BM vascular 3D murine in vitro models include both hybrid (murine and human cells) and murine-derived cells (Tables 1, 2). Rat primary BM-MSCs in collagen gel were cocultured with primary rat aortic ECs. The ECs were either bound to dextran-coated Cytodex 3 microcarrier beads embedded in a 3D fibrin gel underneath to allow vascular sprouting (99), or cultured on the surface of collagen gel-BM-MSCs modules grown under shear stress (107). In this model, the empty spaces created by the collagen modules randomly packed were considered perfusable EC-lined channels (107).

Murine tri-culture has been realized by coculturing mouse ECs from the yolk sack (C166 cells) with BM-MSCs and leukemic B-ALL cells in a 3D fibrin hydrogel, in the two inner regions of a 3 concentric rings microfluidic chip. The niche cells, BM-MSCs, and ECs were placed in the middle ring area, while the B-ALL cells were in the central region (123).

A murine tetra-culture was generated in a microfluidic chip with two channels divided by a semi-porous membrane to physically separate the vascular and the BM niche components. The ECs C166 were grown as a monolayer on the lower side of the membrane under fluidic conditions (30 µl/h). On the apical side, the BM niche composed of primary BM HSPCs or whole BM cells and primary BM-MSCs in a fibrin-collagen gel was topped with a layer of BM-MSCs-derived osteoblasts (108).

An engineered murine BM vascular in vitro model was generated directly in vivo, implanting subcutaneously a polydimethylsiloxane (PDMS) device with a central cylindrical cavity functioning as a scaffold, resulting in a bone-like cylindrical disk tissue with a central region of blood-filled marrow. This engineered BM was surgically removed 8 weeks after implantation, punctured in multiple places to allow access to the culture medium, and then cultured in a 5 layers PDMS microfluidic device. Here, a compatible central cylindrical chamber to accommodate the engineered BM was encapsulated in a sandwich-like structure with two porous membranes and microfluidic channels (109).

One of the most important vascular features consists in the hemodynamic forces applied to the vascular walls formed by ECs, therefore advanced vascular models require shear stress to mimic in vivo blood flow. Two hybrid human-murine 3D microfluidic BM in vitro models that include shear stress on the endothelium were generated. The first is a bioreactor-on-a-chip composed of an upper and a lower microfluidic channel separated by a series of pillars (2 µm apart). The upper channel was injected with a 3D Matrigel or alginate gel embedding murine primary MKs, and a monolayer of primary HUVECs was cultured on fibronectin in the lower channel to mimic the BM vessel wall. In this model, the endothelial monolayer was directly exposed to a controlled flow (12.5 µl/h) by syringe pumps (110). In the second one, a 3-channel microfluidic model developed by Jeon and colleagues a flow of 2 μl/min was introduced via the lateral channel to condition the central channel vasculature network flow resulting in wall shear stress of 0.25 dynes/cm² (124).

To increase the complexity of the 3D BM vascular models, a great effort was put into isolating and/or employing human-derived cells both in static and microfluidic conditions (Tables 1, 2). These models mainly combined non-BM-derived primary ECs with BM stromal cells by culturing HUVECs or a few different primary ECs such as ECs from CB or derma as an endothelial component. Therefore, the human BM niche-stromal components were the real BM tissue-specific cells in these models.

A tri-culture of HUVECs mixed with primary CD138⁻ BM-MSCs and patient-derived cancer cells was performed in a fibrinogen scaffold generated from patient BM-derived fibrinogen, naturally found in the plasma of BM supernatant (125).

HUVECs-MSCs coculture has been also used to generate a 3D static and dynamic tetra-culture with osteoblasts differentiated from primary umbilical cord-derived MSCs, and primary CB-derived CD34⁺ HSPCs. Oliveira et al. generated a BM in liquefied capsules, to facilitate cell movement and self-organization, following three main sequential steps: (i) engineering the endosteal niche by seeding human MSCs with surface-functionalized microparticles, (ii) bioencapsulation of the endosteal niche with HUVECs and HSPCs in alginate, and (iii) production of a multilayered membrane layer-by-layer followed by core liquefaction (111). Another 3D BM static model employed spheres was, generated by the magnetic levitation method, to avoid the use of scaffolds or exogenous matrices and to allow the cells to form the extracellular matrix (108).

