Extracellular vesicles in diabetes mellitus induce alterations in endothelial cell morphology and migration

Longitudinal and cross-sectional cohorts of euglycemic and diabetic individuals were chosen from the Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) study of the National Institute on Aging (NIA) Intramural Research Program, National Institutes of Health (NIH) [32]. HANDLS is a longitudinal study that follows community-dwelling participants, with the goal to better understand age-associated diseases in relation to race and socioeconomic status. Individuals were categorized as diabetic if they reported having received a diagnosis by a healthcare provider, were taking medication for diabetes mellitus or had a fasting serum glucose of > 125 mg/dL. Collection of plasma samples took place after the individual fasted overnight [33].

Previously, we examined EV characteristics in a longitudinal cohort of individuals [27]. This cohort was matched based on body mass index (BMI) category and included participants who had blood samples that were collected during two visits, with a time gap of approximately 5 years (4.95 ± 0.23). Participants included 19 individuals who were euglycemic at both visits, 19 individuals who were euglycemic on the first visit and diabetic during the second, and 20 individuals who were prediabetic on the first visit and diabetic during the second (Table 1). The cross-sectional cohort was matched for BMI and included nine euglycemic and nine diabetic individuals from one visit (Table 1). BMI was categorized as either underweight/normal (< 25 kg/m²), overweight (25 to < 30 kg/m²), obese class I (30 to < 35 kg/m²), and obese class II/III (≥ 35 kg/m²). Homeostatic model assessment (HOMA) of insulin resistance (HOMA-IR) and of β-cell function (HOMA-B) were calculated based on fasting serum glucose and insulin levels [34]. HOMA-IR and HOMA-B were natural log-transformed prior to statistical analysis.

Fasting plasma samples were collected as previously described [33]. For both cohorts, EVs were isolated, as previously reported [27], from 0.5 mL of plasma using ExoQuick Exosome precipitation solution (System Biosciences). ExoQuick offers the most reproducible results as well as the greatest ease of isolation for large human cohorts [33]. Samples isolated from ExoQuick were used in the proteomics experiment.

EVs were isolated from plasma using differential ultracentrifugation. Plasma was added to phosphate-buffered saline (PBS) and then centrifuged at 500 g for 10 min, at 2500 g for 10 min, and 10,000 g for 30 min. For the 10,000 g spin, a Beckman Coulter ultracentrifuge was used with a SW 55 Ti rotor (K = 48) to isolate pellets for immunoblotting and electron microscopy and a SW 32 Ti rotor (K = 204) for the rest of the experiments. Samples were then centrifuged at 120,000 g for 2 h to isolate pellets for immunoblotting and electron microscopy using a SW 55 Ti rotor (K = 48) and for the other experiments using a SW 32 Ti rotor (K = 204). The pellet in each sample was collected and then resuspended in PBS prior to a final spin at 120,000 g for 2 h (SW 55 Ti rotor). Pellets from the 10,000 g (10K) and 120,000 g (120K) spins were resuspended in sterile PBS. For the experiments that involved a vesicle count, the 10K fraction underwent another spin as previously mentioned [27]. For immunoblotting, the EV pellets were directly lysed in 50 µL of Mammalian Protein Extraction Reagent (M-PER) with phosphatase and protease inhibitors. For electron microscopy, the EV pellets were resuspended in sterile 4-(2-Hydroxyethyl)piperazine-1-ethanesulfonic acid (HEPES) buffer at a physiological pH.

CEM (T lymphoblast) cell lysate and equal amounts of lysed EVs were run on an SDS-PAGE gel and then immunoblotted with known EV protein markers, including Alix (sc-271975; Santa Cruz Biotechnology), Flotillin 1 (ab133497; Abcam), CD81 (EXOAB-CD81A-1; System Biosciences), and EV purity marker GM130 (ab52649; Abcam).

Electron microscopy images were taken by the Johns Hopkins University Neurology Microscopy Core using a Veleta camera (Olympus) as previously described [33]. Grids were viewed on a Libra 120 TEM at 120 kV (Zeiss).

