Three-dimensional human bone marrow organoids for the study and application of normal and abnormal hemato-immunopoiesis

Abstract

Hemato-immunopoiesis takes place in the adult human bone marrow (BM), which is composed of heterogenous niches with complex architecture that enables tight regulation of homeostatic and stress responses. There is a paucity of representative culture systems that recapitulate the heterogeneous three-dimensional (3D) human BM microenvironment, and that can endogenously produce soluble factors and extracellular matrix that deliver culture fidelity for the study of both normal and abnormal hematopoiesis. Native BM lymphoid populations are also poorly represented in current in vitro and in vivo models, creating challenges for the study and treatment of BM immunopathology. BM organoid models leverage normal 3D organ structure to recreate functional niche microenvironments. Our focus herein is to review the current state-of-the-art on the use of 3D BM organoids, focusing on their capacities to recreate critical quality attributes of the in vivo BM microenvironment for the study of human normal and abnormal hematopoiesis.

Introduction

Hematopoiesis, the process of producing blood and immune cells from hematopoietic stem and progenitors cells (HSPCs), takes place in the adult human BM through a coordinated hierarchy responding to both positive and negative signals towards lineage commitment and maturation(1–3). HSPCs dynamically interact with the BM microenvironment to self-renew, differentiate or proliferate in order to maintain tissue homeostasis and respond to stress, changing as humans age(4). Interaction occurs in specialized and personalized heterogeneous cell niches comprised of hematopoietic and non-hematopoietic cells(5), such as mesenchymal and stromal cells (MSC)(6, 7), nerves, osteoblasts, adipocytes, endothelial cells, extracellular matrix (ECM) molecules that orchestrate cell fate by communicating through cell-to-cell contact and humoral factors including cytokines, chemokines, and other physico-chemical cues and gradients (e.g. oxygen)(8, 9). These factors determine HSPC self-renewal, commitment, expansion and migration to sustain BM function throughout the lifetime of the host(4, 10, 11). There is now general consensus that two HSPC BM niches exist, which function in lineage commitment, proliferation, quiescence and mobilization: an endosteal niche adjacent to the trabeculum and a perivascular niche adjacent to the sinusoid (Figure 1)(12–14). The purpose of this brief review is not to recapitulate the most recent knowledge of BM structure and function in the normal and diseased state, and for which we refer the reader to many excellent reviews(1–6, 9, 15–17), including on resident lymphoid populations(18). Rather, our focus is to undertake a wide view of the current state-of-the art in recapitulating the three-dimensional (3D) structure and function of the BM organ (both normal and abnormal; Figure 2), the critical quality attributes for a successful BM organoid and how these models may be applied for the study of normal hematopoiesis and in the field of immunology and immunopathophysiology of BM disease, such as bone marrow failure syndromes and acute myeloid leukemia (AML)(19, 20).

Figure 1. Human BM structure and organization.

A) Adult human BM is located within cancellous tissue of mainly the long bones (shown here), pelvis, sternum, ribs, vertebrae and skull, and has a rich arterial supply. B) Paratrabecular regions are lined with osteoblasts and other supportive stromal cells, and constitute the endosteal compartment supporting one of the HSPC niches. The perivascular niche abuts the sinusoid towards which multilineal differentiated cells progress as they mature for release into the peripheral blood circulation. Extracellular matrix permits cell, cytokine and other soluble factor retention, and niche attachment as well as maintenance of the complex architecture and heterogenous cellular maturation during homeostasis and stress states. Erythroblastic islands are key components of the BM and highlight the dependence of normal hematopoiesis on the function of innate immune cells, finely tunable to the cytokine and stress response.

Figure 2. 3D BM models for the culture of human HSPCs or leukemia cells.

Stromal feeder cells are often added to create a supportive microenvironment. A) 3D Static models and B) Perfused models utilize a porous scaffold or hydrogel. C) Murine/Scaffold hybrid models use a murine host to implant a scaffold seeded with human cells, generating a humanized microenvironment with recruitment of murine cellular and humoral factors, including vascularization. D) Microfluidic devices recreate select BM environments on-a-chip. Each discrete and interconnected chamber recreates a particular BM compartment wherein few cells form niches to mimic BM interactions, e.g. adhesion, while receiving shear stress.