Primary ECs from derma, primary BM-MSCs, and primary CD34⁺ HSPCs from CB were combined in a tri-culture, assembled in the 3D Bio AssemblerTM System that mixed all cells (112).

A static penta-culture 3D BM model was realized by embedding in Matrigel or alginate hydrogel primary cells: CB-derived HSPCs and EPCs, BM-MSCs, osteoblasts, and BM-MSCs-derived adipocytes (113).

The outstanding research of Palikuqi et al. identified a new method to obtain a durable hemodynamic, self-organizing, large-volume 3D vascular system in a Matrigel-free matrix, composed of a mixture of laminin, entactin, and type-IV collagen (LEC matrix). It is based on the “reset” of vascular endothelial cells (R-VECs) through the transient reactivation of the embryonic-restricted ETS variant transcription factor 2 (ETV2). These findings could lead to a revolution in the organoid field thanks to the ability to vascularize organoids and tumoroids (136). In the context of in vitro BM modeling, it would open the possibility of recreating the vascular niche in a complex 3D environment for pharmaceutical studies but also offer the opportunity to develop functional and perfused implantable tissues ex vivo.

A vascularized BM organoid based on human induced pluripotent stem cells (iPSCs) was developed in a hydrogel composed of collagen I and IV, and Matrigel. The used medium was supplemented with vascular endothelial growth factor C and the characterization reveals the presence of iPSC-derived ECs, MSC, HSPCs, myelomonocytic, megakaryocytes and immature erythrocytes with human BM cells characteristics. Moreover, this human BM organoid were engrafted with different types of cells to mimic human diseases; specifically: iPSC-derived HSPCs derived from patients with myelofibrosis or healthy individuals, primary cancer cells from patients affected by multiple myeloma, acute lymphoblastic leukemia (ALL), and myeloid leukemias (126).

The aforementioned ring shape microfluidic chip (123), used to culture murine cells, was also employed in an all-human fibrin hydrogel-based 3D model. In this case, HUVECs were seeded in the central sinus region, a mixture of HUVECs, primary BM-MSCs, and leukemia cancer cells in the inner ring, and a mixture of osteoblasts (hFOB cells) and leukemic cancer cells leukemia in the outer ring channel (123) (Figure 2D).

A 3D hydroxyapatite porous scaffold placed in, and, connected to a perfusion bioreactor was used to coculture adipose tissue-derived stromal vascular fraction, which includes primary mature ECs, endothelial progenitor cells, pericytes, and MSCs, with osteo-differentiated primary BM-MSCs. This engineered vascularized BM niche was perfused at the rate of 300 μl/min with a suspension of CD34⁺ CB-HSPCs (114).

To mimic the in vivo vascular structure, 3D human BM in vitro models were developed based on self-assembled vascular networks (Tables 1, 2). HUVECs have been cocultured in tri-cultures with primary BM-MSCs and cancer cells in a hydrogel drop of approximately 200–300 Pa stiffness, optimal for the development of a robust endothelial network (127).

Microfluidic human BM 3D in vitro models are characterized by controlled and/or induced flow for nutrient exchange in all-human cell cultures. 3D vascular models made of capillary-like structure networks were formed by HUVECs cocultured with primary BM-MSCs. The mesenchymal cells adopted pericyte-like localizations among the network system, both in a 3D decellularized bone matrix (128) and in fibrin matrixes within 3D microfluidic chips with different designs and geometries (115–117).

Maturano-Kruik et al. developed a capillary-like structures network model seeding the ECs and MSCs within a 3D decellularized bone tissue's trabecular scaffold that after a maintenance phase, was placed in the microfluidic chip and exposed to a controlled flow (128).

