Human BMSCs (herein denoted as BMSC 1–3) were isolated from bone marrow samples obtained from three donors (Supplementary Table S1) undergoing orthopedic surgery at the Tampere University Hospital Department of Orthopedics and Traumatology. Bone marrow samples were obtained with the donor’s written informed consent and processed under ethical approval of the Ethics Committee of the Expert Responsibility area of Tampere University Hospital, Tampere, Finland (R15174). The cells were isolated as described previously (Wang et al., 2019). The BMSCs were cultured in α-modified minimum essential medium (α-MEM; Gibco) supplemented with 5% human serum (HS; BioWest or Serana), 100 U/ml penicillin, 100 µg/ml streptomycin (Pen/Strep; Lonza), and 5 ng/ml basic fibroblast growth factor (bFGF; Miltenyi Biotec) and used between passages 3 and 6.

Human ASCs (herein denoted as ASC 1–3) were isolated from subcutaneous abdominal tissue samples obtained from three donors (Supplementary Table S1). Tissue samples were obtained at the Tampere University Hospital Department of Plastic Surgery with the donor’s written informed consent and processed under ethical approval of the Ethics Committee of the Expert Responsibility area of Tampere University Hospital (R15161). The cells were isolated as described previously (Kyllönen et al., 2013). The ASCs were cultured in α-MEM supplemented with 5% HS (BioWest, PAA or Serana), 100 U/ml penicillin, and 100 µg/ml streptomycin and used between passages 3 and 5. The mesenchymal origin of BMSCs and ASCs was confirmed by surface marker expression analysis with flow cytometry (Supplementary Table S2) and assessment of adipogenic and osteogenic differentiation potential (Supplementary Figure S1).

Pooled human umbilical vein endothelial cells (HUVECs) expressing green fluorescent protein (GFP) were commercially obtained from Cellworks. GFP-HUVECs were cultured in Endothelial Cell Growth Medium-2 Bullet Kit (EGM-2; Lonza) consisting of Endothelial Cell Growth Basal Medium (EBM-2) and Endothelial Cell Growth Medium-2 Supplements. Instead of the fetal bovine serum supplied with the Kit, 2% HS (BioWest) was used. The cells were used between passages 3 and 6.

Microvascular networks were formed inside microfluidic chips (DAX-1; Aim Biotech), consisting of a gel channel (1.3 mm wide and 250 µm high) flanked by two medium channels. Microvascular networks were formed through self-assembly of GFP-HUVECs and MSCs in fibrin hydrogel according to Jeon et al. (2014) with modifications. Briefly, GFP-HUVECs were combined with either BMSCs or ASCs at 5:1 cell ratio (final cell density 6 million cells/ml) and spun down. The cell pellet was then resuspended in 2 IU/ml human thrombin (Sigma-Aldrich, St. Louis, MI, United States) in EBM-2 and combined at 1:1 volume ratio with 5 mg/ml human fibrinogen (Sigma-Aldrich, St. Louis, MI, United States) in Dulbecco’s Phosphate Buffered Saline (DPBS; Lonza). The resulting mixture (10 µl) was quickly injected into the gel channel of the microfluidic chip and allowed to polymerize in a humidified chamber for 15–20 min at room temperature (RT). Microvascular networks were cultured in EGM-2 with 2% HS (BioWest). To enhance vascular network formation, interstitial flow was used throughout the culture period (Kim et al., 2016; Abe et al., 2019). The interstitial flow was generated by adding 90 µl of culture medium into both reservoirs of one medium channel and 50 µl of medium into both reservoirs of the other medium channel. This procedure was repeated daily when changing medium and the flow direction was kept constant. In order to prevent tracer leakage from the medium channel into the hydrogel matrix and improve microvascular network perfusability, GFP-HUVECs were seeded into the medium channels of microfluidic devices used for bead flow assay and immunostainings at day 3 or 4 according to Campisi et al. (2018). To take into account the donor variability, cells derived from three different donors for each MSCs type (BMSCs and ASCs) were tested. GFP-HUVECs cultured without any supporting cells (hereafter referred to as “EC alone”) were prepared and cultured at a final cell density of 5 million cells/ml in the same manner and served as a control for evaluating microvasculature parameters and vasculature perfusability.

