace CFSE in PBS for 20 min at 37°C.After labeling, EVs were washed in PBS by two sequential steps of ultracentrifugation at 100,000g (4°C, 90 min).The whole amount of isolated EVs was used to treat 3 × 10 4 HUVECs in EGM-2.After 20 hours, HUVECs were detached and intercellular protein transfer was analyzed as above.In experiments in which recipient cells were treated with equal numbers of isolated CM-EVs or MBVs, vesicles were quantified by nanoparticle tracking analysis and used for the treatment.To measure the EV-mediated transfer of THY1, CM-EVs and MBVs were isolated from 5 × 10 4 pCAFs, and the whole amount was used to treat 2.5 × 10 4 HUVECs in EGM-2.After 3 hours, HUVECs were detached, stained for THY1, and analyzed as above.If not otherwise stated, all the protein and THY1 transfer experiments were performed on 1% gelatin-coated dishes.
To determine whether the ECM influences the transfer of proteins and THY1, decellularized ECM was prepared by seeding pNFs or pCAFs at 100% confluence on 0.2% gelatin, which was cross-linked using 1% glutaraldehyde for 7 days.ECM was decellularized with 20 mM NH 4 OH and 0.5% Triton X-100 (TX-100, Sigma-Aldrich) in PBS.CM-EVs and MBVs were isolated from 1.5 × 10 5 pCAFs and labeled as described above.After isolation, EVs were coated overnight on the decellularized ECM or on dishes previously coated with 1% gelatin.Unbound EVs were removed by washing with PBS, and 3 × 10 4 HUVECs were seeded on top in EGM-2.After 20 hours, HUVECs were detached and intercellular protein and THY1 transfer were analyzed as above.
In protein and THY1 transfer experiments, recipient cells that were seeded without donor cells (also referred to as monoculture condition) or that were untreated with EVs were used as control to determine the levels of autofluorescence.In addition, in THY1 transfer experiments, recipient cells seeded in monoculture were stained for THY1 to determine basal amounts.
To evaluate protein transfer by immunofluorescence, pCAFs were labeled with CellTrace CFSE as described above and seeded on glass coverslips in a 24-well plate.Once they adhered, ECs were seeded in coculture with pCAFs.After 24 hours, cells were fixed in 4% PFA.DAPI and Alexa 647 Phalloidin (1/100, Life Technologies) were used for nuclear and F-actin staining, respectively.HUVECs seeded in monoculture condition were used as control.Images were acquired with a Zeiss LSM 880 confocal microscope in Airyscan mode (Carl Zeiss, Plan-Apochromat 63×/1.4Oil DIC M27 objective, zoom 1.8, z-stacks of 5 to 9 μm, 28 to 48 slices).Images were Airyscan processed with Zen software (version 3.7) using default settings.The 3D reconstruction and analysis were performed using Imaris software (version 9.5, Bitplane, Oxford Instruments).

pCAF sorting
Cultured pCAFs were detached with Accumax solution (Sigma-Aldrich) and resuspended in FACS buffer.To sort pCAFs on the basis of α-SMA protein abundance, pCAFs were fixed and permeabilized by using the eBioscience Intracellular Fixation & Permeabilization Buffer Set (Life Technologies) according to the manufacturer's instructions.Cells were resuspended in permeabilization buffer (100 μl/10 6 cells) and incubated with anti-α-SMA antibody [1/1000, (1A4) ab7817, Abcam] for 1 hour and then with Alexa Fluor 488 or 647 secondary antibody (1/250, Life Technologies) supplemented with 2% donkey serum for 1 hour.Unstained pCAFs and pCAFs incubated with the secondary antibody only were used as controls.To sort pCAFs on the basis of TNFRSF12A protein abundance, 10 6 cells were incubated with BD Horizon BV421 anti-TNFRSF12A antibody (1/140, 565712 BD Biosciences, RRID:AB_2739337) and human TruStain FcX (1/200, BioLegend) in 100 μl of FACS buffer for 45 min on ice.Unstained pCAFs were used as control.pCAFs were sorted into α-SMA high and α-SMA low or TNFRSF12A high (10% of cells with the highest expression) and TNFRSF12A low (10% of cells with the lowest expression) using a BD FACSAria (BD Biosciences).

Collagen quantification in sorted pCAFs
TNFRSF12A high and TNFRSF12A low pCAFs were plated at confluence for 7 days.They were incubated with 1 μM of the fluorescent collagen binding CNA35-mCherry (83) for 1 hour, fixed in 4% PFA, and counterstained with DAPI (1/5000).Images were taken on a Zeiss LSM 710 confocal microscope, and collagen staining was quantified using ImageJ software.