The microfluidic chips of Jeon et al. (115) and Mykuliak et al. (117) were designed with a central gel channel, for HUVEC-MSC coculture, between two lateral channels for culture medium, separated by pillars. In this last HUVEC-MSC coculture model, the shear stress on ECs was exerted by interstitial flow and it was estimated to be 0.0065 Pa (117).

Carrion et al. used a 3D penta-channel microfluidic chip with three central parallel gel channels and two lateral channels for medium supply. The central gel chamber was filled with fibrin gel only, and the two adjacent channels with fibrin-embedded ECs and MSCs respectively (116).

Capillary-like structures were also obtained by culturing HUVECs in a 3D gel in the absence of perivascular MSCs (103, 129). In one case, cells were cultured in a microfluidic chip with three parallel microchannels, partitioned by trapezoid-shaped pillars. The central channel was filled with collagen gel, and one lateral channel with ECs monolayer, grown on the pillar-collagen interface of the central channel to provide the space for angiogenic sprouting and neo-vessel formation. The other lateral channel was seeded with leukemia cancer cells followed by injection in the channel reservoirs of BM-MSCs (HS5 cells) in collagen gel (129).

In a second work, the microfluidic chip was composed of two lateral media channels and three main parallel gel channels (103). The two lateral channels mimic the BM vascular niche and the endosteal niche, hosting the vascular network generated by HUVECs and hFOB osteoblasts in collagen fibrin gel, respectively. The central channel was loaded with primary HSPCs in collagen fibrin gel. Another lateral channel, separated by the medium channel from the vascular compartment, holds mitomycin C-treated fibroblasts (NHLF cells) and was used as a cytokine-secreting compartment (103).

The BM perivascular and endosteal niches were also simultaneously reproduced in a microfluidic device with two hexagonal chambers connected by three symmetric two-way ports that allow nutrient diffusion and cell migration. Primary CB-derived ECs were cultured in both chambers in fibrin hydrogel to form a vascular network, mixed either with hFOB 1.19 osteoblasts to form the endosteal niche or primary BM-MSCs to form the perivascular niche on-a-chip (130) (Figure 2B).

The biological complexity of a 3D BM in vitro model can be enhanced by adding other cell types in culture, to better mimic the tissue microenvironment. Other models characterized by the formation of a BM vascular network and its niche were realized in 3D microfluidic devices with parallel 3- or 5-channels. In the central channel, HUVECs were cultured with primary BM-MSCs, osteo-differentiated primary BM-MSCs, and primary HSPCs (118) or with primary BM-MSCs and macrophages (115) in ECM gel. Hence, this type of model allows the mixing of the vascular and the endosteal niches' cellular components. In the central channel of the 3-channel microfluidics device, a microvascular network with highly branched structures was grown in 3D fibrin, supported by medium from the two lateral channels (115). In the central channel of the 5-channel microfluidic system BM-M-SCs were cultured on a mix of collagen and dopamine-HCl to differentiate in osteoblasts as a monolayer, then a fibrin-collagen hydrogel with a mixture of endothelial, mesenchymal and hematopoietic cells was injected on top of the osteoblasts to obtain a 3D human BM-on-a-chip. The two side channels serve for medium supply, and the two external channels were filled with BM-MSCs in fibrin-collagen hydrogel to provide further support to the BM niche (118).

To mimic the vascular geometry and endothelial distribution in vivo, 3D vascular microfluidic models were designed to grow ECs on the channel walls to generate an endothelial monolayer creating a lumen.

A microfluidic device with three parallel central channels and two lateral channels for medium supply was designed to create a lumen (137). In this device, a HUVEC monolayer was cultured on Matrigel in the central channel to form a vascular wall-like geometry lumen with a squared section where medium and circulating cells were flown trought (131). In the side channel, in contact with the endothelial central channel, osteo-differentiated primary BM-MSCs were cultured in 3D collagen to generate an osteo-conditioned microenvironment (131) (Figure 2F).