Isolated EVs from differential ultracentrifugation were diluted to 1:50 in sterile PBS. Concentrations and size distributions of the samples were determined using nanoparticle tracking analysis (NTA) on the Nanosight NS500 (Malvern Instruments). At a camera level of 14 and detection level of 3, five videos of 20 s each were recorded for every sample. The NanoSight Software NTA 3.2 Build 3.2.16 was used for analysis. For accuracy, the samples for each cohort or experiment were measured around the same time period, on the same instrument, and by the same operator. Calculation of total EV concentration from plasma was done as reported before [33].

Plasma-derived EVs from individuals in the longitudinal cohort were lysed in M-PER, containing protease and phosphatase inhibitors. Protein content was quantified using the Bradford assay and adjusted to the same level across all samples. Equal amounts of protein were used as suggested by Olink^® Proteomics and since protein amounts do not always correlate with particle counts in EV preparations [35, 36]. 22 μg of protein in 20 µL (f.c 1.1 μg/μl) of each EV lysate were added to 96-well plates and then analyzed with Olink^® Proteomics biomarker Inflammation Panel using Proximity Extension Assay (PEA) technology (Olink^® Proteomics). PEA allows for sensitive and specific detection of multiple proteins. Experiments were performed blind of group status. Internal controls were used in each step, including a negative control that accounted for background levels and an interplate control that accounted for different plates. Protein data was normalized to inter- and intra-assay controls and represented as normalized protein expression (NPX) units on a log₂ scale. In total, 92 proteins were tested using the Olink^® Inflammation panel. We found 66 proteins that were detectable in the samples. Out of the 66 proteins, 33 proteins met our threshold for being less than 30% at the lower limit of detection, meaning that each of those proteins was present in more than 70% of all the EV samples. Linear mixed model regression was used to analyze the longitudinal data for the remaining proteins. CCL4, CXCL6, and FGF-21 all showed positively skewed distributions, and hence the data for those proteins were natural log-transformed.

Human aortic endothelial cells (HAECs; CC-2535; Lonza) were grown in EBM-2 Endothelial Cell Growth Basal Medium-2 (CC-3156; Lonza) with the EGM-2 Endothelial SingleQuots Kit (CC-4176; Lonza). In preparation for the migration assays, HAECs were incubated in serum-free EBM-2 media for 2 h. HAECs were detached from the plate with trypsin, followed by the addition of neutralization solution and HEPES buffer. The cells were then centrifuged and resuspended in serum-free EBM-2 media prior to experimentation.

The undersides of Transwell polycarbonate inserts (8.0-μm pore; 3422; Corning) were coated with collagen (30 µg/mL) for 1 h at 37 °C and then blocked with 1% BSA in PBS for 1 h at 37 °C. EVs were isolated using differential ultracentrifugation and the fractions from either the 10K or 120K pellets were used. For each test group, EVs from three individuals from the cross-sectional cohort were pooled and added at a dose of ~ 1.8 × 10⁸ EVs to the bottom chamber of the well containing 500 µL of serum-free EBM-2 media. VEGF-A (0.05 μg) was used as a positive control. Serum-starved HAECs, at dosages of ~ 2.5 × 10⁵ cells for the sEV (120K) assay and ~ 1.2 × 10⁵ cells for the medium-sized (10K) EV assay, were added to the transwells and allowed to migrate for 3 h at 37 °C. Filters were then washed with PBS. The remaining HAECs from the top of the transwells were removed. HAECs were fixed with 3.7% formaldehyde in PBS for 15 min and permeabilized with 0.5% Triton in TBS for 3 min. Cells were washed with PBS and the nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI; 32670; Sigma), followed by another PBS wash. Filters were then cut from the transwells and mounted on slides with an antifade medium. Images were acquired using a Zeiss Axio Observer D1 Inverted Fluorescence microscope with an AxioCam1Cc1 camera at a 40× objective. DAPI-stained nuclei in each field were counted.

HAECs (~ 15,000 cells) were plated on collagen-coated glass cover slips in EBM-2 media with 5% exosome-depleted fetal bovine serum (A2720803; Thermo Fisher Scientific) for 24 h. For another 24 h, HAECs were incubated with EVs isolated from the 10K fraction of plasma from an individual in the cross-sectional cohort and a previously reported dose of ~ 1000 vesicles per cell was used [33]. Three replicate experiments of each category, euglycemic and diabetic, were tested. HAECs were washed twice with PBS and fixed in 3.7% formaldehyde/PBS for 15 min, and then permeabilized using 0.5% Triton X-100 for 3 min. HAECs were washed twice in 1X TBS prior to actin-staining with Rhodamine Phalloidin (R415; Thermo Fisher Scientific). HAECs were washed with TBS and nuclei were stained with DAPI. Coverslips were washed and mounted on slides with an antifade medium. Images were acquired using the same microscope and objective as employed for the migration assays.