Adhesion assay
To measure the binding of THP-1 cells to HUVECs that were directly cocultured with pCAFs, control and THY1-silenced pCAFs were labeled with 2.5 μM CellTracker Green CMFDA Dye (Life Technologies) in PBS for 25 min at 37°C, and 2 × 10 4 labeled pCAFs were seeded in each 1% gelatin-coated well of a 96-well plate.HUVECs were labeled with 1 μM CellTracker Deep Red Dye (Life Technologies) in PBS for 20 min at 37°C, and 4 × 10 4 labeled cells were seeded in coculture with pCAFs.After 24 hours of coculture, 8.5 × 10 3 THP-1 cells, which were labeled with 2 μM CellTracker Orange CMTMR Dye (Life Technologies) in PBS for 20 min at 37°C, were added to each well in M199 medium (Life Technologies) supplemented with 10% FBS.After 45 min, unbound THP-1 cells were removed by three washes in 1% BSA in PBS with calcium and magnesium (Sigma-Aldrich).Cells were fixed in 4% PFA and DAPI was used for nuclear staining.
To measure the binding of THP-1 cells to HUVECs that were treated with EVs, 1.5 × 10 5 control pCAFs, THY1-silenced pCAFs, and pNFs were seeded in serum-free medium in 6-cm cell culture dishes, and EVs were isolated as described above.CM-EVs and MBVs were isolated from control and THY1-silenced pCAFs, and MBVs only were isolated from pNFs.A total of 4 × 10 4 HUVECs were seeded in a 1% gelatin-coated well of a 96-well plate and treated overnight with 5 × 10 8 of isolated EVs.Unbound EVs were removed by washing the well with PBS with calcium and magnesium, and 8.5 × 10 3 THP-1 cells were added as described above.For each well, 25 to 45 images (adhesion assay in coculture conditions) or 77 images (adhesion assay after EV treatment) were acquired on an Opera Phenix high-content imaging system (20× objective and z-stacks of 2 μm for the adhesion assay in coculture conditions and 10× objective for the adhesion assay after EV treatment, PerkinElmer).Image analysis was performed using Harmony imaging analysis software (PerkinElmer, version 4.9).For each well, the number of THP-1 monocytes that bound HUVECs was averaged (adhesion assay in coculture condition) or summed (adhesion assay after EV treatment).In the adhesion assay in coculture conditions, only THP-1 cells overlapping the ECs at least for the 30% of their cellular body were counted.