A 3D 1-channel BM microfluidic model, part of a body-on-a-chip, consists of primary HSPCs embedded in a 3D fibrin gel and a HUVECs monolayer placed on opposite site of a porous membrane. Here the endothelial monolayer was directly exposed to controlled shear stress (0.2 dynes/cm²) (132).

A similar microfluidic model included also the primary BM-MSCs with primary HSPCs embedded in fibrin in the top compartment, and a HUVEC monolayer grown under shear stress (1.2–1.6 μl/min) as a vascular compartment in the bottom (119) (Figure 2H).

The BM endosteal niche was partially recapitulated in a 3D microfluidic chip, as part of a vascularized multi-organ tissue system (133). Here, the bone tissue was obtained by coculturing both BM-MSCs differentiated osteoblasts and primary CD14⁺ monocytes-derived osteoclasts within a decellularized bone scaffold on top. To mimic blood vessel walls, the porous membrane was seeded with pericytes-mimicking BM-MSCs on the upper surface and with HUVECs mixed to a small fraction of BM-MSCs on the lower side grown under shear stress (1.88 dynes/cm² max) (133).

A 3D microfluidic vascular network that mimics the vasculature geometry was created by growing HUVECs on a collagen gel grid characterized by a series of squared blocks design, to allow the formation of a rounded-shaped monolayer under gravity-mediated flow. In the collagen gel, the BM components were embedded as primary BM-MSC or BM fibroblasts cell lines (134) (Figure 2E).

A tubular-shaped vascular microfluidic device made with a silk microtube surrounded by a porous silk sponge was seeded with primary CB- or derma-derived ECs in the silk microtube lumen and exposed to shear stress. Then, primary HSPCs-derived human MKs were seeded into the silk sponge (105) (Figure 2G).

Finally, the 3D BM vascular in vitro model, which better mimics the in vivo BM vasculature was generated using organ-specific primary BMECs, isolated via magnetic separation from frozen human BM-MNCs. Arterial and sinusoidal ECs were isolated via double selection for CD146⁺/NG2⁺ or CD146⁺/NG2^lo/−, respectively (135). These cells were grown to replicate the major 3D BM niches in a deconstructed way in a microfluidic chip designed with 4 non-communicating parallel channels that can be exposed to control flow, where the four different niche cell populations were encapsulated in ECM-derived hydrogel: (I) arteriolar BMECs, (II) sinusoidal BMECs, (III) primary BM-MSCs and (IV) osteoblasts derived from osteo-differentiated primary BM-MSCs (135).

In the next sections, we describe how the reported models have been employed for either cardiovascular or cancer research.

Leukaemia and myeloma begin during hematopoiesis with cell genetic alterations. Most leukaemia involves white blood cells, and some affect other types of blood cells. When these cells over over-proliferate, they crowd out normal BM cells with consequent hematopoiesis alterations leading to the development of the disease. Leukemic cells leave the bone marrow niche through the BM vasculature to enter the bloodstream, where they can spread to other organs. Leukaemia is classified based on the speed of progression, that which could be acute or chronic, and on the type of blood cell involved, myeloid or lymphoid (146).

Acute leukaemias (AL) are characterized by immature blood cells with high intrinsic proliferative potential that leads to a fast progression of the disease. Altered immature blood cells with a myeloid commitment, cells able to differentiate into MKs, red or non-lymphocyte white blood cells during hematopoiesis in health, cause acute myeloid leukemia (AML), which is the most common type of acute leukemia in adults (147).

