Matrix mechanics and water permeation regulate extracellular vesicle transport

Abstract

Cells release extracellular vesicles (EVs) to communicate over long distances, requiring EVs to traverse the extracellular matrix (ECM). However, given that the size of EVs is usually larger than the mesh size of the ECM, it is not clear how they can travel through the dense ECM. Here we show that in contrast to synthetic nanoparticles, EVs readily transport through nanoporous ECM. Using engineered hydrogels, we demonstrate that the mechanical properties of the matrix regulate anomalous EV transport under confinement. Matrix stress relaxation allows EVs to overcome confinement, and a higher crosslinking density facilitates fluctuating transport motion through the polymer mesh, leading to free diffusion and fast transport. Furthermore, water permeation through aquaporin-1 mediates EV deformability, which further supports EV transport in hydrogels and decellularized matrix. Our results provide evidence of the nature of EV transport within confined environments and demonstrate an unexpected dependence on matrix mechanics and water permeation.

Extracellular vesicles (EVs) are cell-derived particles found in the extracellular matrix (ECM) [1] and described as ranging from 50–500 nm in diameter [2]. However, the ECM has a range of mechanical properties and often features average mesh sizes smaller than EVs [3] (Fig. 1a).

Figure 1. Extracellular vesicles transport within decellularized lung tissue.

a, EVs exist within extracellular matrices (ECM), where often the mesh size is smaller than the size of EVs. Mechanisms and dependencies directing their potential transport under confinement are unclear. b, (Top Left) Representative images of decellularized lung tissue with EVs passively loaded, imaged using second harmonic multiphoton microscopy. Scale bars = 15 μm. (Bottom Right) A pixel intensity chart drawn along the dotted line in the combined image, demonstrating that EVs exist along fibers as K2S pixel intensity is correlated strongly with collagen pixel intensity. The mean Pearson’s correlation coefficient is reported for N = 9 ROIs analysed across 3 pairs of background-subtracted images. *, p < 10⁻¹⁵ via unpaired two-tailed t-test. Error bars denote standard error of the mean (SEM). c, Mean % EV release from decellularized lung tissue over time with t_1/2 = 24.7 hours. N = 5 tissue slices across 3 independent experiments. d, Representative images of EV load (after 72 hours) and release (after 24 hours) in decellularized lung tissue. The axis scale is fluorescence intensity counts (AU). Scale bars = 2 mm. Error bars denote standard deviation (SD).

To evaluate the extent to which EVs transport through the interstitial ECM, we engineered EVs from mouse mesenchymal stromal cells (MSCs) to contain the EV marker CD63 fused with Katushka2S (K2S, a far-red fluorescent protein [4]) in order to visualize them after passive loading by incubation in decellularized matrix from lung tissue. MSCs were chosen as the source of EVs because in vivo they are often present in interstitial regions surrounded by matrix [5]. Expression of CD63-K2S in EVs (K2S-EVs) does not alter their expected size distribution (diameter [d] ~ 50–150 nm) (Supplementary Fig. 1a). Multiphoton second harmonic imaging analysis shows that EVs are distributed throughout collagen fibers within the matrix (Fig. 1b). Despite a nanoscale mean porosity (Supplementary Fig. 1b, c) of the matrix, ~50% of loaded CD63-K2S-EVs are released from the matrix within ~24.7 hours (Fig. 1c, d), suggesting that EVs readily transport through naturally derived nanoporous matrices.

Decellularized matrix exhibits complex shear modulus magnitude G* ~ 750 Pa with loss tangent (viscous modulus/elastic modulus, G”/G’) ~ 0.15 (Fig. 2a and Supplementary Fig. 2a), and stress relaxation behavior (t_1/2 ~ 15s) (Fig. 2b). To determine whether matrix mechanics mediates EV transport, we engineered alginate-based hydrogels with a range of mechanical properties known to be present in tissues [6]. Importantly, alginate-based hydrogels are bio-inert, non-degradable and exhibit homogeneous nanoporous structures [7], making them ideal to model ECM without the influence of biochemical or degrading interactions. Hydrogels can be crosslinked physically through divalent cations or covalently through click chemistry, and G* is tunable for both (Fig. 2c, Left and Supplementary Fig. 2b). Physical crosslinking leads to stress relaxing hydrogels and covalent crosslinking leads to elastic hydrogels as indicated by loss tangent (Fig. 2c, Right) and stress relaxation times (Fig. 2d) [8]. We consider G* ~ 500 Pa ‘soft’ and G* ~ 3000 Pa ‘stiff’. Alginate-based hydrogels are nanoporous like the decellularized matrix (Supplementary Fig. 2c) regardless of crosslinking density or type. This is consistent with the egg-box model of crosslinking between alginate chains [9], where more crosslinking is not expected to dramatically alter mesh size. As expected, after dextran-FITC (hydrodynamic radius ~15 nm [10]) molecules are encapsulated in hydrogels, most release completely within 24 hours (Fig. 2e). In contrast, minimal release is observed for polystyrene nanoparticles (NPs; d ~80–100 nm) (Supplementary Fig. 2d). Like decellularized tissue, some EVs release from hydrogels; however, surprisingly, EV release is greater from stress relaxing hydrogels with a higher G*. This effect occurs for EVs from other cells (Supplementary Fig. 2e), suggesting its generalizability across cell type. Liposomes with a similar size (Supplementary Fig. 2f) and lipid content as EVs [11] do not exhibit higher release from stress relaxing hydrogels with a higher G* (Supplementary Fig. 2g). Hydrogels do not undergo degradation or loss of mass over the tested time period (Supplementary Fig. 3a), confirming independence of degradation. Importantly, this observation is independent of Ca²⁺, since treatment with ionomycin or EGTA do not affect release (Supplementary Fig. 3b and 3c). To test whether EV release is mechanosensitive in a more natural ECM composition, an interpenetrating network (IPN) hydrogel of alginate and collagen-I polymers was fabricated [12] where the hydrogel G* is tunable independent of collagen-I concentration (Supplementary Fig. 3d). While EV release from the IPN is generally lower depending on collagen concentration, release remains mechanosensitive (Supplementary Fig. 3e).