The immunostaining protocol was adapted from Honkamäki et al. (2021) and performed on 2–4 technical replicates for each cell donor. After 6 days of culture microvascular networks were washed with DPBS and fixed with 4% paraformaldehyde (Sigma-Aldrich) in DPBS for 30 min at RT following by three washes with DPBS. The samples were then permeabilized with 0.1% Triton X-100 (Sigma-Aldrich, St. Louis, MI, United States) for 10 min and non-specific binding was blocked with blocking buffer [10% normal donkey serum (NDS; Millipore), 1% bovine serum albumin (BSA; Sigma-Aldrich, St. Louis, MI, United States), 0.1% Triton X-100 in DPBS] for 2 h at RT. We used primary antibodies against vascular endothelial cadherin (VE-cadherin; BD Pharmingen, 555,661; 1:200) to identify the presence of endothelial adherens junctions, collagen type IV (Sigma-Aldrich, C1926-.2ML; 1:150) as a marker of the basement membrane, and alpha-smooth muscle actin (α-SMA; Abcam, ab7817; 1:300) and PDGFR-beta (PDGFR-β; Abcam, ab32570; 1:100) as perivascular cell markers. Primary antibodies were diluted in PBS with 1% NDS, 1% BSA, 0.1% Triton X-100 and applied for 3 days at 4°C. The samples were then washed for at least 6 times with 1% BSA in DPBS over a period of 2 days in total. Alexa Fluor 568 conjugated secondary antibodies (Thermo Fisher, A11031 and A10042) diluted 1:400 in DPBS with 1% BSA were applied overnight at 4°C. The samples were washed 5 times over a period of 2 days, stained with 0.67 µg/ml DAPI for 2 h at RT followed by three washes in DPBS. The chips were imaged using a Zeiss LSM 800 laser scanning confocal microscope with high-sensitivity two-channel spectral PMT detector and LSM 780 laser scanning confocal microscope with a Quasar spectral GaAsP detector (all from Carl Zeiss) and processed with Fiji software (Schindelin et al., 2012) and Adobe Photoshop 2020. Negative controls for immunocytochemical stainings are shown in Supplementary Figure S7.

To assess perfusability of the microvascular networks, 2 µm yellow-green fluorescent microbeads (FluoSpheres Carboxylate-Modified Microspheres, 2% solids, Thermo Fisher) were introduced as a fluorescent tracer into microfluidic devices containing living cells at day 7 of culture and time-sequential images were captured. Briefly, medium was aspirated from all reservoirs, 25 µl of sterile water was added into the reservoirs of one channel and 50 µl was added into the reservoirs of the other channel followed by 50 µl of fluorescent microbeads diluted 1:2,000 in sterile water. Sterile water was used instead of DPBS to avoid the formation of microbead aggregates. The generated hydrostatic pressure between the two medium channels resulted in luminal flow through the microvascular network. The time-sequential images were acquired every 10 s during 3 min using EVOS FL cell imaging system (Thermo Fisher, Nungambakkam, TamilNadu). Animated GIF images were then generated using Fiji software.

Microvascular networks morphology was analyzed using Fiji software. The GFP signal from the endothelial cells was used to quantify vascular area percentage, vascular network length, and average vessel diameter. The samples were fixed at day 6 and confocal z-stacked images were captured with 5 µm step (45 slices) using a Zeiss LSM 800 confocal microscope. Three regions of interest (ROIs; 798,63*798,63 µm²) per chip were selected from the co-culture area excluding the hydrogel-restricting microposts and solely including microvascular networks. Quantitative image analysis was not performed blinded. Images were processed as described previously (Jeon et al., 2014; Campisi et al., 2018; Haase et al., 2019) with minor modifications. Briefly, the maximum intensity projections were generated, brightness and contrast were adjusted by the “window/level” tool (35/70), noise was filtered by applying “Gaussian blur” filter (sigma 3), and images were converted to a binary format by applying the “triangle” threshold method. Artefacts were removed using the “erode” and “remove outliers” (radius 5 pixels for bright and 10 pixels for dark) functions. All values were calculated per ROI. The vasculature area percentage and the total vasculature area were computed using “measure” function. The 2D “skeletonize” function was applied to calculate the total network length. Finally, the average diameter was computed by dividing the total vasculature area by the total network length. Three chips per donor cell line and 3 ROIs per device were used for quantified analysis and statistical testing.