In leukemic patients it has been observed that an increased BM vascularization is a key feature, however, this event has not been largely studied for solid tumor progression and metastasis (148). Angiogenesis was studied with a 3D BM microfluidic chip showing that ECs (HUVECs) sprouting and neo-vessel formation were increased by the presence of AML cells (HL60 cells) compared to control (EC medium). In the presence of HS5BM-SCs, to mimic the BM niche, angiogenesis was more pronounced, and, in coculture with AML cells, it was further enhanced, but characterized by a larger number of isolated endothelial tips without multicellular stalks, reflecting what was found in AML patient biopsies. ECs migrated for longer distances in the collagen matrix, with a higher number of endothelial tips but lower vascularization were found using chronic myeloid leukemia (CML) cells (K562 cells) in the absence of stromal cells (129). CML originates from partially mature myeloid white blood cells and is characterized by slower progression compared to AML. In turn, ECs play an important role in CML. Coculture of HUVECs with a murine AML cell line in a 2D model showed that ECs support the expansion of rare leukemic cells, known as LICs representing a subset of self-renewing cells able to generate an aggressive AML (120). This was also shown in the 3D BM vascular model developed by Bray et al., where both AML cell lines (KG1a, MOLM13, MV4-11, OCI-AML3) and primary patient-derived AML cells exerted preference to adhere and proliferate along the endothelial network (HUVECs-BM-SCs), further highlighting the importance of the interaction between AML cells and the vascular niche. In accordance with these findings, this model enabled further research on drug resistance, one of the main challenges in clinical treatment. It has been observed that the 3D tri-culture model was more resistant to two antineoplastic drugs commonly used to treat AML compared to simpler 2D and 3D mono-cultures (AML cells only) (127). In a perfusable 3D microfluidic vascular model with tubular HUVECs vessels cocultured with stromal BM fibroblast cell lines (HS27a or HS5), patient-derived primary AML cells adhesion and extravasation were tested under shear stress in combination with primary monocytes. Interestingly, AML cells did not exhibit preferential adhesion or extravasation in the presence of either stromal cell type, in contrast, leukemic cells without monocytes adhered and extravasated more to and across vessels cocultured with HS27a when compared to HS5 cells. These findings underline the importance of complex 3D models that mimic in vivo multicellular components (134) (Figure 2E).

To explore the AML cells-BM niche components specific interaction, a deconstructed microfluidic BM niches on-a-chip (NOC) with arterial (BMAECs), mesenchymal (BM-SCs), sinusoidal (BMSECs) and osteoblastic niches was perfused with AML cells (MOLM13). AML cells were preferentially lodged within the osteoblastic and the arterial niches rather than into the mesenchymal or sinusoidal niches (135). Acute lymphoblastic leukaemia (ALL) is another form of acute leukaemia and the most common cancer in childhood, characterized by the overproduction of immature and dysfunctional lymphoblast in the BM (149). A ring-shaped leukaemia-on-a-chip 3D BM model (Figure 2D) that resembles the in vivo spatial architecture and cellular composition of the leukaemia BM tissue was developed and adopted to study the heterogeneity in B-ALL human BM microenvironments based on their chemo resistance. It has been found that Ph⁺ SUP B-ALL cells cocultured with BM niche cells in the biomimetic device were more resistant to conventional chemotherapic than REH B-ALL, consistent with the insensitivity of Ph⁺ B-ALL to these types of drugs in the clinic. Furthermore, the leukemia-on-a-chip 3D BM model was designed to rescue the niche cells for downstream assay like single-cell RNAseq, a very powerful tool to characterize the BM microenvironment with limited cell input number. Finally, using a simpler ring-shaped murine 3D BM model with ECs and BM-MSCs it was observed that B-ALL cells attracted ECs. Niche cells promoted B-ALL cell cluster formation and reduced their motility over time (123).

The second most common hematologic malignancy is the MM, where clonal terminally differentiated B lymphocytes undergo abnormal proliferation followed by accumulation in the BM crowding out healthy blood cells. In MM abnormal monoclonal antibodies, paraproteins, are produced causing overly thick blood and kidney issues (150). To investigate the pathophysiology of MM, a 3D tissue-engineered BM (3DTEBM) derived from MM patients-BM supernatant that incorporated the BM components MM (MM1s, H929, RPMI8226 or primary), stromal, and ECs was established. It was observed that the presence of both stromal and ECs increased MM cell proliferation compared to single culture or coculture using other commercially available 3D systems, thanks to the 3DTEBM scaffold with fibrinogen cross-linking (125).