Figure 2. Complex shear modulus and stress relaxation time regulate bulk release of EVs from hydrogels.

a, Mean rheological properties of N = 5 decellularized lung tissue slices calculated at 1 Hz. (Left) Complex shear modulus. (Right) Loss tangent. b, Decellularized lung tissue exhibits stress relaxation with t_{1/2 =} 14.9 seconds. Data represent the mean of N = 3 tissue slices. c, Rheological properties of hydrogels calculated at 1 Hz. (Left) Complex shear modulus. (Right) Loss tangent. d, Stress relaxation properties of hydrogels. For (c) and (d), Data represent the mean of N = 3 hydrogels. Error bars are SEM. e, Release of EVs but not dextran or NPs is affected by hydrogel complex shear modulus for hydrogels exhibiting stress relaxation. (Left) 500 kDa dextran release from hydrogels. (Middle) NP release from hydrogels. (Right) EV release from hydrogels. N = 3 hydrogels across 3 independent experiments. Data represent the mean and error bars denote SEM. *, p = 0.0095 via two-way ANOVA followed by Tukey’s test for multiple comparisons. Unless stated otherwise, error bars denote SEM.

To study whether EV release from engineered hydrogels corresponds to individual EV transport, we developed a 3D particle tracking approach utilizing high-speed 3D microscopy with deconvolution to visualize (Fig. 3a and Supplementary Movies) and calculate the mean square displacement (MSD) of CD63-K2S-EVs over time in different environments. Particles were tracked immediately after hydrogel formation to capture initial behaviors possibly affected by hydrogel swelling. Data were collected every Δt = 0.267 seconds over a total time T ~ 8 seconds. Next, data were ensemble-averaged over numerous tracks and fit to the power law form [13]

$$<MSD\left(t\right)>=K_{\alpha }{t}^{\alpha }$$ (1)

to calculate an effective ensemble exponent α. The effective diffusion coefficient

$$D_{\tau }=\raisebox{1ex}{$MSD\left(\tau \right)$}\!\left/ \!\raisebox{-1ex}{$6\tau $}\right.$$ (2)

was calculated for each track over each interval τ = 4Δt ~ 1.06s [14] to give

$$D_{1.06s}=\raisebox{1ex}{$MSD\left(\tau =1.06s\right)$}\!\left/ \!\raisebox{-1ex}{$6\left(1.06s\right)$}\right. .$$ (3)

Figure 3. Individual EVs exhibit anomalous transport in matrix that is more rapid and diffusive in stiff stress relaxing matrix.

a, Representative 3D particle tracks for EVs in matrix. See supplementary information for videos. b, Values of α calculated for a non-linear fit of tracking data for EVs in matrices (Equation 1). Error bars represent 95% CI. c, EVs in stiff stress relaxing matrix (Left, N = 279) exhibit more diffusive ensemble-averaged transport (α ~ 0.89) relative to EVs in soft stress relaxing (Middle, N = 263) or stiff elastic matrix (Right, N = 89). Data represent the mean and error bars represent SEM. d, Mean D_1.06s calculated for tracks in (b). *, p = 6.9 × 10⁻⁷ via one-way ANOVA with Tukey’s test for multiple comparisons. e, EV transport in matrix displays dynamic heterogeneity, indicated by a higher standard deviation of D_1.06s for measured tracks versus simulated tracks σ_meas/σ_sim. N = 5 simulations. *, p < 10⁻¹⁵ via one-way ANOVA with Tukey’s test for multiple comparisons. f, Distributions of the change in diffusion coefficient ΔD_0.53s calculated at time (t) ~4 seconds are broader for EVs in stiff matrix, indicating fluctuating motion. g, Escape from cages of confinement for EV tracks in (b). (Left) Fraction of EVs able to escape cages. (Right) Time elapsed before EVs escape cages. *, p = 7.5 × 10⁻⁸ via unpaired two-tailed t-test. h, Radius of gyration R_g for EV tracks in (b). *, p < 10⁻¹⁵ via one-way ANOVA with Tukey’s test for multiple comparisons. Each tracking condition was performed across 2 independent experiments. For (e) and (f), NPs in 80% glycerol are analysed for N = 32 tracks. Unless stated otherwise, error bars denote SEM.

Multiple values for D_1.06s(τ) are obtained for a single track for each interval τ and averaged to obtain a single D_1.06s for each track (see Methods). We validated our method by measuring transport of NPs in glycerol solutions with different solution viscosities and thus different expected transport speeds. NPs in these solutions show an α ~ 1 (Supplementary Figs. 4a and 4b), indicating diffusive transport. Furthermore, they exhibit diffusion coefficients D_1.06s like that expected from conventional Stokes-Einstein theory (Supplementary Fig. 4c). In contrast, NPs in stiff stress relaxing matrix exhibit a sub-diffusive (α ~ 0.39) slower (D_1.06s ~ 0.01 μm²/s) transport (Supplementary Fig. 4d), indicating confinement. Strikingly, EVs in stiff stress relaxing matrix show α approaching that of NPs transporting in solution (α ~ 0.88) (Figs. 3b and 3c). EVs in soft stress relaxing matrix exhibit a significantly lower D_1.06s (Fig. 3d) with sub-diffusive transport (α ~ 0.49), while EVs in stiff elastic matrix show more pronounced sub-diffusive transport (α ~ 0.045), indicating that matrix stress relaxation allows EVs to overcome confinement.