The Challenge

A significant limitation to understanding normal and abnormal hemato-immunopoiesis in the context of heterogenous BM niches is a paucity of representative and robust culture systems that recapitulate the 3D human BM organ structure, function and microenvironment with inherent diverse stromal and endothelial niches, native lymphoid populations, and metabolism to reflect primary human myelo-biology in homeostatic and stressed states. Current in vitro two-dimensional (2D; flask and plate) cultures lack the heterogeneity and complexity of the human BM microenvironment and have significant limitations: they require addition of serum with inherent inter-batch variability, supra-physiological concentrations of cytokines and/or allogenic feeder layers resulting in the introduction of artefact and an oversimplification of in vivo BM function and hematopoiesis (Table 1)(21–23). Murine in vivo models are most widely used for the study and pre-clinical experimentation in hematopoiesis and AML, though remain unsatisfactory as model systems due to discrepancies between human and murine physiology(24, 25). Even AML Patient-Derived Xenograft (PDX) mouse models are inadequate, particularly for the study of microenvironment and stromal-AML interactions, due to (1) stromal damage caused by irradiation required for human grafts to take, (2) 50% failure rate of engraftment and (3) a 12-week minimum for engraftment to occur(26–28). When successfully engrafted, PDX models are beneficial for studying clearance of AML in the blood and spleen in response to therapy, but they are limited when examining the BM microenvironment wherein leukemic cells are protected(29, 30). The use of any ex vivo model for the study of BM failure syndromes will require the support of different native lymphoid populations (deficient in the NSG and similar mouse backgrounds) and non-hematopoietic components (e.g., sympathetic innervation) that are implicated in the pathophysiology of the disease, such as in aplastic anemia or myelodysplastic syndromes, or its treatment. Though addition of cytokines, e.g. interferon (IFN)-γ or tumor necrosis factor (TNF)-α, implicated in these conditions may be added to established organoids for mechanistic studies of pathophysiology, cytokines cannot entirely substitute for the actions, responses and drug-targeting of native lymphoid populations. Such ex vivo models will need to support tissue-like densities and sustain long-term dynamic cultures which are tunable and easily interrogated.

Table 1. Principal Advantages and Disadvantages of Models to Study Human Normal and Abnormal Hematopoiesis.

Model	Advantages	Disadvantages
2D Models	Ease of Use	Homogeneous Feeder Layer
	Reproducible	Serum & Cytokine Dependence Gradient-poor
	Quick Results	Spatial Organization Cell-Cell/Microenvironment Interactions
	Cheap	Nutrient Access Short-term Cultures
3D Models	Moderate Heterogeneity Spatial Organization	Feeder Layers
	Cell-Cell Interactions	Serum & Cytokine Dependence
	Soluble Factor Gradients	Lower Reproducibility
	Niche/Microenvironment Formation	Lack of Lymphoid Populations Cheap (Static) / Expensive (Perfusion)
	Long-term Cultures Tissue-like Densities
Animal Models	Complex	Animal Physiology
	Heterogeneous Niche Formation	Stromal Damage
	Long-term Studies	Low Engraftment Rate
	Serum- & Cytokine-Free Whole Animal Studies	Lack of Lymphoid Populations Expensive

