Stromal vascular fraction (SVF) cells were isolated as previously described (Guven et al., 2011). Lipoaspirates or excision fat samples were obtained during routine liposuction, after informed consent and following protocol approval by the ethical committee of the local Government (Permit number 78/07 of the Ethikkommission beider Basel, Kanton Basel-Stadt, Basel, Switzerland). SVF cells were isolated from 6 healthy donors ranging in age from 28 to 56 years. Adipose tissue samples were digested in 0.075% collagenase type 2 (Worthington) on an orbital shaker for 45 min at 37°C. The digested tissues were centrifuged at 300 g for 10 min. The pellet, containing the stromal vascular fraction (SVF), was rinsed once with PBS (Gibco), resuspended in alpha-minimal essential medium (α-MEM, Gibco) and filtered through a 100 μm and 70 μm strainer (BD Falcon). SVF nucleated cells were counted with 0.01% Crystal Violet (Sigma) in PBS.

The amino acid sequence of mouse Sema3A (NCBI Reference Sequence: NP_033178.2) is 96% identical with that of human Sema3A (NCBI Reference Sequence: NP_006071.1). Further, signalling activity has been shown to be widely conserved across different species (mouse, human, rat and chicken). To produce the engineered cross-linkable form of mouse Semaphorin 3A (Sema3A), before insertion into the expression vector pXLG (provided from the Protein Expression Core Facility (PECF) of the Ecole Polytechnique Federale de Lausanne (EPFL), Switerland), the cDNA for mouse Sema3A was amplified by PCR using primers designed to add the transglutaminase (TG) substrate sequence NQEQVSPL, including the 8 N-terminal residues of α2-plasmin inhibitor (α2PI_1–8, hereafter referred to as TG-), onto the N-terminus of the amplified cDNA (Schense et al., 2000). An 8x histidine was then also added at the N-terminus. The engineered protein was expressed in mammalian cells, Human Embryonic Kidney (HEK)-293 F (ThermoFischer Scientific, Waltham MA, United States). Cells were transfected with the TG-Sema3A containing plasmid using PEI-Max and cultured for 7 days under shaking conditions at 37°C, 5% CO₂ in FreeStyle 293-F medium. The supernatant was then collected and purified using an HisTrap column and Äkta Pure FPLC machine (GE Heatlhcare, Chicago IL, United States) using standard procedures according to the manufacturer. Sema3A monomers and dimers were then separated using size exclusion with a HiLoad 16/60 Superdex 75-pg column (GE healthcare, Chicago IL, United States). Fractions corresponding to Sema3A dimers were pooled, concentrated with Amicon tubes and filtered through a 0.22 μm filter. Sema3A dimers purity was assessed by SDS/PAGE and found to be >99%. Endotoxin level was under 0.05 EU/mg of protein after measurement with the HEK-Blue mTLR4 assay (Invivogen, San Diego, California, United States).

The engineered cross-linkable form of aprotinin was produced as previously described (Burger et al., 2022). Briefly, the cDNA for bovine aprotinin was modified adding an 6x histidine tag and a thrombin cleavage site at its N-terminus, and the cDNA of the TG substrate sequence NQEQVSPL at its C-terminus. The cDNA was subcloned into a pXLG vector for the expression in HEK-293F mammalian cells. HEK-293F cells were cultured in suspension for 7 days after transient transfection with polyethylenimine (PEI). The supernatant was then collected and purified via immobilized metal affinity chromatography (HisTrap columns, GE healthcare, Chicago, Illinois, United States) using an Äkta Pure FPLC system (GE healthcare, Chicago, Illinois, United States). The purified proteins were then dialyzed and the histidine tag was cleaved using thrombin (50 U/mg protein) for 24 h at room temperature. The proteins were purified again using HisTrap and benzamidine columns (GE healthcare, Chicago, Illinois, United States) to remove the cleaved his-tag and thrombin. The purified TG-aprotinin was dialyzed against TBS, concentrated using Amicon tubes, filtered through a 0.22 μm filter and stored at −80°C.