Perivascular distribution of MSCs was analyzed using Fiji software and quantified through the α-SMA signal in terms of pericyte area percentage and pericyte coverage. Immunofluorescent stained chips that had been fixed after 6 days of culture, were imaged completely with the Zeiss LSM 780 confocal microscope using Z stacks with 5 µm steps (41 slices). Maximum intensity projections were prepared and five ROIs (750*750 µm²) per chip were selected. Quantitative image analysis was not performed blinded. The pericyte area percentage was quantified from the α-SMA channel as described above for vascular parameters with minor adjustments. The pericyte coverage was quantified as the overlapping area of α-SMA and GFP signals (Erickson et al., 2020). For that, an “AND” operation of Fiji image calculator was performed on α-SMA and GFP channels of the same ROIs pre-processed separately. The vasculature coverage by pericytes was calculated by dividing the computed values by the total vasculature area within each ROI. Three chips per donor cell line and 5 ROIs per device were analyzed.

To characterize the gravity-driven flow generated daily, interstitial flow across the hydrogel area was measured in EC-BMSC and EC-ASC cocultures using fluorescence imaging. The flow measurements of EC-BMSCs and EC-ASCs were performed at days 0, 2, 4 and 7 using initial volumes of gravity-driven flow mimicking the flow condition. Measurements were performed at +37°C. First, medium was aspired from all reservoirs, 50 µl of 1xDPBS was added into two of the medium reservoirs (Figure 1A) followed by the addition of 90 µl of 100 µg/ml 70 kDa Rhodamine B isothiocyanate–Dextran (R9379, Sigma) in 1xDPBS to the two other reservoirs located opposite sides of the hydrogel area (Figure 1A). This generated a time-dependent hydrostatic pressure inside the chip resulting in a gravity-driven interstitial flow across the hydrogel area. To quantify the interstitial flow, sequential images of the fluorescent flow front across the hydrogel area were acquired by using a wide-field fluorescence microscope (Olympus IX 51) with sCMOS camera (Orca Flash4.0LT, Hamamatsu) using 5 seconds measurement intervals (Figure 1B). Each chip was manually placed on the microscope so that the center of the culture area was imaged. All measurements were performed as independent biological replicates N = 2 for each flow condition per coculture of one BMSC/ASC donor cell line.

Image analysis was performed using MATLAB (Version R2018b, The MathWorks, Inc., Natick, MA, United States) and scripts made in-house. First, stack images (.vsi) were read using BioFormats toolbox, greyscale images were binarized using 0.5 threshold value, and average locations of the flow front were tracked from the binarized images (Figure 1B). Based on differences in flow front locations between two consecutive indexes, known pixel size (0.615 µm), and known time interval between these two image indexes, an average flow velocity was calculated at each time point (Figure 1C, Supplementary Figure S6). An average flow rate at each time point was then calculated by multiplying the average velocities and the cross section of the hydrogel area (10,8 mm × 250 µm) (Figure 7). It should be noted that the accuracy of the developed analysis decreases within lower flow velocity, as smaller flow front movement between two consecutive image indexes is measured. This results in oscillation of the measured flow rate on day 4 (in Figure 7C), in which the flow rate was measured remarkably low (below 0.5 µl/min).