BM organoids, which were originally introduced over 30 years ago in the study and application of normal and diseased processes(31, 32), are gaining traction. In the most simple form, these 3D models are based on the premise of applying a bio-mimetic material similar to that of human BM in terms of architecture, porosity, nutrition/oxygen diffusion, metabolite gradients and surface biochemistry(31, 33, 34). As in 2D systems, the addition of prescribed exogenous feeder layers (whether derived from autologous, allogenic, xenogeneic, or from cell line sources) in order to recreate a 3D BM microenvironment introduces artefact due to cell source and culture expansion conditions, with skewing of subpopulations during culture. The use of high concentrations of cytokines and serum to sustain proliferation is not physiological, differs from patient-to-patient in homeostasis and disease states and enables clonal selection even in 3D culture(35–37). Current BM organoids are mostly indifferent to maintenance of lymphoid populations, focusing mainly on the myeloid, stromal and leukemic elements under study. In order to maximize representativity, robustness, reproducibility and fidelity, 3D BM models should incorporate certain critical quality attributes that are universal to the BM organ and define its function: (1) heterogeneous cell niches that include hematopoietic, immune and non-hematopoietic cells, diverse ECM molecules (collagen, fibronectin, laminin), and different humoral factors (cytokines and chemokines) forming niche-specific gradients, (2) be structurally complex with porosity and pore size similar to that of human BM in order to support heterogenous niches, (3) support long-term dynamic cultures to reflect the hematopoietic changes that normally occur over time, (4) enable metabolite turn-over, oxygen gradients and shear stress inherent in normal human BM, (5) be self-sustainable with autocrine and paracrine production of physiological cytokines reducing the requirements for additives (except for critical factors not produced normally in BM, e.g. hormones, erythropoietin), (6) enable spontaneous niche formation and re-modelling that is required for BM organoid personalization and, (7) be robust platforms with reproducible results (Table 2).

Table 2. Required Critical Quality Attributes for a 3D Human BM Biomimicry.

Model	Critical Quality Attribute
	Niche Heterogeneity1		Structural Heterogeneity2	Serum & Cytokines	Long-Term Dynamics	Personalized3
	Cell	ECM
2D Culture	Low	Low	Low	Exogenous	No	Low
3D Static	Medium / High	Medium / High	Medium / High	Exogenous/Endogenous	Some	Medium / High
3D Perfusion	Medium / High	Medium / High	Medium / High	Exogenous/Endogenous	Yes	Medium / High
Microfluidics	Medium / High	Medium / High	Low / Medium	Exogenous	No	Low
Animal Models	Medium / High	High	High	Endogenous	Yes	Medium
Hybrid Models	Medium / High	High	Medium / High	Endogenous	Yes	Medium / High

¹Low (1 component); Medium (2 components); High (3 or more components).

²Low (homogeneous structure); Medium (homogeneous structure with heterogeneous sections); High (fully heterogeneous structure).

³Low (organoid self-organized structure within defined cell proximities only); Medium (organoidselforganized structure in specific sections); High (organoid fully self-organized and structured).

3D organoid models of human BM as the new golden standard

Two main types of 3D BM organoids can be distinguished: (1) Porous scaffolds and (2) Hydrogels. Scaffolds vary in bio-material and can be of natural (e.g. collagen, gelatin, or Matrigel) or synthetic (e.g. Polystyrene (PS), Polyethylene glycol (PEG), Poly-ε-caprolactone (PCL), Poly(lactide-co-glycolide) (PLGA), Polyurethane (PU)) origin(38). Hydrogels are polymer networks containing a large volume-percentage of water and can be particularly effective as cell support matrices due to their similarity to the ECM(39, 40); their mesh size is often not large enough to allow for cellular motility, resulting in cells either being encapsulated or settled on the surface of the gels, reducing cell-cell interactions(39–42). Gels can be modified for desired stiffness(43), water percentage, or encapsulation of growth factors(44) and biological cues(45) impacting cell morphology(46), HSPC differentiation(47, 48) and proliferation(49). Commercially available gels often used in hematopoietic culture systems, such as Matrigel, have an undefined composition containing exogenous factors that introduce artefact into the system undermining culture fidelity, reproducibility and robustness(50). In contrast, porous scaffolds provide an open structure allowing for cell motility, so much so, that often cell attachment becomes an issue unless they undergo biofunctionalization with ECM molecules for contact-dependent signaling and cell maintenance within the structures during culture perturbation. Pores can be manufactured using different methods including salt leaching, freeze-drying and 3D printing(41). Due to improved porosity compared with that of hydrogels, porous scaffolds exhibit higher oxygen, nutrient, metabolite, soluble factor diffusion, and higher cell motility with heterogenous cell niches(51) wherein cell-cell and cell-environment interactions can be captured in tissue-like densities(38, 52).