MS proteomic analysis
For trans-SILAC experiments, heavy-labeled pCAFs were labeled with CellTracker Green CMFDA Dye as described above, and 1.5 × 10 6 pCAFs were seeded in a gelatin-coated 15-cm dish.After 16 hours, 7.5 × 10 5 HUVECs were seeded in coculture with pCAFs.HUVECs seeded without pCAFs (also referred to as monoculture condition) were used as control.After 4 and 24 hours, cells were detached with Accutase and resuspended in FACS buffer, and HUVECs were sorted as "CellTracker Green CMFDA Dye" negative cells by using a BD FACSAria.Sorted HUVECs were lysed in 6 M urea/2 M thiourea supplemented with 10 mM tris(2-carboxyethyl) phosphine (TCEP) and 40 mM chloroacetamide (CAA) in 75 mM NaCl and 50 mM tris-HCl (Sigma-Aldrich) and sonicated using a metal tip.Proteins (25 to 120 μg) were digested with trypsin and were fractionated using high-pH reverse-phase fractionation.Briefly, dried peptides were resuspended in 200 mM ammonium formate adjusted to pH 10 with ammonium hydroxide solution (Sigma-Aldrich).Then, peptides were loaded on pipette-tip columns of ReproSil-Pur 120 C18-AQ (5 μm) (Dr.Maisch HPLC GmbH), eluted in seven fractions using an increasing amount of acetonitrile, and analyzed by MS.
For the proteomic analysis of THP-1, α-SMA high and α-SMA low pCAFs, cells were washed three times in PBS and lysed in 6 M urea/2 M thiourea supplemented with 10 mM TCEP and 40 mM CAA in 75 mM NaCl and 50 mM tris-HCl and sonicated using a metal tip.Proteins (10 μg) were digested with trypsin, desalted using C18 StageTip (84) and analyzed by MS.
For the analysis of the pCAF1 proteome, cultured pCAFs were washed three times with PBS and cultured in serum-free DMEM.After 24 hours, cells were washed with PBS, lysed in 2% SDS with 1 mM DTT in 100 mM tris-HCl (pH 7.4), incubated at 95°C for 5 min, and sonicated using a metal tip.Tryptic peptides were generated from 150 μg of proteins using filter-aided sample preparation using filtration units with molecular weight (MW) cutoff of 30 kDa (85,86).Briefly, lysates were loaded on the filter units, incubated for 20 min with 55 mM iodoacetamide (IAA, Sigma-Aldrich) in 50 mM ammonium bicarbonate (pH 8.0), digested with trypsin, and eluted with 50 mM ammonium bicarbonate (pH 8.0) (Sigma-Aldrich).Peptides (60 μg) were fractionated using high-pH reverse-phase fractionation as described above and analyzed by MS.
For the analysis of the EV proteome, pCAFs were cultured in DMEM supplemented with 10% ultracentrifuged FBS.pCAFs were washed three times in PBS and cultured in serum-free DMEM.After 48 hours, CM-EVs and MBVs were collected as described above from 2 × 10 7 and 10 7 pCAFs, respectively.After ultracentrifugation, EVs were collected in 200 mM Hepes (pH 8.0), and 2,2,2-trifluoroethanol (TFE) was added 1:1 (Sigma-Aldrich).EVs were sonicated three times at 10-μm amplitude for 10 s (with 20 s on ice in between each sonication) and were incubated at 60°C for 1 hour at 1000 rpm and sonicated again.EV lysates were incubated with 10 mM TCEP and 40 mM CAA for 1 hour.TFE concentration was reduced to 10% by adding 200 mM Hepes, and EV lysates were digested with trypsin.Peptides were desalted using C18 StageTip, dried, resuspended in 200 mM Hepes, and incubated with 0.1 mg Tandem Mass Tag (TMTzero, Thermo Fisher Scientific) label reagent for 2 hours at 26°C and 450 rpm.Samples were dried, resuspended in 0.1% formic acid, acidified by adding trifluoroacetic acid (TFA; Sigma-Aldrich), desalted using C18 StageTip, and analyzed by MS.
For the proteomic analysis of pCAFs and pNFs, cultured cells were washed with PBS, lysed in 2% SDS in 100 mM tris-HCl (pH 7.4), incubated at 95°C for 5 min, sonicated using a metal tip, and centrifuged at 16,000g for 10 min.Lysates were mixed 1:1 with an internal standard composed of a mix of SILAC heavy-labeled cCAFs/cNFs.Protein lysates were mixed with NuPAGE LDS sample buffer (4×) and 1 mM DTT.Proteins were separated using 4 to 12% gradient Nu-PAGE Novex bis-tris gel, which then stained with Coomassie Blue.Gel lanes were cut into slices; proteins were in gel digested with trypsin (87); and peptides were desalted using C18 StageTip and analyzed by MS.

MS analysis with a Q Exactive HF (trans-SILAC experiment and pCAF1 proteome)
Each of the seven fractions was dried down and resuspended in 2% acetonitrile/0.1% TFA in water and separated by nanoscale C18 reverse-phase liquid chromatography performed on an EASY-nLC II 1200 coupled to a Q Exactive HF mass spectrometer (Thermo Fisher Scientific).Elution was carried out for a total run time duration of 65 min (fraction 1), 105 min (from fractions 2 to 5), and 135 min (fraction 6 and 7), using an optimized gradient.Peptides were eluted into a 50-cm (trans-SILAC) or 20-cm (pCAF1 proteome) fused silica emitter (New Objective Inc., Littleton, MA) packed in-house with ReproSil-Pur C18-AQ, 1.9-μm resin (Dr.Maisch HPLC GmbH).The emitter was kept at 50°C (trans-SILAC) or 35°C (pCAF1 proteome) by means of a column oven integrated into the nanoelectrospray ion source (Sonation).Eluting peptides were electrosprayed into the mass spectrometer using a nanoelectrospray ion source (Thermo Fisher Scientific).An active background ion reduction device (ABIRD, ESI Source Solutions) was used to decrease air contaminants signal level.Xcalibur software (Thermo Fisher Scientific) was used for data acquisition.Full scans over mass range of 375 to 1500 mass/charge ratio (m/z) were acquired at 60,000 resolution at 200 m/z.Multiply charged ions from two to five were selected through a 1.4-m/z window and fragmented.Higher-energy collisional dissociation fragmentation was performed on the 15 most intense ions using normalized collision energy of 27, and the resulting fragments were analyzed in the Orbitrap at 15,000 resolution, using a maximum injection time of 25 ms or a target value of 10 5 ions.Former target ions selected for MS/MS were dynamically excluded for 20 s.