In the gravity-driven system, when assuming zero capillary forces, pressure difference Δp(t) between inlet and outlet reservoirs is based on time-dependent hydraulic pressure p _hyd (t). As described earlier, fluorescent dextran solution was used as a tracer for determining the spatial change of the flow front preventing the possibility to record fluorescent intensity differences after the complete saturation of the hydrogel area. However, the constantly decreasing time-dependent gravity-driven flow rate thus flow duration can be estimated using following equations (Mäki et al., 2015). First, hydraulic pressure p _hyd (t) is based on the liquid plug height difference Δh(t) between inlet and outlet reservoirs and can be calculated by Δ p ( t ) = p h y d ( t ) = ρ g Δ h ( t ) where ρ density of liquid and g is gravitational acceleration.

Hydraulic resistance of the system, R _hyd , was approximated based on measured initial flow rate from [(90 µl + 90 µl)+(50 µl + 50 µl)] volume condition and estimated initial hydraulic pressure p _hyd (0) using following equation. Here, it is assumed that R _hyd is constant during the experiment. Δ p ( t ) = R h y d Q ( t ) ⇒ R h y d = Δ p ( 0 ) / Q ( 0 )

Then, the time-dependent flow rate Q(t) can be estimated using equation Q ( t ) = p h y d ( t ) R h y d = ρ g Δ h ( t ) R h y d = ρ g Δ h ( t ) ρ g Δ h ( 0 ) Q ( 0 ) = Δ h ( t ) Δ h ( 0 ) Q ( 0 )

In gravity-driven flow, Δh(t) is continuously decreasing as liquid flows from the inlet reservoirs to the outlet reservoirs, thus reducing the hydraulic pressure and the flow rate. With known reservoir dimensions (a cylinder shape with a diameter of 5 mm), the plug height difference between inlet and outlet reservoirs at different time points can be calculated. This equation was solved using Simulink (The MathWorks). In theory, gravity-driven flow would continue infinitely long time as the flow rate is constantly decreasing, but in practice, flow will be stopped due to small friction losses. To account this issue, a minimum flow rate threshold was set to 0.02 µl/min thus ending the simulation when the flow rate dropped below this threshold value. The simulation end-point time was then used as an estimation of total flow rate duration.

Shear stress τ induced by the interstitial flow velocity v can be estimated using equation (Yang et al., 2021) τ ≈ μ v K p in which K _p = 6.67 × 10⁻¹² m² is the permeability of the hydrogel material (Bachmann et al., 2018) and µ is the dynamic viscosity of the flow medium. Here, medium was assumed to have water-like properties at 37°C, thus µ = 0.6913 × 10⁻³ Pa s was used.

At day 6 of culture, the medium was aspirated from all the media ports and chips were stored frozen at −80°C prior RNA extraction. The total RNA of EC-BMSC and EC-ASC co-cultures were extracted using TRIzol reagent (Invitrogen). Fibrin matrices containing cells were flushed out from the chips with TRIzol reagent and three chips were pooled for each BMSC/ASC donor cell line to collect higher amount of RNA. The samples were processed according to manufacturer’s instructions and RNA concentration was measured with a spectrophotometer (Nanodrop 2000; Thermo Fisher). Complementary DNA was synthesized using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems) and quantitative real time reverse transcription polymerase chain reaction (qRT-PCR) was performed with QuantStudio 12K Flex Real-Time PCR System (Applied Biosystems) using approximately 20 ng cDNA per reaction, TaqMan Fast Advanced Master Mix (Thermo Fisher) and Custom TaqMan Array Plates (Applied Biosystems) according to manufacturers’ instructions. TaqMan assay probes for genes of interest are listed in Supplementary Table S3. All measurements were performed using technical triplicates. 18S ribosomal RNA (18S), glyceraldehyde 3-phosphate dehydrogenase (GAPDH) and platelet and endothelial cell adhesion molecule 1 (PECAM1) were used as reference genes for normalizing qRT-PCR data. Gene expression analysis was performed for each EC-MSC co-cultures of three donor cell lines for both BMSCs and ASCs. Relative quantification of gene expression was performed using the 2^−ΔΔCt method (Livak and Schmittgen, 2001). Gene expression levels are shown relative to the average expression in EC-ASC co-cultures. Statistical analysis was performed on the mean ΔΔCt values from each donor cell line.