Stress relaxing matrix systems can give rise to ‘dynamic heterogeneity’ [15] wherein particles can escape confinement or ‘cages’ formed by the matrix. To determine an expected standard deviation of D_1.06s for particles in a homogeneous system, tracks were simulated matched to measurement conditions (see Methods). Simulated tracks follow measured tracks for NPs transporting in solutions (Supplementary Fig. 4e). The standard deviation of experimentally determined D_1.06s (σ_meas) was calculated and normalized to the standard deviation of D_1.06s for simulated trajectories (σ_sim) to measure degree of heterogeneity of D_1.06s [16]. While NPs in solution follow their simulated trajectories with lower degree of heterogeneity σ_meas/σ_sim (Supplementary Fig. 4f), EVs in matrix show a higher σ_meas/σ_sim (Fig. 3e), indicating a more heterogeneous distribution of D_1.06s. To investigate this behavior, we analysed how individual EVs exhibit changes in transport motions over time by defining another 3D diffusion coefficient (D_0.53s) with shorter intervals τ = 2Δt ~ 0.53s to capture local transport behaviors. D_0.53s was calculated for each interval τ_i within tracks to express each track as D_0.53s (τ). Next, the difference of D_0.53s (τ) between consecutive intervals τ_i and τ_i+1 (τ₁ ~ 0.53s, τ₂ ~ 1.06s …) was taken to calculate ΔD_0.53s

$$\mathrm{\Delta }{D}_{0.53s}\left({\tau }_i\right)=D_{0.53s}\left({\tau }_{i+1}\right)-D_{0.53s}\left({\tau }_i\right)$$ (4)

which indicates the magnitude of changes in diffusion coefficient over time within a track. To compare the spread of ΔD_0.53s between groups, values for ΔD_0.53s are normalized to the mean ΔD_0.53s for each group (Normalized ΔD_0.53s). From a theoretical perspective, particle motion is facilitated when ΔD_τ > 0, particle motion is hindered when ΔD_τ < 0, and particle motion remains constant when ΔD_τ ~ 0 (Supplementary Fig. 5a). ΔD_0.53s values are close to zero for NPs transporting in solution (Supplementary Fig. 5b), suggesting that ΔD_0.53s ~ 0 for particles undergoing free diffusion. However, individual tracks of EVs in stiff matrix show a much broader distribution of ΔD_0.53s (Fig. 3f, Supplementary Figs. 5c and 5d), suggesting that stiff matrix drives fluctuating transport motions within tracks. Furthermore, ΔD_0.53s values are ~50% both positive and negative (Supplementary Fig. 5e), indicating that this behavior is associated with zero-mean fluctuations in transport motion.

To calculate the extent to which EVs escape confinement, we modeled the matrix as a system of ‘cages’ with defined size c that transporting particles must overcome (Supplementary Fig. 6) [17–19]. Since NPs in stiff stress relaxing matrix are confined with α ~ 0.39, c was defined as the plateau MSD for this condition (c ~ 0.09 μm²). Tracks were analysed to determine whether their MSD exceeds c (fraction of particles escaping from ‘cages’) and if so, the elapsed time before the MSD exceeds c (escape time). A significant amount of EVs in stiff stress relaxing matrix demonstrate the ability to escape cages and they do this more rapidly (~1.3s) than EVs in soft stress relaxing matrix (Fig. 3g). In contrast, EVs in stiff elastic matrix less readily escape cages, further showing that matrix stress relaxation is crucial for allowing EV transport. Furthermore, we calculated the radius of gyration R_g [20] for each particle, defined as the root mean square distance from the center of the trajectory. EVs in stiff stress relaxing matrix explore more space than EVs in soft stress relaxing matrix indicated by a higher R_g (Fig. 3h).

Because EVs show the ability to transport in confined spaces, we hypothesized that intrinsic EV properties also drive their transport. While lyophilized (freeze-dried) EVs possess the same size distribution as freshly isolated EVs (Supplementary Fig. 7a), they do not exhibit greater release from the stiff stress relaxing hydrogel (Fig. 4a) – this is further confirmed by a decrease in D_1.06s by ~10-fold and α to ~0.25. (Fig. 4b). Non-lyophilized EVs with integral membrane structure are likely required for mechanically sensitive transport, since lyophilizing EVs [21] can compromise their membrane integrity. This is supported by addition of the cryoprotectant trehalose to EV preparations during lyophilization [22], which recovers release behavior (Supplementary Fig. 7b). We speculated that transport may be regulated by EV surface interactions within hydrogels or actomyosin contractility within EVs. However, tethering the integrin binding ligand RGD (~0.8 μM) within hydrogels or treating hydrogels with drugs against myosin-II (blebbistatin) and Rho-associated protein kinase (Y27632) do not affect EV release (Supplementary Fig. 7c and 7d). Importantly, ATP within EV preparations exists at a concentration much less than in cells (Supplementary Fig. 7e), and EVs from cells partially (~50%) depleted of ATP do not release differently (Supplementary Fig. 7f), indicating that EV transport mechanisms are likely metabolically passive rather than active.