MS analysis with an Orbitrap Fusion Lumos (THP-1 cells, sorted pCAF and EV proteomes)
Desalted peptides were separated by nanoscale C18 reverse-phase liquid chromatography performed on an EASY-nLC 1200 coupled to an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific).Elution was carried out for a total run time duration of 265 min (THP-1 and sorted pCAF proteome) or 135 min (EV proteome), using a binary gradient with buffer A (water) and B (80% acetonitrile), both containing 0.1% of formic acid.Peptide mixtures were separated at 300 nl/min flow, using a 50-cm fused silica emitter (New Objective Inc.) packed in-house with ReproSil-Pur C18-AQ, 1.9-μm resin (Dr.Maisch GmbH).The packed emitter was kept at 50°C by means of a column oven integrated into the nanoelectrospray ion source (Sonation).The eluting peptide solutions were electrosprayed into the mass spectrometer via a nanoelectrospray ion source (Sonation).An ABIRD (ESI Source Solutions) was used to decrease ambient contaminant signal level.Samples were acquired on an Orbitrap Fusion Lumos mass spectrometer.The mass spectrometer was operated in positive ion mode and used in data-dependent acquisition (DDA) mode.Advanced peak determination was turned on, and monoisotopic precursor selection was set to "peptide" mode.A full scan was acquired at a resolution of 120,000 (THP-1 and sorted pCAF proteome) or 60,000 (EV proteome) at 200 m/z, over mass range of 375 to 1500 m/z (THP-1 and sorted pCAF proteome) or 375 to 1400 m/z (EV proteome).The top 20 (THP-1 and sorted pCAF proteome) or 15 (EV proteome) most intense ions were selected using the quadrupole, fragmented in the ion routing multipole, and analyzed in the linear ion trap (THP-1 and sorted pCAF proteome) or analyzed in the Orbitrap at 15,000 resolution (EV proteome), using a maximum injection time of 35 ms (THP-1 and sorted pCAF proteome) or 125 ms (EV proteome) or a target value of 2 × 10 4 ions (THP-1 and sorted pCAF proteome) or 1.5 × 10 5 ions (EV proteome).Former target ions selected for MS/MS were dynamically excluded for 60 s (THP-1 and sorted pCAF proteome) or 30 s (EV proteome).

MS analysis with an Orbitrap elite (pNF and pCAF proteomes)
Digested peptides were separated by nanoscale C18 reverse-phase liquid chromatography performed on an EASY-nLC II (Thermo Scientific) coupled to a Linear Trap Quadrupole -Orbitrap Elite mass spectrometer (Thermo Fischer Scientific).Elution was carried out using a binary gradient with buffer A (water) and B (80% acetonitrile), both containing 0.1% of formic acid.Peptide mixtures were separated at a flow of 200 nl/min, using a 20-cm fused silica emitter (New Objective) packed in-house with ReproSil-Pur C18-AQ, 1.9-μm resin (Dr.Maisch GmbH) for a total duration of 255 min.Packed emitter was kept at 35°C by means of a column oven integrated into the nanoelectrospray ion source (Sonation).Eluting peptide solutions were automatically (online) electrosprayed into the mass spectrometer by a nanoelectrospray ion source (Sonation).An ABIRD was used to decrease ambient contaminant signal level.General mass spectrometric conditions of Linear Trap Quadrupole -Orbitrap Elite were as follows: spray voltage, 2.1 kV; ion transfer tube temperature, 200°C.The mass spectrometer was operated in positive ion mode and used in DDA mode.A full scan (FT-MS) was acquired at a target value of 1 × 10 6 ions with resolution R = 120,000 over mass range of 300 to 1650 amu.The top 10 most intense ions were selected for fragmentation in the linear ion trap using higher-energy collision dissociation using a maximum injection time of 150 ms or a target value of 4 × 10 4 ions.
Multiply charged ions from two to five charges having intensity greater than 40,000 counts were selected through a 3-amu window and fragmented using normalized collision energy of 30.Former target ions selected for MS/MS were dynamically excluded for 60 s.