Static organoids: hydrogels

Hydrogel matrix stiffness plays a role in HSPC lineage commitment. Examples of this include a Pullulan/dextran hydrogel used to induce CD34⁺ HSPC differentiation towards megakaryopoeisis(53) using stem cell factor (SCF; 20 ng/mL) and thrombopoietin (TPO; 50 nM) for 30 days, and a 3D collagen matrix supplemented with SCF (100 ng/mL) that resulted in HSPC expansion and myeloid differentiation after 2 days(54). A similar collagen hydrogel was applied to HSPC/MSC co-culture with addition of SCF 50ng/mL, TPO 20ng/mL, Flt3-Ligand (Flt3L) 50ng/mL and Interleukin- (IL) 6 10ng for 14 days, and achieved two different microenvironments: maturing cells in the supernatant suspension and a matrix microenvironment within the collagen fibers(55). Within the hydrogel, MSCs formed endosteal-like niches that supported CD34⁺ cells, expansion and migration of early myeloid progenitors, whereas cells in suspension were maturing myeloid cells and CD56⁺ Natural Killer (NK) cells. Removal of the MSC feeder layer enabled differentiation of CD56⁺ NK cells at the expense of myelopoiesis suggesting that HSPCs had reduced capacity to differentiate into NK cells whilst myeloid commitment and maturation was occurring in the presence of MSCs and offered an intriguing model to study myeloid-lymphoid lineage switching with respect to niche interactions. In another example, a macroporous PEG hydrogel was used to co-culture CD34⁺ HSPCs for 4 days in serum- and cytokine-containing media either with an osteoblast-like cell line or with allogeneic MSCs(56). The latter supported higher proliferation and survival and was better able to preserve multi-lineal potential both within the scaffold and in the supernatant when compared with 2D controls, underscoring the platform’s capacity to preserve HSPC function. Seeking to create a system that could sustain multiple BM populations, Matrigel and alginate were used to culture MSCs, endothelial cells and HSPCs in a single platform exposed to different oxygen concentrations(57). Alginate hydrogels produced MSC-derived adipocytes and osteoblasts and an even more diverse culture that supported HSPC survival and increased colony number after 10 days. Matrigel enabled HSPC expansion, maintenance of immature progenitors and some CD38⁺ differentiation, with more cell-cell contact observed between HSPCs and supporting niche cells, stressing the importance of understanding how different materials confer different properties to the hematopoietic system being studied.

Hydrogel 3D-printing enables custom-designed matrices and has been used to culture chronic lymphocytic leukemia (CLL) cell lines using RGD/laminin based bioink in 10% FBS(58), preserving leukemic immunophenotype and maintaining higher viability over a 28 day culture compared with 2D controls. The 3D-printed BM organoids with specified attributes could deliver the reproducibility and robustness required in general hematopoiesis research, assuming that all elements of the microenvironment are recapitulated, including lymphoid populations, to study immune cell dynamics, the stress response and or cell-cytokine interactions. Although 3D printing is an appealing strategy to create “designer” BM organoid scaffolds with inclusion of customized cell populations of different types and directed towards different niches, it is unsuitable for personalized study of disease and precision therapeutic responses.