Engineered constructs were generated using 60 mm³ of silicate-substituted calcium phosphate granules (1–2 mm in size, Actifuse^®; Apatech-Baxter, Elstree, United Kingdom), 2 × 10⁶ of SVF cells and 60 µL fibrin gel. The fibrin hydrogel was prepared by mixing 25 mg/mL human fibrinogen (plasminogen-, von Willebrand Factor-, and fibronectin-depleted; Milan Analytica AG, Rheinfelden, Switzerland), 3 U/mL factor XIII (CSL Behring, King of Prussia, Pennsylvania, United States), and 6 U/mL thrombin (Sigma-Aldrich, St. Louis, Missouri, United States) with 2.5 mM Ca²⁺ in 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (Hepes, Lonza, Basel, Switzerland). The engineered constructs displayed a spherical shape with a diameter of about 6 mm. Decoration with 51 μg/mL of TG-Aprotinin and 0.1 or 100 μg/mL of TG-SEMA3A were obtained by adding the engineered proteins to the cross-linking enzymes solution before mixing with fibrinogen. Fibrin was allowed to polymerize at 37°C for 10 min after mixing with all the other components before in vivo implantation. The engineered constructs were implanted subcutaneously (4 constructs/animal) in 5-week old female nude mice (CD1-Foxn1 ^nu , Charles-River, Sulzfeld, Germany). Animals were treated in agreement with Swiss legislation and according to a protocol approved by the Veterinary Office of Canton Basel-Stadt (permission #1797). After 1, 6 and 12 weeks, mice were sacrificed by inhalation of CO₂ and constructs were explanted.

After harvesting, constructs were washed with PBS and fixed overnight at 4°C with freshly prepared 1% paraformaldehyde (Sigma-Aldrich, St. Louis, Missouri, United States) in PBS. They were then decalcified in a solution containing 7% w/v EDTA (0.5M, pH 8, Sigma-Aldrich, St. Louis, Missouri, United States) and 10% w/v sucrose (Sigma-Aldrich, St. Louis, Missouri, United States) and incubated at 37°C on an orbital shaker. The decalcification solution was changed daily until the samples were fully decalcified, as estimated by the degree of sample stiffness (for a maximum of 20 days). Finally, the samples were embedded in OCT compound (CellPath LTD., Newtown, United Kingdom), frozen in freezing 2-methylbutane (isopentane) (Sigma-Aldrich, St. Louis, Missouri, United States) and sectioned with a cryostat with a thickness of 10 µm.

The following primary antibodies were used for immunofluorescence staining: rat anti-mouse CD31 (clone MEC 13.3, BD Bioscience, San Jose, California, United States) at 1:100; mouse anti-human CD31 (clone JC70, Santa Cruz, Dallas, United States) at 1:200; rat anti-mouse CD11b (clone M1/70, Thermofisher, Waltham, Massachusset, United States) at 1:200; rabbit polyclonal anti-Neuropilin 1 (Abcam, Cambridge, United Kingdom) at 1:200, rabbit polyclonal anti-Laminin (Abcam, Cambridge, United Kingdom, 11575) at 1:200. Secondary antibodies labeled with fluorescent fluorochromes (Invitrogen, Thermo Fisher Scientific, Waltham, Massachusetts, United States) were used at 1:200.

Images were acquired with a Nikon Ti2 Eclipse microscope (Nikon, Tokyo, Japan) and a Nikon Ti2 inverted microscope (Nikon, Egg, Switzerland) equipped with a X-Light V3 Spinning disk confocal unit (CrestOptics) (50 μm pinhole size with 250 μm spacing), a Celesta Light Engine (LumenCor), and a Prime 95B (Photometrics) camera. All image measurements were performed with FIJI software (ImageJ, http://fiji.sc/Fiji), NIS-Elements (Nikon, Tokyo, Japan) and QuPath software (v 0.4.4 https://qupath.github.io/). All subsequent quantifications were performed by 3 independent observers who were blinded to the sample identities.