Heatmaps were generated from the mean ΔCt values for each donor cell line using the gplot package in R (Warnes et al., 2015). Hierarchical clustering and dendrograms were computed based on Euclidean distances.

All data are represented as mean ± SD. Statistical analysis was performed using GraphPad Prism 9.0.0 (GraphPad Software, www.graphpad.com). Unpaired student’s T-tests were used for significance testing between BMSCs and ASCs. p values were corrected to control the false discovery rate using the two-stage step-up method of Benjamini, Krieger and Yekutielio (Benjamini et al., 2006). The differences were considered statistically significant for p < 0.05 and represented as follows: * denotes p < 0.05, ** ‒ p < 0.01, *** ‒ p < 0.001, n. s ‒ non-significant.

The data sets generated and analyzed for the current study are available from the corresponding author on reasonable request.

We examined two types of human primary MSCs—BMSCs and ASCs, for their ability to support microvascular network formation and maturation in fibrin matrix and act as perivascular cells when co-cultured with ECs in a microfluidic chip environment. First, we examined morphological characteristics including 3D structure, lumen-containing vessel formation, basement membrane deposition, EC-EC cell junctions of the microvascular networks formed by EC-BMSC and EC-ASC co-cultures. By day 6 both EC-BMSC and EC ASC co-cultures gradually developed into lumenized interconnected vascular networks with randomly aligned morphology spanning the entire hydrogel (Figure 2A). However, the microvascular networks formed by EC-BMSC co-cultures were more organized, interconnected, and denser compared to EC-ASC co-cultures. Daily observations of the microvascular network formation revealed minor vasculature degradation in the middle part of the chip in EC-ASC co-cultures starting from day 5 (Supplementary Figure S2).

Importantly, cross-sectional images of the vessel segments confirmed the presence of hollow lumens in both EC-BMSC and EC-ASC co-cultures as well as in EC alone (Figure 2B). In contrast to MSCs supported co-cultures, ECs cultured alone formed multicellular aggregates of different sizes which eventually developed into primitive structures containing lumens but failed to form interconnected microvascular networks. Notably, unlike in EC-BMSC and EC-ASC co-cultures, not all ECs were engaged in vasculature formation when ECs were cultured alone.

We studied vascular maturation further by immunofluorescence staining for collagen type IV, a predominant vascular basement membrane component (LeBleu et al., 2007), and for VE-cadherin, a specific marker for endothelial cell-cell adherens junctions (Giannotta et al., 2013). Collagen type IV was deposited in the perivascular area surrounding the vessels in both EC-BMSC and EC-ASC co-cultures (Figure 3A). In EC-ASC co-cultures, however, collagen type IV staining was weaker and less prominent (Supplementary Figure S3). Moreover, here we detected regions of collagen type IV-positive basement membrane surrounding an empty lumen without ECs. VE-cadherin was strongly expressed at the intercellular contacts along the EC border in both EC-BMSC and EC-ASC co-cultures (Figure 3B) demonstrating formation of adherens junctions between ECs. No differences were observed between EC BMSC and EC-ASC co-cultures with VE-cadherin staining intensity.

Next, we examined perfusability of the microvascular networks formed by EC-BMSC and EC-ASC co-cultures, and ECs alone. The perfusability was assessed using 2 µm fluorescent microbeads after 7 days of culture. Microbeads were loaded into one of the medium channels and hydrostatic pressure drop was created between two medium channels so that the luminal flow across the microvascular network allows microbeads to travel through the established microvasculature. Only vessels that developed open lumens toward the media channels at the hydrogel-medium interface were able to carry beads through the network. To facilitate formation of microvasculature with open lumens, the ECs were also seeded into both medium channels of the microfluidic devices at day 3 or 4 and allowed to adhere to the hydrogel-medium interface and connect to the vascular network. With visual observation, we detected a great improvement in perfusability of the microvascular networks (data not shown). All three EC BMSC co-cultures resulted in formation of perfusable microvascular networks with multiple entry points upstream of the luminal flow and several exit points downstream of the luminal flow (Supplementary Video S1). In contrast, microvascular networks resulting from EC-ASC co-cultures demonstrated poor perfusability regardless of ASC line. Two ASC lines induced formation of partially perfusable microvessels (in only one chip out of 3 tested) with a couple entry points and only one exit point for microbeads (Supplementary Video S2). As expected, EC alone did not form perfusable microvascular networks as shown by bead flow assay (Supplementary Video S3).