MS proteomic data analysis
The. RAW files were processed with MaxQuant software (version 1.5.5.1 for the proteome analysis of pCAF/pNF proteome, version 1.6.3.3 for all the other experiments) (88) and searched with the Andromeda search engine.The following setting was used: minimal peptide length seven amino acids, trypsin-specific digestion mode with maximum of two missed cleavages, carbamidomethyl (C) as fixed modification, and oxidation (M) and acetylation (Protein N-term) as variable modifications.For the analysis of the EV proteome, TMTzero was added as fixed modification, and a maximum of four missed cleavages was allowed.Minimum peptide ratio count was set to 2, except for the trans-SILAC experiment and the analysis of the pCAF1 proteome, in which this parameter was set to 1. "Unique + razor" peptides were used for quantification in the analysis of the THP-1 and the pCAF and pNF proteomes; "unique" peptides were used for quantification in all the other experiments.The "match between runs" option was enabled for the analysis of pCAF/pNF proteome.For SILAC experiments, multiplicity was set to 2: light labels were Arg0 and Lys0; heavy labels were Arg10 and Lys8.Label-free quantification (LFQ) setting was enabled for all the other experiments.The false discovery rates at protein and peptide levels were set to 1%.
Perseus software (version 1.5.5.3 for the analysis of pCAF/pNF proteome and version 1.6.2.2 for all the other experiments) (89) was used for data analysis.Potential contaminants and reverse peptides and proteins only identified by a modification site were filtered out.Only proteins identified with at least one unique peptide were kept for the analysis.To define the transferred proteins in the trans-SILAC experiment, we selected proteins with a "ratio H/L count" value higher in HUVECs cocultured with CAFs compared with monoculture.In addition, we selected proteins with an intensity value in the heavy channel (intensity H) but not in the light one (intensity L).The exogenous fraction was calculated as 1 − [1/(x + 1)], where x is the "ratio H/L" value.The exogenous fraction of proteins with an "intensity H" value but not the "intensity L" one was set to 1.The proteins with an exogenous fraction in at least three of five biological replicates were selected.We filtered out proteins not identified in the pCAF1 proteome (donor cells).For the analysis of the THP-1 proteome, the intensity value of each protein was divided by the MW and transformed by log 2 .The adhesion molecules were selected on the basis of the Gene Ontology Biological Process (GOBP) category of cell adhesion (GO:0007155) and on the basis of the subcellular location's annotations retrieved from UniProt.For the analysis of the pCAF1, CM-EV, and MBV proteomes, the intensity value of each protein was divided by the MW and transformed by log 2 .For the analysis of the pCAF and pNF proteomes, the SILAC ratio was inverted, transformed by log 2 , and normalized by subtracting the median from each column.For the analysis of the α-SMA high and α-SMA low pCAF proteome, LFQ intensity was transformed by log 2 , three valid values were required for at least one pCAF1 or pCAF3 subpopulation, and one valid value was required for at least one pCAF4 subpopulation.Missing values were replaced from the normal distribution using the recommended setting in Perseus software and proteins with a fold change ≥1.5 and P ≤ 0.05 (two-tailed t test) in at least two of the three pCAF lines were selected.The z-score was calculated by row.The cell surface proteins were selected on the basis of the subcellular location's annotations retrieved from UniProt.