Immunostaining for CD31 was used to asses vascular invasion and vascular density. Reconstructions of whole sections from the central part of each sample were acquired and the invaded area was measured by tracing the area occupied by mouse blood vessels (mCD31⁺ structures) and expressed as percentage of the total graft area. To quantify vessel density at week 1, at least 15 images were acquired per sample within the invaded areas and vessel length density (VLD) was measured tracing manually the total length of vessels, distinguishing those of mouse and human origin, in the fields and by normalizing it to the field area (mm/mm²). The same images were also used to quantify the amount of perfused vessels, which were identified by the presence of red blood cells (RBC) within the lumen. VLD at 6 and 12 weeks was quantified on 15 randomly acquired images, covering all the area of the tissue section, since constructs were completely invaded by these time-points. Every quantified image covered a field of 0.3 mm² and therefore a total area of 4.5 mm² was quantified for each sample. Since the total area of the histological section through the center of each graft had an area of approximately 20–25 mm², about 20% of the total area was quantified in each sample, in order to minimize the effects of non-homogeneous vessel distribution within the constructs.

The number of NEM recruited to the constructs after 1 week in vivo was quantified by immunostaining for NRP-1 in combination with a monocyte marker (CD11b) and the DNA dye DAPI. Images of whole sections for each condition were acquired and a concentric layer of 1 mm depth from the external surface was traced. NRP⁺/CD11⁺ cells were automatically detected and counted by using the open-source software QuPath v0.4.4. The DAPI channel was used for cell identification and pixel classifiers were generated for the NRP-1 and CD11b channels. A composite classifier was created for the colocalization of the three individual channels.

Data are presented as mean ± standard error of the mean (SEM). The significance of differences was assessed with the GraphPad Prism software (Version 9.5.1, GraphPad Software, San Diego, California, United States). The normal distribution of all data sets was tested and, depending on the results, multiple comparisons were performed with the parametric one-way analysis of variance (ANOVA) followed by the Dunnet test, or with the non-parametric Kruskal–Wallis test followed by Dunn’s post-test. Differences were considered statistically significant if p < 0.05.

Grafts were generated with fibrin decorated with two doses of TG-Sema3a (0.1 and 100 μg/mL) or without the recombinant factor as control, together with 2 × 10⁶ freshly isolated human adipose tissue-derived stromal-vascular fraction cells (SVF) (Figure 1). The two doses were chosen to be far apart, in order to cover a wide biologically relevant spectrum, as previously found for the factor VEGF (Grosso et al., 2023). In order to provide the grafts with structural stability and persistence similar to a typical application like for bone tissue engineering, silicate-substituted calcium phosphate granules (Actifuse^®) were also included (Burger et al., 2022). Further, to ensure a slow degradation of the fibrin matrix over a period of at least 4 weeks, a TG-version of the fibrinolysis inhibitor aprotinin was included at the concentration of 51 μg/mL, as previously optimized (Sacchi et al., 2014). Constructs were then implanted in subcutaneous pockets in immune-deficient nude mice and harvested after 1, 6 or 12 weeks.

After 1 week the human SVF cells had efficiently self-assembled into microvascular networks throughout the constructs, as evidenced by specific immunostaining for human endothelium (hCD31) embedded within a laminin-positive basal lamina (Figure 2A). VLD quantification showed that the efficiency of this initial self-assembly step was not affected by the presence of Sema3A at any dose (Figure 2B; Control = 2.7 ± 0.5 mm/mm²; Sema3A 0.1 = 1.9 ± 0.7 mm/mm²; Sema3A 100 = 3.2 ± 0.6 mm/mm², p = n.s.). On the other hand, host-derived vessels could be identified by staining for mouse-specific CD31 and were found to have started invading the outer part of the grafts (Figure 2C). The density of host vascular networks was also not affected by Sema3A (Figure 2D; Control = 1.9 ± 0.4 mm/mm²; Sema3A 0.1 = 3.3 ± 1.0 mm/mm²; Sema3A 100 = 2.6 ± 0.9 mm/mm², p = n.s.).

After 1 week, the outer part of the grafts was populated by blood vessels of both human and mouse origin. By co-staining with species-specific anti-CD31 antibodies, it was found that in this outer portion human and mouse microvascular networks were connected in all conditions, as shown by abundant hybrid vascular structures containing both human and mouse endothelium within a common basal lamina (Figure 3A). Furthermore, these chimeric vessels were functionally connected to the host circulation, as evidenced by the presence of red blood cells within the lumens (Figure 3B). Quantification of the amount of human-derived blood vessels that were perfused by red blood cells showed that the self-assembled networks were about 90% functional in all conditions (Figure 3C; Control = 88.8 ± 2.6%; Sema3A 0.1 = 86.2 ± 4.4%; Sema3A 100 = 90.2 ± 1.5%, p = n.s.). Conversely, we investigated whether the ingrowth of host blood vessels could be influenced by Sema3A. Histological quantification of the percentage of graft areas invaded by mouse vascular networks showed similar results in all conditions (Figure 3D; Control = 2.4 ± 0.6%; Sema3A 0.1 = 4.6 ± 1.6%; Sema3A 100 = 1.8 ± 1.0%, p = n.s.). Together with the data reported in Figure 2D, these results show that Sema3A did not affect the initial host vascular growth inside the grafts, neither in its invasion speed nor in its density.