Static organoids: porous scaffolds

Porous scaffolds composed of PCL and fibrin could successfully expand HSPCs co-cultured with allogeneic BM-MSCs, supplemented with SCF 10ng/mL, TPO 20ng/mL, fibroblast growth factor-1 (FGF-1) 10ng/mL, and angiopoietin like-5 (Angptl-5) 500ng/mL(59). The output HSPCs could engraft NSG mice, better than those cultured without MSCs, and differentiated in vivo into CD13⁺, CD3⁺ and CD19⁺ progenitors. Separately, a PU scaffold coated with collagen I was developed to expand cord blood mononuclear cells (CBMNCs) in the only long-term (28 days) serum- and cytokine-free 3D organoid model(60). PU scaffolds were manufactured by thermally-induced phase separation resulting in a structure that resembled human BM with well interconnected macro (100 – 300 μm) and micro (50 – 100 μm) pores and a total porosity around 90%. The scaffolds could be bio-functionalized with collagen (62.5μg/ml) or RGD(61) to ensure cell adhesion and improved proliferation. Human CBMNCs were successfully cultivated in serum-free 3D cultures exposed to fluctuating oxygen schedule between hypoxia (5%) and normoxia (20%) with near-physiological concentrations of EPO (0.1U/mL), with or without SCF (10ng/mL), in the absence of other exogenous growth factors or feeder layers. After 4 weeks, the cytokine-free 3D biomimicry supported: (a) progenitor clonogenic capacity, (b) cell growth in “niche-like” structures, (c) committed multi-lineal differentiation in cytokine-free conditions and, (d) a supportive microenvironment with self-organizing heterogenous stromal niches (STRO-1, NES, CD68), osteoblasts (OPN, Osx) and hematopoietic cells (CD45) and endogenously-produced ECM proteins laminin and fibronectin, and soluble growth factors and adhesion molecules (e.g. V-CAM1, SDF-1, IL-6, GM-CSF and G-CSF) that enabled a self-sustaining culture and could maintain lymphocytes over two months of culture (unpublished data). The platform was robust, even accounting for the inter-sample variability inherent in primary samples, sensitive to hypoxia and small perturbations in the system lending itself as a good model to study stress hematopoiesis and microenvironmental dynamics.

Inverted colloidal crystals (ICC) provide more macroporous highly organized structures with increased porosity and connectivity compared with other scaffolds(62). ICC scaffolds coated with nanocomposite were used to co-culture CD34⁺ cells with the BM stromal cell line HS-5, human fetal osteoblasts and the cytokines IL-2 5ng/mL, IL-7 20ng/mL, Flt3L 20ng/mL, SDF-1 20ng/mL, BMP-4 4ng/mL and IL-3 10ng/mL for HSPC expansion, and CD40L 5ng/mL, IL-4 10ng/mL, IL-5 10ng/mL, IL-6 10ng/mL, IL-10 10ng/mL, IL-2 5ng/mL, IL-7 20ng/mL, Flt3L 20ng/mL, SDF 20ng/mL, and IL-3 10ng/mL for B-cell differentiation in 28 days of culture(63). These same ICC scaffolds were compared with two different nanofibrous gels, Matrigel and Puramatrix, over a period of 14 days for the support of HSPCs. Interestingly, ICC supported more direct cell contact, whereas the hydrogels depended on soluble factor secretion, suggesting that different platforms could be used to study different interactions in hematopoiesis(64).

For 3D BM disease models, it is critical that the scaffold material has dynamic biomimetic properties and can remain intact for the culture time required for evaluation of disease biology and therapeutic testing. Scaffolds composed of different materials, such as Poly(methyl methacrylate) (PMMA), PS, Poly(D,L-Lactic acid) (PDLLA) and PCL, have differentially been able to support AML cell lines(65). However, only PLGA and PU could remain intact and supported the culture over 2 months with improved biofunctionalization through coating with collagen I and/or fibronectin. Combinations of different scaffold materials have also resulted in improved AML cell line proliferation for 7 days when PLLA was combined with PU in 40:60 ratio and coated with fibronectin(66). Such combinations confer microenvironments for the survival of AML cell lines when exposed to cytarabine chemotherapy due to better cell adhesion and Bcl-2 pathway activation. Scaffold stiffness contributes to chemo-sensitivity through ECM-drug interactions, as observed in alginate-RGD scaffolds with drug sequestration or release dependent on scaffold fluidity(67).