Cells were scored positive for lamellipodia/ruffles if there was at least one thick, curled, actin-rich structure per cell, similar to quantification previously described [37]. Only cells with less than one cell border were counted. Cells were quantified as a percentage of the total number of DAPI-stained nuclei. Approximately 50 cells were counted for each experiment and the averages of three different experiments were calculated.

In Table 1, for the longitudinal cohort, continuous variables were analyzed using one-way ANOVA, and for the cross-sectional cohort, the Student’s t test was used. The categorical variables were analyzed in the longitudinal cohort using the χ² goodness-of-fit test, and in the cross-sectional cohort using Fisher’s exact test. Data from the PEA assay was analyzed using R, version 3.3.2 [38]. The longitudinal cohort was analyzed using linear mixed models accounting for age, sex, race, matching across BMI groups, and repeated measurements. For EV concentration and size, the differences were analyzed using the Mann–Whitney test. The cell migration and morphology data were analyzed using one-way analysis of variance (ANOVA) and Tukey’s post hoc test.

Previous data suggested that EVs from diabetic individuals may contribute to heightened inflammation [27]. Hence, we examined whether the inflammatory protein content differed between EVs isolated from diabetic individuals and those of euglycemic controls. Previously, we constructed a longitudinal cohort comprised of individuals who became diabetic over a ~ 5 year time period (4.95 ± 0.23). Individuals were either euglycemic or prediabetic and then became diabetic or they were euglycemic at both time points (Fig. 1). We previously reported that individuals who developed diabetes over this time interval had a higher concentration of plasma EVs than euglycemic controls [27]. Demographic information about this cohort is listed in Table 1 and further characterization of EVs from this cohort was previously described [27]. A flow diagram describing the cohort design is shown in Fig. 1.

In order to analyze the inflammatory protein content of EVs from this cohort, the EVs were lysed and analyzed using a multiplex PEA. This assay quantifies inflammatory proteins in EVs using a sensitive and specific detection method [39‐43]. As shown in Table 2, many significant associations were found between the levels of EV inflammatory proteins and the acquisition of diabetes in our longitudinal cohort. Proteins with significant associations included CCL28, CD40, CD5, STAMBP, TWEAK, and VEGF-A, which are indicated in the Table. We also performed a cross-sectional analysis of inflammatory protein levels at time 2 of the longitudinal cohort, where additional significant associations were found between EV inflammatory protein levels and diabetes status. Proteins with significant associations included CD5, MCP-1, and VEGF-A. Some proteins were only significantly associated with one of the diabetes diagnoses groups. However, these groups had different starting conditions (i.e. non-diabetic or pre-diabetic), which may have affected the magnitude of change and the strength of the associations with the EV protein levels.

Homeostatic model assessment is a mathematical model that measures insulin resistance and β-cell function [34]. Many clinical and epidemiological studies employ the HOMA model to measure the severity of diabetes in individuals [44]. We wanted to examine the various inflammatory proteins in relation to a different quantitative assessment of diabetes mellitus. In our proteomics analysis, we found significant associations between EV inflammatory protein levels and HOMA-B as well as HOMA-IR in our longitudinal analyses and also in our cross-sectional analysis at time 2. According to the longitudinal analysis, proteins with significant associations with HOMA-B included IL-10RB, IL-18R1, SCF, and VEGF-A, and proteins with significant associations with HOMA-IR included DNER, HGF, IL-10RB, IL-18R1, and VEGF-A. In the cross-sectional analysis of the longitudinal cohort at time 2, proteins with significant associations with HOMA-B included SCF and uPA, and proteins with significant associations with HOMA-IR included DNER, HGF, IL-18R1, and VEGF-A.