After 6 weeks in vivo, graft invasion by the host vessels was completed and the microvascular networks were comprised of both human and mouse components, which were still connected and giving rise to clear hybrid vessels similar to those at 1 week (Figure 4A). Vessel density quantification showed that in the control condition the human vascularization was diminished compared to the 1-week time-point (VLD 1 week = 2.7 ± 0.5 mm/mm² vs. 6 weeks = 1.0 ± 0.5 mm/mm², p = 0.05). In contrast, the presence of TG-Sema3A caused a significant expansion of the human SVF-derived vasculature at a low dose of 0.1 μg/mL (1 week = 1.9 ± 0.7 mm/mm² vs. 6 weeks = 5.3 ± 0.9 mm/mm², p < 0.05), but not at the higher dose (TG-Sema3A 100 μg/mL = 2.1 ± 0.7 mm/mm², p < 0.05 vs. TG-Sema3A 0.1 μg/mL and p = n.s vs. Control). This effect appeared to be specific for the self-assembled human vascular networks. In fact, the invading mouse vascularization was not affected by the presence of TG-Sema3A (Figure 4C; Control = 3.2 ± 0.6 mm/mm²; Sema3A 0.1 = 2.8 ± 0.4 mm/mm²; Sema3A 100 = 4.6 ± 0.3 mm/mm², p = n.s.).

By 12 weeks in vivo the self-assembled human vascular networks had completely regressed in the control condition but were still clearly visible in the presence of TG-Sema3A, especially at the low dose of 0.1 μg/mL, whereas mouse-derived vessels were present in all conditions (Figure 5A). Correspondingly, hybrid human-mouse vascular structures were not detected in the absence of TG-Sema3A. Quantification of vessel densities confirmed the specific protective effect of 0.1 μg/mL of TG-Sema3A on the human vascular component (Figure 5B; Control = 0.03 ± 0.03 mm/mm²; Sema3A 0.1 = 2.9 ± 0.4 mm/mm²; Sema3A 100 = 0.4 ± 0.2 mm/mm², p < 0.0001), but not on the mouse vessels (Figure 5C; Control = 2.6 ± 1.0 mm/mm²; Sema3A 0.1 = 5.9 ± 2.7 mm/mm²; Sema3A 100 = 3.8 ± 0.6 mm/mm², p = n.s.).

We previously found that endothelial-derived Sema3A promotes the stabilization and survival of VEGF-induced angiogenesis in skeletal muscle through the early recruitment of a specific population of NRP-1-expressing monocytes (NEM) (Groppa et al., 2015). Therefore, we investigated NEM recruitment in the grafts 1 week after in vivo implantation by co-staining for NRP-1 and the monocyte marker CD11b (Figure 6A). Endothelial cells were positive for NRP-1, but not for CD11b, while classic monocytes were CD11b⁺ and NRP1^-. Double-positive NEM could be identified in the outer layer of the grafts, in the areas that were invaded by mouse vessels and therefore accessible to the circulating cells. A quantification of their frequency revealed that NEM recruitment was significantly greater in the grafts with 0.1 μg/mL of TG-Sema3A compared to both controls and the higher Sema3A dose (Figure 6B; Control = 24.0 ± 9.7 NEM/mm²; Sema3A 0.1 = 158.1 ± 38.9 NEM/mm²; Sema3A 100 = 22.1 ± 7.7 NEM/mm², p < 0.01). These data show that the frequency of NEM recruited to the grafts correlated closely with the improvement of long-term persistence of the self-assembled human microvascular networks by a low dose of 0.1 μg/ml TG-Sema3A.