Proteomic datasets
EV proteomic data were downloaded from three publicly available datasets (57)(58)(59).For each dataset, we considered the proteins unique to each EV subpopulation and those with an abundance significantly different between the two subpopulations as statistically analyzed by the authors; except for (59), we considered proteins with at least a twofold change and P < 0.05.The selected proteins were matched by gene name with the proteins whose abundance was significantly different between CM-EVs and MBVs (two-tailed t test, P < 0.05).The z-score was calculated by row.
In vivo study of RFP transfer BALB/c C.FVB-tg(Acta2-DsRed)1RK1/J mice (JAX stock #031159, generated by R. Kalluri, University of Texas MD Anderson Cancer Center and provided by C. D. Madsen, Lund University) were used for the in vivo experiments.All mouse procedures were in accordance with ethical approval from University of Glasgow under the revised Animal Act 1986 (Scientific Procedures) and the EU Directive 2010/63/EU authorized through UK Home Office Approval (project license number 70/8645).For FACS analysis, 2.5 × 10 4 4T1 cells were resuspended in 100 μl of PBS and injected in the tail veins of 6-to 8-week-old RFP-expressing female mice.Littermate α-SMA-RFP female mice that had not been injected with 4T1 cells were used as control.Mice were culled 3 weeks after the injection.Lungs were collected, minced finely, and digested in prewarmed PBS (with calcium and magnesium) with collagenase A (2 mg/ml; Roche) for 1 hour on a rotating wheel at 37°C.The pieces of lung tissue were then passed through a 14G needle.Isolated cells were resuspended in M199 medium supplemented with 10% FBS, passed through a cell strainer (70 μm), and washed several times by centrifugation at 300g for 5 min.Cells were resuspended in FACS buffer and incubated with the following antibodies (1/100, BioLegend): CD31-Alexa Fluor 488 [(390) 102414, RRID:AB_493408] and CD45-APC/cyanine7 [(30-F11) 103116, RRID:AB_312981] and with mouse TruStain FcX (1/200, BioLegend) for 45 min on ice.DAPI was used as a live/dead marker.ECs were identified as CD31 + CD45 − cells.Cells were analyzed using an Attune NxT flow cytometer and FlowJo software version 10.7.1.
For immunofluorescence analysis, 2.5 × 10 4 4T1 cells were resuspended in 100 μl of PBS and injected in the tail veins of 4-to 5-month-old female mice expressing or not (control mice) RFP.Mice were culled 3 weeks after the injection.A small incision was made in the trachea, and 1 ml of 2% low-melting point agarose was introduced slowly into the lungs through a 22G needle.Lungs were excised and fixed in 4% PFA for 2 hours at 4°C.Then, lungs were sliced into 300-μmthick sections by using a vibrating microtome (Campden Instruments Ltd).Slices were permeabilized and blocked for 5 hours in PBS with 1% BSA, 10% normal goat serum (NGS, Sigma-Aldrich), 0.3% TX-100, and 0.05% sodium azide (VWR International), incubated overnight with anti-CD31 antibody [1/200, (2H8) MA3105 Invitrogen, RRID:AB_223592], and then for 3 hours within Alexa Fluor 647 secondary antibody (Jackson ImmunoResearch Labs) diluted in the same buffer.DAPI was used for nuclear staining.Slices were fixed in 4% PFA for 30 min, incubated for 45 min with Ce3D clearing solution, and mounted with the Ce3D solution (90).Images were acquired with a Zeiss LSM 880 confocal microscope in Airyscan mode (Carl Zeiss, Plan-Apochromat 63×/1.4Oil DIC M27 objective, zoom 1.8, z-stacks of 10 to 14.5 μm, 41 to 58 slices).Images were Airyscanprocessed in Zen software (version 3.7) using default settings.Imaris software (version 9.5) was used to generate the 3D images and to calculate both the distance between RFP and surface CD31 and the volume of RFP surface.

Hematoxylin and eosin staining of lungs
Hematoxylin and eosin (H&E) staining was performed on 4-μm formalin-fixed paraffin-embedded sections (FFPE) that had previously been heated at 60°C for 2 hours.H&E staining was performed on a Leica autostainer (ST5020).FFPE sections were dewaxed and taken through graded alcohols before being stained with Haem Z (RBA-4201-00A, CellPath) for 13 min.Sections were washed in tap water, differentiated in 1% acid alcohol (three dips), and washed in tap water, and the nuclei were blued in Scotts tap water substitute (made in-house).After washing, the sections were placed in Putt's eosin (made in-house) for 3 min.To complete H&E staining, sections were rinsed in tap water, dehydrated through graded ethanols, and placed in xylene.The stained sections were coverslipped in xylene using DPX mountant (SEA-1300-00A, CellPath).

Orthotopic 4T1 mammary tumor experiments
pCAFs were transduced with a lentiviral vector encoding shCtrl (SHC016, Sigma-Aldrich) or shTHY1 (sc-42837, Santa Cruz Biotechnology).Cells were selected using puromycin (1.5 μg/ml).A total of 0.25 × 10 5 4T1 cells and 1.25 × 10 5 CAFs expressing either shCtrl or shTHY1 were mixed in a volume of 50 μl PBS and coinjected orthotopically into the fat pads of 8-week-old female BALB/c nude mice (Charles River).Mice were randomly allocated to the two groups.The tumors were harvested 14 days after inoculation.FFPE sections were stained for CD31 (1/75, ab28364 Abcam) on the Agilent autostainer using TRS high retrieval buffer (Agilent K8004) and for CD11b (1/5000, ab133357 Abcam) on the Leica Bond autostainer using Epitope retrieval buffer 2 (Leica AR9640) for 20 min.Quantitative analysis was performed on serial FFPE mouse tumor sections using Halo software (version 3.1.1076.363,Indica Labs).Veins were selected on the basis of the morphology and CD31 staining in the adjacent section.Software parameters were set that defined the stain of interest, and all sections were analyzed using the same settings.