Figure 4. Aquaporin-1 mediates the ability of EVs to transport in engineered and decellularized matrices by increasing EV deformability.

a, After lyophilization, mean % EV release is decreased from stiff stress relaxing hydrogels. N = 3 hydrogels for each condition. **, p = 0.012 via two-way ANOVA followed by Tukey’s test for multiple comparisons. b, (Left) Ensemble MSD curves for untreated (N = 279) versus lyophilized (N = 618) EV tracks in stiff stress relaxing matrix. (Middle) Values of α from a non-linear fit by Equation 1. Error bars represent 95% CI. (Right) Mean D_1.06s. *, p = 2.9 × 10⁻¹² via unpaired two-tailed t-test. c, Hypertonic medium (3% polyethylene glycol, 300 kDa) significantly increases mean % EV release from stress relaxing hydrogels. N = 3 hydrogels for each condition. *, p = 0.026 (soft), p = 5 × 10⁻³ (stiff) via unpaired two-tailed t-test. d, EVs from cells treated with siRNA against AQP1 (N = 6) exhibit significantly higher mean Young’s Modulus (E) than EVs from cells treated with a scrambled siRNA control (SCR, N = 7). *, p = 0.005 via unpaired two-tailed t-test. e, EVs depleted of AQP1 exhibit significantly lower mean % release from stress relaxing hydrogels. N = 3 hydrogels for each condition. *, p = 0.021 (soft), p = 8.6 × 10⁻³ (stiff) via unpaired two-tailed t-test. f, Mean % release of AQP1-depleted EVs (N = 7) from decellularized lung tissue is significantly reduced versus a control (N = 8). *, p = 0.010 via unpaired two-tailed t-test. g, (Left) Ensemble MSD curves for AQP1-depleted EV tracks (N = 613) versus control (N = 659) EV tracks. (Middle) AQP1-depletion does not change α values. Error bars are 95% CI. (Right) AQP1-depletion significantly decreases mean D_1.06s. *, p = 1.3 × 10⁻⁸ via unpaired two-tailed t-test. h, Analysing tracks from (g), AQP1-depleted EVs exhibit significantly slower mean escape time than control EVs in stiff stress relaxing matrix. *, p = 2.1 × 10⁻⁷ via unpaired two-tailed t-test. Unless stated otherwise, error bars denote SEM.

Water permeation via aquaporins drives migration of spatially confined cells independent of myosin-II [23]. Since aquaporins are partitioned into EVs [24], we hypothesized that water permeation through aquaporins regulates EV transport. EV release in both stiff and soft stress relaxing hydrogels is increased by addition of 3% polyethylene glycol (Fig. 4c) but does not occur if EVs are freeze-dried (Supplementary Fig. 7g). We then tested whether aquaporins are required for EV release. AQP1 is the dominant aquaporin isoform expressed in MSCs (Supplementary Fig. 8a and Supplementary Table 1). Treating cells with siRNA against AQP1 leads to a ~80% mRNA knockdown in cells (Supplementary Fig. 8b) and a ~60% reduction in AQP1 protein packaged into EVs (Supplementary Fig. 8c). AQP1 depletion in EVs significantly increases their Young’s modulus (Fig. 4d and Supplementary Figs. 9a, 9b), suggesting that water permeation makes EVs more deformable. AQP1 depletion in EVs significantly decreases EV release from hydrogels (Fig. 4e), and AQP1-depleted EVs show impaired release from decellularized matrices (Fig. 4f and Supplementary Fig. 9c), indicating that greater deformability via AQP1 enhances EVs ability to transport in matrix. While AQP1 depletion reduces D_1.06s by ~3-fold, α remains unchanged for individual EVs (Fig. 4g). Liposomes encapsulated in the stiff stress relaxing matrix exhibit an α ~ 0.65 (Supplementary Fig. 9d) with a much lower D_1.06s (Supplementary Fig. 9e), suggesting the presence of lipid membrane alone is not sufficient for enhanced EV transport. Pulling values from all experimental groups of EVs in matrix shows that α increases with increased D_1.06s but becomes saturated near α ~ 1.0 when D_1.06s is higher than 0.1 μm²/s (Supplementary Fig. 9f), suggesting that a 3-fold decrease in D_1.06s via AQP1 depletion is less likely sufficient to significantly decrease α. Consistent with these results, AQP1 depletion decreases the time required for EVs to escape cages (Fig. 4h). Finally, AQP1 depletion does not affect spread of ΔD_0.53s (Supplementary Fig. 9g), indicating independence of AQP1 with fluctuating transport motion.

Results describe the ability of EVs to transport in polymer matrix with an absence of matrix degradation, despite EVs being larger than the average mesh size of matrices. Matrix stress relaxation allows EVs to readily escape cages formed by the polymer network (Fig. 5). Stiff matrix increases fluctuating EV transport motions, and thus the combination of stiffness and stress relaxation leads to greatly enhanced EV transport. EVs are also subject to water permeation through AQP1, which allows EVs to become more deformable by altering their volume, enabling their escape from confinement. This behavior is reminiscent of a model of hopping diffusion of nanoparticles in entangled polymer matrices [25–27], where it has been hypothesized that nanoparticles show the ability to slide through matrix under some conditions. Phospholipid contents of EVs vary [28], and thus it will be interesting to determine whether and how these contents affect EV transport in matrix, as lipid asymmetry was shown to affect EV membrane stability [29]. The observation that AQP1 mediates EV deformability and resulting transport in ECM is important because deformability of synthetic nanoparticles with lipid bilayers was recently shown to dramatically affect their accumulation in tissues both in vitro and in vivo [30]. Future studies will test whether the presence of water channels on lipid vesicles alone is sufficient or if other membrane components are also necessary to facilitate EV transport under confinement in matrix. Furthermore, the 3D particle tracking approach utilized here can be extended to study EV transport in various environments, for investigating or treating diseases implicating EVs. Finally, the results may inform how therapeutic EVs can potentially be modified to better facilitate their delivery through tissue ECM. In sum, this study opens new avenues of investigations into EV transport behaviors occurring in the ECM.

Figure 5. Model for EV transport under confinement.