The classical feature of BM myelofibrosis consist in the deposition of reticulin and collagen fibers by marrow stroma (151) and the iPSC-based organoid model was able to recapitulate them when the donor cells derived from patients with myelofibrosis but not when derived from healty donor: increased soluble TGFβ levels, collagen 1 and α-SMA and reduced vascularization was observed (126).

Solid tumors such as breast cancer can spread from the primary site intravasating in the bloodstream to form secondary tumors at distant sites called metastasis. Metastasis is a multistep process where circulating tumors cells (CTCs) interact with ECs, forming the microvascular vessel wall, adhering and extravasate extravasating into the target organ to form metastasis (Figure 1). These key events of metastasis formation take place under hemodynamic shear stress exerted on ECs by blood flow (152). Triple-negative breast cancer (negative for oestrogen receptor, progesterone receptor, and HER2) is a highly invasive type of breast cancer, resistant to both hormonal therapy and treatments against HER2 (Trastuzumab), that develops metastasis in the bone with high frequency (153). Numerous BM vascular models in vitro were developed to study metastasis formation by human triple triple-negative breast cancer cells (MDA-MB-231) (121, 124, 128, 130, 131). It has been shown that CTCs firm adhesion to ECs (HUVECs) under shear stress was higher when HUVECs were cocultured with BM niche specialized cells compared to HUVECs cocultured with muscle cells, resulting in cancer cells organotropism for bone vasculature (121) (Figure 2C). In a model of self-assembled capillary-like network (HUVECs), cancer cell extravasation in the matrix increased in the presence of BM cells (BM-MSCs and osteoblasts). In contrast, it was observed a decreased extravasation in a shear stress-preconditioned vascular network or in the presence of inflammatory macrophages (inflammation model) within the BM niche (124). The extravasation step was further investigated in a microfluidic model with HUVECs forming a vascular lumen. MDA-MB-231 cells transmigrated across the HUVECs monolayer into the osteo-differentiated BM-MSC gel and travelled within the matrix more compared to the control. Extravasated cancer cells were able to proliferate and form micrometastasis of various sizes, ranging from 4 cells to 60 cells or more (131). A 3D BM vascular model was employed to investigate cancer cell migration to BM-specific niches. An increased MDA-MB-231 cell migration to the BM niches compared to the control was observed, without differences between perivascular and endosteal niches. The migrated MDA-MB-231 cells expressed more Ki-67, a proliferation marker, than non-migrated cells, consistent with high Ki-67 expression in bone lesions (130) (Figure 2B). All these 3D BM vascular models showed how the BM niche is important for the breast cancer cell metastatic process and enabled to highlight the cancer cell organotropicity. Finally, a 3D model with a bone matrix scaffold, to maintain the ECs viability and branching without the need of for specialized growth-promoting conditions culture medium, was developed to study metastatic colonization of MDA-MB-231 cells under static and interstitial flow conditions. It was reported that the interstitial flow promoted vascular branching and prevented cancer cell growth rate within the niche (128). The BM vasculature-on-a-chip models were also employed for drug toxicity studies. Two models were realized as part of a body on a chip (132, 133) to test the uptake and side effects of common chemotherapeutic drugs given to treat several cancers such as cisplatin and doxorubicin. The perfusable vasculatures were challenged in the luminal side when in contact with the apical bone marrow niche. Cisplatin was flowed through the vascular channel causing myeloid toxicity in the HSPCs BM niche, reducing the total number of cells in the HSPCs BM niche, in particular neutrophils and erythrocytes, recapitulating the known side effects of cisplatin in patients (neutropenia and anaemia) in vitro (132). In a vascularized bone model which includes a lot of components in common with the BM endosteal niche, osteoblasts were found to be more sensitive than osteoclasts when exposed to doxorubicin, as observed in pre-clinical studies, and, as expected, the endothelium showed decreased resistance (133).