Quantifying microvascular network parameters revealed a significantly larger vasculature area, wider vessels, and greater total vasculature length in EC-BMSC co-cultures compared to EC-ASC co-cultures (Figure 4). Microvessels occupied 52% (46–60%) of area in EC-BMSC co-cultures and 31% (28–35%) of area in EC-ASC co-cultures, while those occupied 24% of area in EC alone. As expected, MSCs supported microvascular networks covered larger area than ECs alone. EC-BMSC co-cultures formed microvascular networks with an average vessel diameter of 57 µm (54–60 µm) and a total vasculature length of 5,877 µm (5,288–6,373 µm). The average vessel diameter in EC-ASC co-cultures was 45 µm (41–48 µm) and the total vessel length 4,411 µm (4,048–4,731 µm). The average vessel diameter in EC monocultures was 49 µm and the total vessel length 3,072 µm.

To investigate the spatial distribution of BMSCs and ASCs as perivascular cells, the EC-BMSC and EC-ASC co-cultures were stained for known perivascular cell markers (Armulik et al., 2011; Holm et al., 2018; Sweeney and Foldes, 2018) PDGFR-β and α-SMA after 6 days of culture. Immunofluorescence staining showed that both BMSCs and ASCs express PDGRF-β and α-SMA when co-cultured with ECs (Figures 5A,B). Nearly all BMSCs and ASCs stained positively for perivascular cell marker PDGFR-β. Interestingly, BMSCs were found in close proximity to the vessel structures while ASCs were more frequently localized in areas distant from the vessels. On the contrary, immunostaining for α-SMA differed notably between EC-BMSC and EC-ASC co-cultures. A large fraction of BMSCs stained positively for α-SMA, whereas only few ASCs expressed α-SMA. The α -SMA positive MSCs had elongated morphology similar to in vivo and were wrapped around the microvessels or localized closely to the ECs in both EC-BMSC and EC-ASC co-cultures (Figures 5B,C).

We further quantified MSCs transition toward perivascular cells using the α-SMA signal in terms of pericyte area percentage and pericyte coverage (Figures 6A,B). Quantification revealed that α-SMA positive BMSCs occupied an area (40%; 36–46%) that was more than 7-fold larger than the area occupied by α-SMA positive ASCs (6%; 3–8%) indicating stronger potential of BMSCs to function as perivascular cells compared to ASCs. We found that in EC-BMSC co-cultures, 54% (45–64%) of the vascular network was covered by α-SMA positive perivascular cells whereas in EC-ASC co-cultures – 12% (9–14%), suggesting that BMSCs have more potential to interact with and stabilize microvascular networks.

To demonstrate the presence of gravity-driven interstitial flow, we estimated the average flow rates by measuring spatial change of the wave front and computed the flow duration within the microfluidic chip using the gravity-driven model, both presented in Section Characterization of the Gravity-Driven Flow in the Microfluidic Chip (Figure 7). We verified the presence of gravity-driven flow across the hydrogel area within the microfluidic chip. At day 0, approximated initial flow rates Q (0) were 3.1 µl/min in EC-BMSC culture and 3.2 µl/min in EC-ASC culture (Figure 7A). Based on this, the estimated flow durations, where Q(t) > 0.02 µl/min, were approximately 1 h in both co-cultures (Figure 8). At day 2, measured Q (0) were approximately 1,1 µl/min and 0.55 µl/min (Figure 7B), prolonging the estimated flow durations from 1 h to two and 4 h for EC- BMSC and EC-ASC, respectively (Figure 8). At day 4, Q (0) were 0,35 µl/min and 0,25 µl/min (Figure 7C), resulting in >5 h and >6 h of estimated total flow duration for EC- BMSC and EC-ASC cultures, respectively (Figure 8). Flow characterization measurement for day 7 demonstrated unequal fluorescence wave front as we observed luminal flow via the formed vascular network (Supplementary Figure S6). The measured maximum interstitial flow velocity was v ≈ 0.024 mm/s, resulting in maximal shear stress τ ∼ 0.0065 Pa, which falls within the physiological shear stress range (Yang et al., 2021).