Out of all the examined proteins, EV-associated vascular endothelial growth factor A (VEGF-A) levels showed significant associations with most of the variables analyzed (Table 2). Levels of VEGF-A in EVs were significantly associated with diabetes status in our analyses of the longitudinal cohort at both times and at time 2. Individuals who were prediabetic or euglycemic at time 1 and diagnosed with diabetes at time 2 showed significantly higher levels of VEGF-A in their EVs than individuals who were euglycemic at both time points (Fig. 2). Figure 2 visualizes the predicted values from linear mixed model regression showing that individuals who developed diabetes had higher levels of EV VEGF-A over time. This analysis included two time points for each person, and thus we have depicted time using age in the graph. Figure 2 shows the change in EV VEGF-A levels by age in the three different diabetes diagnosis groups. Furthermore, EV VEGF-A levels were significantly associated with levels of HOMA-B and HOMA-IR in our longitudinal analysis, and HOMA-IR in our cross-sectional analysis at time 2. VEGF-A is a widely known angiogenic factor and its binding to receptors, including VEGFR-1 and VEGFR-2, regulates angiogenesis [45]. VEGF-A plays an important role in the formation of vasculature by stimulating a variety of processes, including endothelial cell migration [46].

Given the association that we found between EV inflammatory protein content and diabetes status, we examined whether EVs from diabetic individuals affected cell behavior in vitro. Diabetes is known to be associated with an increased risk for vascular disease, which may arise from interactions between circulating EVs and the endothelium [31]. Hence, we used endothelial cells to see whether EVs from diabetic individuals can induce functional effects.

For these studies, plasma EVs from the cross-sectional cohort were isolated using differential ultracentrifugation (Fig. 1). We chose to use this isolation method since precipitation reagents contain polyethylene glycol, which may confound the outcomes of cell migration assays [47, 48]. Guidelines from the International Society for Extracellular Vesicles were followed in characterizing and validating the presence of isolated EVs [49]. EVs were isolated based on size via differential ultracentrifugation, in which the 10,000 g (10K) and 120,000 g (120K) fractions contain, respectively, medium-sized and sEVs [23]. The 10K and 120K fractions showed signal for known EV markers and no signal for the EV purity marker (Fig. 3a). Images taken with electron microscopy show intact, round vesicles for both the 10K and 120K samples (Fig. 3b, c). The size distributions for the 10K and 120K fractions were quantified using NTA (Fig. 3d). The EVs from the 120K fraction have a typical size distribution of plasma EVs and show a peak at approximately 200 nm. In comparison, EVs from the 10K fraction have a broader distribution and a significantly lower concentration of particles (Fig. 3d, e). Comparison of EV mean and mode sizes indicate that EVs isolated from the 10K fraction were significantly larger than EVs isolated from the 120K fraction (Fig. 3f, g). Hence, our samples isolated from differential ultracentrifugation exhibit size and morphological characteristics of EVs.

Based on our proteomic analysis showing that EV inflammatory protein levels are significantly associated with diabetes status, we wanted to see if EVs that differed in molecular cargo also differed in their functional effects. Using the Boyden transwell assay, we allowed HAECs to migrate towards either EVs, isolated from individuals in our cross-sectional cohort, or VEGF-A. EVs from the 120K and 10K spins of differential ultracentrifugation were classified as, respectively, sEVs and medium-sized EVs [23]. Our results showed no differences in endothelial cell migration towards sEVs (120K fraction) isolated from euglycemic or diabetic individuals, with the migration from both groups being similar to that of the untreated endothelial cells (Fig. 4a). In contrast, our subsequent test with medium-sized EVs (10K fraction) showed greater cell migration with EVs derived from diabetic individuals than compared to the euglycemic control (Fig. 4b).

Since endothelial cell migration is mobilized by the actin cytoskeleton [50], we tested whether medium-sized EVs (10K fraction) from individuals with diabetes may lead to structural alterations in target cells. After incubating HAECs with EVs from diabetic or euglycemic individuals, we stained the actin cytoskeleton of the cells. Our results show that HAECs displayed significantly more actin-rich, membrane protrusions and ruffles, called lamellipodia, when treated with EVs from diabetic rather than euglycemic individuals, and that EVs from the latter led to morphology similar to what was seen in the untreated cells (Fig. 5). There were no significant changes in cell size across all categories (Additional file 1: Figure S1).