EVs exist trapped in elastic matrix, while matrix stress relaxation allows EVs to escape confinement. Stiffness in a stress relaxing matrix leads to fluctuating transport motions, further increasing EVs ability to transport. Furthermore, AQP1 present on EVs mediates water permeation within EVs leading to greater EV deformability and enhanced transport under confinement.

Methods

Particle size and number characterization

Particle size and number were obtained using Nanoparticle Tracking Analysis 3.2 (NTA) via NanoSight NS300 (Malvern) using 405nm laser. Samples were introduced by syringe pump at a rate 100 μL/min. Three thirty-second videos were acquired using camera level 14 followed by detection threshold 7. Camera focus, shutter, blur, minimum track length, minimum expected particle size and maximum jump length were set automatically by the software. Samples were diluted as needed to maintain particles per video from 100–2000.

Cell culture

All cells were cultured at 37C in 5% CO₂. HeLa cells (CCL-2, ATCC) were a gift from Dr. Andrei Karginov at UIC. D1 MSC cells (CRL-12424, ATCC), HeLa cells and HEK293T cells (CRL-3216, ATCC) were cultured using high-glucose DMEM (Thermo) supplemented with 10% FBS (Atlanta Biologicals), 1% penicillin/streptomycin (P/S, Thermo) and 1% GlutaMAX (Thermo) to 80% confluency before passaging, no more than 30 times. Human umbilical vein endothelial cells (HUVEC) (#CC-2519, Lonza) were a gift from Dr. Yulia Komarova at UIC. HUVEC were cultured using Ham’s F-12K (Thermo) supplemented with 10% FBS, 1% P/S, 1% GlutaMAX, 0.1 mg/mL heparin (Sigma #H3393) and endothelial cell growth supplement (Sigma #E2759) at passage 5. Human MSCs (hMSCs) were derived by plastic adherence of mononucleated cells from human bone marrow aspirate (Lonza). After 3 days, adherent cells were cultured in the hMSC medium: α-minimal essential medium (αMEM, Thermo) supplemented with 20% FBS, 1% penicillin/streptomycin (P/S, Thermo Fisher Scientific), and 1% GlutaMAX (Thermo). After reaching 70~80% confluence at 10~14 days, cells were split, expanded in the hMSC medium and used at passage 3. Cells were routinely tested for mycoplasma contamination and only used if no contamination was present.

Lentiviral expression of CD63 fused with Katushka2S

A DNA plasmid containing Katushka2S (K2S) was synthesized in a pUC57-Kan backbone (GenScript). The Katushka2S sequence was cloned into a lentiviral construct containing CD63 (LV112335, Applied Biological Materials) so that K2S is fused to CD63 on the C-terminus of CD63. D1 MSCs were transduced with lentivirus containing the CD63-K2S plasmid using standard techniques [31]. Briefly, lentiviral particles were produced with a 2^nd generation lentiviral packaging system (LV003, Applied Biological Materials) using Lentifectin (Applied Biological Materials) in HEK293T cells. Lentiviral particles were purified and applied to D1 MSCs at passage 10 with 8 μg/mL polybrene (Sigma) for 3 days. Cells were expanded over a period of several days to reach ~80% confluency. Then, cells were sorted using a MoFlo Astrios (Beckman Coulter) based on their CD63-K2S signal compared to non-transduced cells of the same passage. Concentrated EV solutions were shown to be positive for CD63-K2S versus EVs from non-transduced cells using IVIS imaging (Living Image 4.0, Perkin Elmer).

Extracellular vesicle isolation and preparation

To isolate EVs from cells, cells were washed twice with Hank’s balanced salt solution (HBSS, Thermo) followed by incubation with serum-free growth medium for 1 hour. Afterwards, medium was exchanged with medium consisting of high-glucose DMEM supplemented with 10% exosome-depleted FBS (Thermo) instead of 10% FBS. The next day, medium was centrifuged at 2,000xg for 10min to remove cell debris followed by centrifugation at 10,000xg to remove particles larger than 500nm [32]. Afterwards, the solution was added to a 100 kDa MW-cutoff column (Amicon) and centrifuged at 5000xg for 20min followed by washing with an equal volume of HBSS. The retentate was resuspended and confirmed to contain concentrated EVs using NanoSight NS300 (Malvern).

Lyophilization of EVs

Concentrated EVs were frozen at −80×C overnight. If applicable, preparations were treated with 4% trehalose (Sigma) before freezing. They were then placed in a lyophilization chamber operating at < 0.1mBar vacuum and < −100×C temperature and allowed to sublimate overnight. The solid was reconstituted in HBSS and confirmed to contain EVs using NanoSight NS300.

Decellularization of lung tissues

All animal procedures were performed in compliance with NIH and institutional guidelines approved by the ethical committee from the University of Illinois at Chicago. Female C57BL/6J mice were purchased from The Jackson Laboratory, housed in the University of Illinois at Chicago Biologic Resources Laboratory, and sacrificed 12 weeks after birth. Lung tissue was harvested and decellularized based on techniques described previously [33]. Briefly, the heart-lung bloc was exposed, and the trachea cannulated with a blunted 18-gauge needle. Lungs were infused with 1mL deionized water containing 5% penicillin/streptomycin (wash solution). The heart-lung bloc was excised and washed through the airway and the right ventricle (RV), incubated in 0.1% Triton-X wash solution overnight at 4C, washed, and incubated in 2% sodium deoxycholate wash solution overnight at 4C. It was then washed, incubated in 1M NaCl wash solution for 1 hour at RT, washed, and incubated in wash solution containing DNAase for 1 hour at RT. The tissue was placed in a solution of liquified 5% low-melting-point agarose (GeneMate) and allowed to solidify at 4C overnight. Slices were prepared using a tissue slicer (Braintree) into 1 mm sections and punched into 5mm discs using a punch (Integra). Discs were placed in HBSS, incubated at 42C for 30min and washed several times.