To investigate differences in gene expression between EC-BMSC and EC-ASC co-cultures, we performed qRT-PCR analysis for both co-cultures considering pericyte-specific (ACTA2, PDFGRB, CSPG4) and endothelial-specific (PECAM1, CDH5, VWF) genes, basement membrane protein (COL4A1), angiogenic growth factors (VEGFA, FGF2), and other genes involved in vasculature morphogenesis and stability (KDR, FLT1, ANGPT1, ANGPT2) (Figure 9, Supplementary Figure S4). The mRNA expression of ACTA2, PDFGRB, CSPG4, PECAM1, COL4A1, VEGFA, FGF2, and ANGPT1 in both co-cultures were measured relative to 18S and GAPDH. To assess changes in the number of ECs between EC-BMSC and EC-ASC co-cultures, we measured the mRNA expression of PECAM1 relative to 18S and GAPDH. Expression of endothelial cell specific-genes VWF, CDH5, KDR, FLT1 and ANGPT2 were measured relative to PECAM1 (Supplementary Table S3, Supplementary Figure S4) in both co-cultures. The qRT-PCR analysis revealed a significantly higher expression of pericyte marker genes ACTA2 and CSPG4 (9.5-fold and 4.5-fold, respectively), also known as α-SMA and NG2, in EC-BMSC co-cultures compared to EC-ASC co-cultures (Supplementary Figure S4A). The changes in the expression of PDGFRB were not significant between EC-BMSC and EC-ASC co-cultures. Furthermore, expression levels of COL4A1, encoding for the major basement membrane protein, were 4-times higher in EC-BMSC co-cultures compared to EC-ASC co-cultures (Supplementary Figure S4C). The qRT-PCR results confirmed our observations from the immunofluorescence stainings for α-SMA, PDGFR-β, and collagen type IV.

The expression of PECAM1, an endothelial marker gene, was markedly higher (3.3-fold change) in EC-BMSC co-cultures (Supplementary Figure S4B). Minor differences were found in the expression of other EC-specific (CDH5 and VWF) genes (Supplementary Figure S4B). The expression of CDH5 encoding VE-cadherin was 1.3-fold higher in EC-BMSC co-cultures than in EC-ASC co-cultures. The expression of main angiogenic growth factor genes, VEGFA and FGF2, did not differ between the co-cultures (Supplementary Figure S4D). In addition, changes in gene expression of VEGFA receptors were either not significant (KDR) or slightly decreased in EC-BMSC co-cultures (FLT1, 0.8-fold change) (Supplementary Figure S4E). Considering expression of ANGPT1 and ANGPT2 genes, main regulators of vessel stabilization and destabilization, we found no difference in ANGPT2 expression between EC-BMSC and EC-ASC co-cultures (Supplementary Figure S4E). Interestingly, the ANGPT1 expression was consistently more than 2.4-times increased in EC-BMSC co-cultures. However, due to donor variation in EC-ASC co-cultures (0.5-2-fold change), the changes were not significant.

Hierarchical gene clustering showed that microvascular networks supported by BMSCs and ASCs had distinct gene expression patterns (Supplementary Figure S5). Even though expression of genes normalized by PECAM1 (ANGPT1, CDH5, KDR, FLT1 and VWF) was similar in EC-BMSC and EC-ASC co-cultures, BMSCs and ASCs still formed their own clusters.