Electron microscopy
CM-EVs and MBVs were isolated from pCAFs as described above, fixed in 4% PFA, ultracentrifuged at 100,000g (4°C, 90 min), and resuspended in PBS.Drops of 5 μl of CM-EV and MBV suspensions were loaded onto carbon-coated 400-mesh copper grids (Agar Scientific Ltd), which had been previously glow discharged (Quorum Q150T ES High Vacuum Unit settings 20 mA/30 s).Samples were left to absorb onto carbon surfaces for 30 min.Grids were floated on 100-μl droplets of PBS followed by fixation on a 50-μl droplet of 1% glutaraldehyde (Agar Scientific Ltd) for 5 min.Grids were washed with distilled water before they were contraststained with uranyl oxalate (Merck UK) (pH 7.0) (10 min in the dark) and embedded in methylcellulose/uranyl acetate (Merck UK) (10 min on ice in the dark).Grids were scooped up on platinum loops and excess fluid gently drained off, leaving thin films.Grids were left to dry before they were picked off and stored in a grid box.Samples were viewed on a JEOL 1200 EX TEM running at 80 kV, and digital images were captured using Olympus ITEM software and a Cantega 2kx2k Camera.
Fluorescent samples were imaged using a Zeiss 880 LSM confocal microscope (Carl Zeiss) in Lambda mode with a 32-channel spectral detector, and spectral unmixing was performed to remove as much tissue autofluorescence as possible.The autofluorescence spectrum was obtained from an unstained control, and fluorescence spectra were obtained from individual dyes (Hoescht, Alexa Fluor 488, Cy3, Alexa Fluor 647) using 405-, 488-, 561-, and 647-nm lasers.Unbiased imaging of entire tissue sections was performed using a Plan-Apochromat 20×/0.8M27 objective using tilescan and z-stack modes.Tile stitching, maximum Z projection and linear unmixing was performed using Zen Black software (version 2.3 SP1), and images were visualized in Zen Blue software (version 2.3).More detailed imaging of three tissue samples was performed using a Plan-Apochromat 40×/1.3Oil DIC M27 objective and z-stack mode.Maximum Z projection and linear unmixing was performed as above.Image processing was performed in Fiji (ImageJ, version 1.53f51).

RT-qPCR analysis
RNA was extracted from cultured cells or cells sorted after coculture.DNase treatment and total RNA isolation were performed using the RNeasy mini kit (Qiagen) according to the manufacturer's instructions.RNA (1 μg) was used to synthesize complementary DNA using the iScript kit (Bio-Rad).DNA was diluted to 10 ng/μl, and 2 μl was used in each RT-qPCR with 10 μl of iTaq Universal SYBR Green Supermix (Bio-Rad Laboratories) and 400 nM forward and reverse primers.PCR runs were performed using a QuantStudio 3 Real-Time PCR System (Thermo Fisher Scientific).Primers are listed in table S1.

Single-cell RNA sequencing
Data were analyzed as described in the original manuscripts (7,16).

Statistical analysis
Statistical analysis was performed on biologically independent replicates (n) using GraphPad Prism software version 9 (GraphPad Software Inc.).A Shapiro-Wilk test was used to test data for normality, and then the P value was calculated as detailed in each figure legend.P ≤ 0.05 was considered significant.

Fig. 1 .Fig. 2 .
Fig. 1.CAFs transfer proteins to ECs in vitro and in vivo.(A) The direct coculture method used in (B) to (F).(B) Representative images (maximum intensity projection processing from confocal z-stack) and corresponding 3D reconstruction of protein transfer from cFSe-labeled pcaFs (fully green cells) to huVecs.actin and nuclei were stained with phalloidin and DaPi, respectively (scale bar, 10 μm).(C and D) Quantification (c) of the protein transfer from pcaFs to huVecs at different ratios between the two cell types.pcaFs:huVecs ratios are 1:5, 1:1, and 5:1.n = 3 biological replicates (MFi, median fluorescence intensity and r, Pearson correlation).Representative histogram (D) of the 1:5 and 5:1 ratio.The y axis is normalized to mode (Mc, monoculture).(E) comparison of the protein transfer from huVecs to pcaFs and from pcaFs to huVecs.n = 4 biological replicates per condition.(F) comparison of the protein transfer from huVecs to huVecs, from pcaFs to huVecs, and from MDa-MB-231 cells to huVecs.n = 4 biological replicates per condition.(G) Proportions of RFP + ecs (cD31 + cD45 − cells) in the lungs of α-SMa-RFP tumor-free mice or mice with lung metastases.n = 4 mice per condition.Paired mice were born on the same day and are indicated with the same color.(H) Representative image (maximum intensity projection processing from confocal z-stack) of the tumor area in the lung of α-SMa-RFP mice stained for cD31.nuclei were stained with DaPi.Scale bar, 10 μm.Representative of two mice.(I) 3D reconstruction of the tumor vasculature and of the RFP signal; the distance between the RFP signal and the endothelium and the volume of the RFP signal are shown.Scale bar, 2 μm.Representative of two mice.Data are presented as means ± SeM. one-way analysis of variance (anoVa) with Tukey's multiple comparison test for (e, F) and two-tailed paired t test for (g).all significant P values are included in the figure.