Multiphoton microscopy

~1 × 10⁹ CD63-K2S EVs were incubated with a ~5 mm tissue slice at 37×C for 3 days followed by washout. EV-loaded tissue slices were imaged using a 20X, 1.00 N. A water immersion objective (Olympus) with a multiphoton microscope (Bruker Fluorescence Microscopy, Middleton WI; formerly Prairie Technologies) equipped with a Coherent Cameleon Ultra II laser employing both second harmonic and 2-photon excited fluorescence signal generation (SHG and 2PEF) [34]. Backward scattering SHG was obtained at 860 nm excitation to capture signals from collagen within tissue and 2PEF was performed at 760 nm excitation to capture signals from CD63-K2S. Three images were taken each for experimental and background (no loaded EVs) conditions. Images were processed by subtracting background fluorescence from the 760 nm channel. Then, three ROIs were chosen for each background-subtracted image and the Pearson’s correlation coefficient (PCC) was calculated. Next, the 760 nm channel signal was randomized using the MATLAB function randblock [35], the PCC calculated again, and distributions were compared.

Lung tissue transport experiments

After loading ~1 × 10⁹ CD63-K2S EVs to a ~5 mm tissue slice for 3 days, loading was confirmed using IVIS. EV transport was determined by measuring tissue fluorescence before and after indicated times. Imaging occurred with a 3 second exposure using fluorescence excitation filter 570 nm and emission filter 640 nm. IVIS software (Living Image 4.0, Perkin Elmer) was used to create an ROI around the tissue pieces where the total fluorescent signal was counted.

Material preparation and hydrogel formation

Raw sodium alginates with different molecular weights (MW), low (5/60, ~40 kDa) and medium (10/60, ~120 kDa), were obtained from FMC Corporation. Alginate was purified through dialysis in a 3.5 kDa membrane submerged in water, followed by treatment with activated charcoal (Sigma) 0.5g per gram alginate. It was then filtered, frozen and lyophilized to obtain a solid polymer. Conjugation of click chemistry reagents or RGD (amino acid sequence GGGGRGDSP, Peptide 2.0) to alginate polymers was performed using a method described previously [36]. Norbornene-amine (Matrix Scientific) was conjugated to 10/60 alginate at a degree of substitution (DS) 75–150 and tetrazine-amine (Conju-Probe) was conjugated to 5/60 alginate to achieve DS18–36. For some experiments, RGD was conjugated to 10/60 alginate at DS10. Physically crosslinked hydrogels were formed as described previously [37]. Briefly, alginate solutions were mixed to be 1% 5/60 and 1% 10/60 (2% total), added to a syringe, and locked to another syringe with CaSO₄ (Sigma) to achieve final calcium concentrations of 12 mM (soft) and 20 mM (stiff). After mixing, solutions were deposited under glass for 2 hours to form a hydrogel. For covalently crosslinked hydrogels, tetrazine-alginate and norbornene-alginate were mixed to be 1% each (2% total), and deposited under glass for 2 hours to form a hydrogel. Interpenetrating network hydrogels of collagen-1 and alginate were created as described [12]. Briefly, hydrogels were prepared as physically crosslinked hydrogels, but solution was mixed with collagen-I to achieve final concentration of 0.75 or 0.375 mg/mL before mixing with CaSO4. To avoid drying, hydrogels were incubated in ‘retention medium’: HEPES-buffered saline at pH 7.75 supplemented with 2mM CaCl₂, an amount shown previously [7] to prevent leaching of calcium from hydrogels without leading to further crosslinking.

Mechanical characterization of hydrogels and tissues

Mechanical properties of hydrogels or tissues were obtained using rheometry via Anton Paar MCR302. Storage (G’) and loss (G”) moduli were measured through a frequency sweep by lowering the geometry (Anton Paar PP08) to 5% normal strain followed by rotation inducing 0.5% shear strain at increasing frequency and measurement of the resulting shear stress. Complex shear modulus G* was calculated [38]

$$G^{*}=\sqrt{{G^{\prime }}^2+{G^{″}}^2} .$$ (5)

Loss tangent was defined as

$$tan\delta =\raisebox{1ex}{$G^{″}$}\!\left/ \!\raisebox{-1ex}{$G^{\prime }$}\right. .$$ (6)

To determine stress relaxation, the geometry was lowered at constant velocity (25 μm/s) through the linear elastic region until reaching 15% strain. Swelling ratios were calculated by leaving samples to dry or swell overnight followed by mass measurements. Swelling ratio Q was calculated through volumes of hydrogels expressed as [39,40]

$$V_s=\frac{m_d}{m_s}=\frac{1}{Q};V_r=\frac{m_d}{m_r}$$ (7)

where m is the hydrogel weight and subscripts d, r and s denote dry, relaxed (before swelling) and swollen hydrogels. Average molecular weight between crosslinks was calculated as

$$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${\overline{M}}_c$}\right.=\raisebox{1ex}{$2$}\!\left/ \!\raisebox{-1ex}{${\overline{M}}_n$}\right.-\frac{\left(\raisebox{1ex}{$\overline{\overline{v}}$}\!\left/ \!\raisebox{-1ex}{$V$}\right.\right)\left[\text{ln}\left(1-V_s\right)+V_s+\chi V_s{}^2\right]}{V_r\left[{\left(\raisebox{1ex}{$V_s$}\!\left/ \!\raisebox{-1ex}{$V_r$}\right.\right)}^{\frac{1}{3}}-\raisebox{1ex}{$V_s$}\!\left/ \!\raisebox{-1ex}{$2V_r$}\right.\right]}$$ (8)

with ${\stackrel{-}{M}}_n$ the average molecular weight of polymers, $\raisebox{1ex}{$\overline{\overline{v}}$}\!\left/ \!\raisebox{-1ex}{$V$}\right.$ molar volume of hydrogel divided by the molar volume of water, and χ the Flory interaction parameter. Values were used to calculate average hydrogel mesh size ξ through the equation

$$\xi =V_s{}^{-\frac{1}{3}}{\left(\frac{2C{\overline{M}}_c}{{\overline{M}}_r}\right)}^{\frac{1}{2}}l$$ (9)

with C the polymer characteristic ratio, ${\stackrel{-}{M}}_r$ the average molecular weight of the polymer repeating unit, and l the carbon-carbon bond length. Differential scanning calorimetry (DSC) was used to perform thermoporometry to measure pore size distributions described previously [41]. Briefly, samples ~10 mg were placed in a sealable aluminum pan inside the DSC instrument (TA Instruments Q2000). Samples were cooled to −30C at a rate of 4C/min, held for 5 min, warmed to 15C at a rate of 4C/min, held for 5 min, and then cooled again to −30C at 4C/min. Distributions were calculated by determining $\raisebox{1ex}{$\mathrm{\Delta }V$}\!\left/ \!\raisebox{-1ex}{${\mathrm{\Delta }R}_p$}\right.$ [42] and then fit to a frequency-normalized histogram.

Bulk transport experiments

Liposomes (FormuMax, #F60103F-F) were obtained with a similar (~45% cholesterol, ~55% phospholipids) content as EVs [11]. Encapsulation of particles or dextran in bulk alginate hydrogels was performed by mixing particles with alginate or click-alginate followed by hydrogel formation. Hydrogels were punched into discs and placed into polystyrene plates with retention medium. If applicable, hydrogels were treated with blebbistatin (Cayman 13013) or Y-27632 (Cayman 10005583). If necessary, gels were digested by adding medium with 3.4 mg/mL alginate lyase (Sigma) and placing at 37C for 30 min. Release was measured using fluorescence for polystyrene nanoparticles (SpheroTech) and FITC-dextran (500 kDa, Sigma). Percent release was determined at the indicated times as the number of particles in the medium P_M divided by P_M plus the number of particles in the digested hydrogel P_G as

$$\%\mathit{\text{Release}}=\frac{P_M}{P_M+P_G}*100\%$$ (10)

For EVs and liposomes, P_M was measured as above using NanoSight NS300 but P_G was determined by calculating the initial number of particles added to the hydrogel using NanoSight NS300. Samples without encapsulated particles were used to account for background.

3D single particle tracking

CD63-K2S EVs were encapsulated in hydrogels, placed on dishes with No. 1.5 coverslip thickness (MatTek), and imaged at 60X with immersion oil of refractive index 1.518 (Cargille) using a DeltaVision OMX microscope (GE). Single channel 1024×1024 pixel (81.92 × 81.92 μm) images were obtained in 2μm thick stacks with 0.125μm spacing (16 images per stack) using conventional imaging mode. Thirty stacks were acquired over ~8 sec for a stack frequency of 3.75 Hz and image frequency 60 Hz. After acquisition, images were processed through deconvolution using softWoRx.

Using IMARIS ‘Spots’ function, a custom particle tracking algorithm was created. Particles were determined using intensity thresholding over regions measuring 10×10×1 pixels followed by tracking their 3D position (x, y, z) over time (t). Tracks could continue if the particle is undetectable for a single timepoint within the track but not for two or more consecutive timepoints.

Analysis of particle tracking data

Mathematical calculation and analysis were performed using MATLAB software. Particle MSD was calculated from positional data as

$$MSD\left(t\right)={\left[x\left(t\right)-x\left(t=0\right)\right]}^2+{\left[y\left(t\right)-y\left(t=0\right)\right]}^2+{\left[z\left(t\right)-z\left(t=0\right)\right]}^2 .$$ (11)

Tracks with less than five measurements of MSD were removed from further analysis. For ensemble-averaged tracks, a lower limit of 20 points and upper limit of 30 points were defined to constrain the tracks considered for analysis, as uneven track sizes can bias results [14]. Because of this, data are shown only until the lower limit of 20 points (t ~ 5 sec). To account for static (or localization) error [43], for each particle type, particles were adhered to glass using (3-aminopropyl)trimethoxysilane (APTMS, Sigma) with a method described previously [44]. The MSD was tracked for adherent particles over time, and static error was defined as the plateau MSD. This error was subtracted from all subsequent MSD measurements for each experimental group.

Ensemble-averaged track data were generated by averaging the MSD for each track i at every time t elapsed since the start of tracking

$$<MSD\left(t\right)>=\frac{1}{N}\sum _{i=1}^{N}MS{D}_i\left(t\right)$$ (12)

where N is number of tracks. Exponent α is calculated for ensemble-averaged tracks using Equation 1. Diffusion coefficient D_1.06s is calculated over intervals τ = 4Δt ~ 1.06s for each track as in Equation 3. Thus, if the total track time is T, a given track will have T/τ values for D_1.06s(τ) which are averaged to provide a singular value for D_1.06s for a given track. The expected D_1.06s for particles was determined based on the Stokes-Einstein relationship

$$D=\frac{k_{b}T}{6\pi \eta r}$$ (13)

where k_bT is the Boltzmann constant multiplied by temperature, r is particle radius, and η is solution viscosity. The viscosity of glycerol solutions was determined previously [45]. Degree of heterogeneity of D_1.06s was defined as described in the main text. For each sample, simulations were performed to obtain an equal number of simulated tracks as the number of tracks measured for each sample. Each MSD(t) was drawn randomly from a zero-mean Gaussian distribution determined for each sample with variance 2D_1.06st [14]. D_1.06s was then calculated for simulated tracks as for experimental measurements (Equation 2). ‘Cages’ of confinement were defined in the text. Tracks were evaluated for their ability to overcome this cage size by exceeding c (particles escaping) or not (particles not escaping). The timepoint which the particle exceeds c is defined as the escape time. Radius of gyration R_g was defined as the time-averaged square displacement of particles over the length of a single track as

$$R_g={[\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$N$}\right.\sum _{i=1}^{N}MSD\left(t_i\right)]}^{\frac{1}{2}}$$ (14)

over each measured timepoint t_i through the duration of the track.