Fig. 3 .Fig. 4 .Fig. 5 .
Fig. 3. CAF-derived THY1 supports the physical interaction between ECs and monocytes.(A) The direct coculture method used in (B) and (D) to (g).(B) mRna expression of THY1 in huVecs in monoculture and after 24 hours of coculture with pcaFs and in pcaFs.THY1 mRna amount was normalized to 18S expression.n = 3 biological replicates per condition.(C) Representative Western blot showing ThY1 protein abundance in pcaFs transfected with sictrl or siThY1.gaPDh was used as loading control.(D and E) Quantification (D) of ThY1 protein abundance in monoculture of huVecs and huVecs that were cocultured with pcaFs transfected with sictrl or siThY1 (n = 5 biological replicates per condition).Representative histogram (e) of ThY1 protein amounts in huVecs (the y axis is normalized to mode).(F) Quantification of the protein transfer from pcaFs transfected with sictrl or siThY1 to huVecs, n = 6 biological replicates for pcaF1 and 5 biological replicates for pcaF3.(G) number of ThP-1 monocytes per field bound to huVecs that were cocultured with pcaFs silenced or not for ThY1.colors indicate the paired independent experiments.n = 4 biological replicates for pcaF1 and 3 biological replicates for pcaF3.(H) Quantification of cD11b + areas adjacent to veins in 4T1 tumors cotransplanted with pcaF1 transfected with shctrl or shThY1.n = 6 mice for shctrl and n = 7 mice for shThY1 condition.(I) Quantification of cD11b + cells within veins in 4T1 tumors cotransplanted with pcaF1 transfected with shctrl or shThY1.n = 6 mice for shctrl and n = 7 mice for the shThY1 condition.(J) Representative images of tumor tissue sections from tumors containing shctrltransfected pcaFs stained for cD11b and cD31.The white arrowheads indicate cD11b + cells within veins; the green arrowheads indicate the areas adjacent to veins where cD11b staining has been quantified.Scale bar, 100 μm.images are representative of six mice.Data are presented as means ± SeM. one-way anoVa with Tukey's multiple comparison test for (B), (D), and (F), two-tailed paired t test for (g), and two-tailed Mann-Whitney U test for (h) and (i).all significant P values are in the figure.

Fig. 6 .PFig. 7 .
Fig. 6.CAFs have an enhanced protein transfer ability.(A) The direct coculture method used in (B).pnF-or pcaF-derived eVs were used to treat huVecs in (c) and (D).(B) Quantification of the protein transfer from pcaFs or pnFs to huVecs.n = 10 biological replicates for pnF1 and pcaF1, 4 biological replicates for pnF2 and pcaF2, and 7 biological replicates for pnF3 and pcaF3.(C) Quantification of the amount of proteins transferred by pnF-and pcaF-derived MBVs to huVecs.The eV amount was derived from the same number of donor cells.colors indicate paired independent experiments.n = 9 biological replicates per cell line.Data are normalized to the MFi of the monoculture of huVecs.The data related to pcaF-derived MBVs also are shown in Fig. 5D.(D) Quantification of the amount of proteins transferred by pnF-derived cM-eVs and MBVs to huVecs.The eV amount was derived from the same number of donor cells.colors indicate the paired independent experiments.n = 9 biological replicates per eV type.Data are normalized to the MFi of the monoculture of huVecs.The data related to pnF-derived MBVs also are shown in (c).(E) Scatter plot showing the correlation between the abundance of caF markers in fibroblasts (data file S4) and the amount of proteins that they transferred to huVecs, which corresponds to the data shown in fig.S5c.Data are in log 2 scale (ρ, Spearman rank correlation).Data are presented as means ± SeM. one-way anoVa with Tukey's multiple comparison test for (B) and two-tailed Wilcoxon matched-pairs test for (c) and (D).all significant P values are included in the figure.

Fig. 8 .
Fig. 8. Characterization of CAFs with high protein transfer ability.(A and B) Violin plot showing the expression of ACTA2 and TNFRSF12A in mycaF and icaF subpopulations.Data in (a) are from (16), and data in (B) are from (7).Two-tailed Mann-Whitney U test.all significant P values are included in the figure.(C) Representative image of TnFRSF12a, α-SMa, and cD31 staining in a tumor tissue section from a patient with breast cancer (maximum Z projection).nuclei were stained with hoechst-33342.Scale bar, 50 μm.(D) Working model showing mycaF-ec communication based on MBV-mediated transfer of proteins.