ATP measurement and pharmacological depletion

ATP concentration was measured using a commercially available luciferase-based assay (Cayman, 700410). Briefly, samples were lysed followed by addition of a mixture which catalyzes a reaction to produce bioluminescence based on the concentration of ATP within samples. Values of bioluminescence were compared to a standard curve with a known concentration of ATP. To deplete ATP, cells were treated with 1 μg/mL oligomycin (Cayman, 1404-19-9) and 1 mM 2-deoxy-D-glucose (Cayman, 154-17-6) for 24 hours.

siRNA transfection

Scrambled siRNA (Dharmacon) or siRNA against AQP1 (AM16708, Ambion) was diluted to 160nM in un-supplemented Opti-MEM medium (Thermo) and combined 1:1 with Opti-MEM supplemented with 2% Lipofectamine RNAiMAX (Thermo) and incubated at RT for at least 20min. Cells were washed with HBSS, and fresh growth medium was added to cells. The transfection solution was added dropwise for a final siRNA concentration of 4nM and cells were incubated for three days followed by EV isolation.

Gene expression analysis

Trizol (Thermo Fisher Scientific) was added directly to cells. 200 μL of chloroform was added per 1 mL Trizol followed by centrifugation for 15 min at 15,000 rpm, 4C. The top layer was collected and RNA precipitated with 500 μL isopropanol for 20 min at 4C. Samples were centrifuged at 12,500 rpm for 15 min at 4C. Supernatant was removed, precipitated RNA was washed with 75% EtOH and centrifuged for 5 min at 7,500 rpm, 4C. EtOH was removed and the purified RNA was resuspended in 15 μL of RNase-free water. RNA concentration was quantified by NanoDrop. cDNA was reverse transcribed by SuperScript-III (Thermo Fisher Scientific). qPCR was performed in the ViiA7 qPCR system with PowerSYBR Green master mix (Applied Biosystem). Samples were analysed in triplicate with 50 ng of cDNA per well. Relative gene expression was computed by the delta-delta Ct method by comparing Ct values to a reference gene (GAPDH). See Table S1 for the list of primers for qPCR.

Atomic Force Microscopy

Vesicles were adhered to freshly cleaved mica by incubation at room temperature for 15 minutes followed by washing [44]. Atomic force microscopy was performed using an MFP-3D-Bio model (Asylum Research) with a pyramidal tip (Bruker; MLCT, triangular, resonant frequency: ~125 kHz) as described previously [30]. Briefly, vesicles with a size range between 50~300 nm were found by scanning in a tapping (AC) mode and indented until reaching 0.5 nN at 250 nm/s to generate a force-displacement curve. The data were analysed and converted to Young’s modulus (E) using MATLAB by modeling EVs as thin elastic shells [46]. The slope of the approach curve was calculated over a sliding interval and the surface of the vesicle was determined by a high and sustained change in slope. The linear region was used to calculate E via the equation

$$F\left(\delta \right)=\frac{aE{t}^2}{r}\delta$$ (15)

With F as measured cantilever force and δ as tip displacement. The constant $\raisebox{1ex}{$a{t}^2$}\!\left/ \!\raisebox{-1ex}{$r$}\right.$ is determined by vesicle geometry and assumed to be ~0.87nm.

Western Blot

Western blot was performed using conventional methods on samples prepared by RIPA buffer. For each lane, 20 μg of protein was added. Immunoblots were performed against AQP1 (sc-20810, SCBT, 1:2000) and GAPDH (600004–1-Ig, Proteintech, 1:5000) using an anti-rabbit or anti-mouse HRP-conjugate secondary antibody (rabbit: 115-035-003; mouse: 115-035-071, Jackson ImmunoResearch Laboratories) combined with Luminol (Santa Cruz) substrate for detection.

Statistical evaluation

Statistics were performed as described in figure captions. All statistical analyses were performed using GraphPad Prism version 8.1.1. Unless otherwise noted, statistical comparisons were made from at least three independent experiments by one-way ANOVA followed by Tukey’s multiple comparison test, and then were considered significant if p < 0.05.

Data availability:

The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request.

Code availability:

The codes used to analyse the data in this study are available from the corresponding author upon reasonable request.

Supplementary Material

Video 1

Tracking data overlaid with imaging data for representative transport of a single EV in stiff stress relaxing matrix shown in Fig. 3A. The length scale is micrometers and the time scale is seconds.

Video 3

Tracking data overlaid with imaging data for representative transport of a single EV in stiff elastic matrix shown in Fig. 3A. The length scale is micrometers and the time scale is seconds.

Video 2

Tracking data overlaid with imaging data for representative transport of a single EV in soft stress relaxing matrix shown in Fig. 3A. The length scale is micrometers and the time scale is seconds.

Video 5

Tracking data overlaid with imaging data for representative transport of multiple EVs in soft stress relaxing matrix. The length scale is micrometers and the time scale is seconds.

Video 4

Tracking data overlaid with imaging data for representative transport of multiple EVs in stiff stress relaxing matrix. The length scale is micrometers and the time scale is seconds.

Video 6

Tracking data overlaid with imaging data for representative transport of multiple EVs in stiff elastic matrix. The length scale is micrometers and the time scale is seconds.