Primary AML cells are challenging to maintain in 2D cultures, even with feeder layers and cytokines. In 3D systems, co-culture of AML cells with stroma or other supportive feeder layers provides the scaffold with a modified microenvironment made of multicellular niches, more representative of the in vivo state. Hydroxyapatite scaffolds coated with collagen I have been used to support co-cultures with allogeneic MSCs, osteoblasts and AML primary samples for 21 days (20% FBS) with resultant preservation of the original AML immunophenotype; MSCs were used to form primary supportive environments that were then used by leukemic cells to form multicellular niches, permitting the study of AML-MSC interactions though transcriptome analysis(68). An endosteum model was engineered to co-culture allogeneic osteoblasts with primary samples from patients with AML or cell lines for 21 days supplemented with 10% FBS in a polystyrene scaffold, and showed that the osteoblasts conferred a chemo-protective effect through AML-osteopontin binding(69). Similar co-cultures with either MSCs or stromal cells and AML cell lines for 7-14 days with 10% FBS, decreased drug-induced apoptosis in PGA/PLLA (90:10) 3D scaffolds(70) and, in a NanobioMatrix–poly(ε-caprolactone) scaffold, lower drug-induced metabolic modulations were observed with chemotherapy exposure(71). Similar chemoresistance was observed in a complex multi-cellular culture of primary AML samples and leukemia cell lines combined with endothelial and mesenchymal stem cells created using a star-PEG heparin hydrogel with RGD motifs, to mimic the vascular and endosteal niches, respectively(72). Since human BM has a heterogenous structure that can be challenging to emulate through the use of synthetic biomaterials, some investigators have instead used primary human decellularized bone as the scaffold. Cord blood derived decellularized Wharton jelly matrixes (DWJM) were harnessed to create scaffolds rich in ECM molecules (hyaluronic acid, collagen and fibronectin) to culture leukemic cell lines for up to 9 days in 5% FBS(73). Cells cultured in DWJM developed spindle-shaped morphology, enhanced clonogenicity and higher resistance to chemotherapy compared with those of 2D cultures. Similarly, plasma from patients with AML, chronic myeloid leukemia and CLL were used to develop a fibrin-mesh scaffold for the co-culture of leukemic cell lines and stromal cells for 7 days(74). Plasma-derived 3D scaffolds contained a mixture of humoral factors (cytokines, chemokines, growth factors) from primary patient samples and created a physiological tumor microenvironment that could support cell proliferation, recreate the protective effect of leukemic niches and allowed for evaluation of drug responses to combinations of nilotinib and integrin-linked kinase (ILK) inhibitor (Cpd22) for chronic myeloid leukemia, cytarabine and venetoclax for AML, and ibrutinib and venetoclax for CLL cell lines. The applied treatments induced over 50% cell death in 2D cultures whereas in 3D cultures cell death was less than 50%, and often none. These BM organoids illustrate the challenges in treating patients with AML – leukemic cells in BM niche microenvironments are protected from the effects of chemotherapy and lead to relapse and therapeutic resistance. Interestingly, 3D niches without MSCs or endothelial feeders have similar effects and shows that the 3D platform can be a powerful tool to dissect relapse and resistance kinetics as well as for use in pre-clinical drug-testing platforms. Since the metabolism of relapsed and resistant AML cells differ from that at diagnosis(75), altering treatment response profiles, stable long-term culture models operating under defined nutrient/metabolite feeding conditions would be required to model patient- and leukemia-specific heterogenous responses, and to create a robust personalized organoid for use in precision therapeutics. Ultimately, choice of 3D BM organoid model and culture conditions will depend on the questions being asked, the need for personalization, and finally must avoid allogenic and xenogeneic additives including cytokines where possible.

Despite the advantages of 3D static models over historic 2D culture methods, most of the state-of-the-art systems published to date do not recapitulate the natural heterogeneity of primary human samples, utilizing either manipulated feeder cells (allogenic, autologous or cell lines) or supplementation of exogenous cytokines and serum for hematopoietic cell propagation, with the inherent risk of cell selection and bias, introducing artefact to the system. These static batch operation culture models suffer from low metabolite turnover, resulting in metabolite accumulation and metabolic stress and/or adaptation to a metabolism that differs from that found in vivo; static cultures consume glucose at higher levels with concomitant accumulation of lactic acid due to poor metabolite turnover which results in metabolic shock that interferes with normal metabolism(34). Perfusion, at rates similar to those of normal BM, has been introduced to the 3D organoid in order to relieve the metabolic stresses of static cultures and as a means to harvest cells in a continuous closed system, akin to the function of the BM sinusoid, as well as to provide shear stress required by some cell populations to mature, mobilize and function.

Perfused 3D BM organoids offer improved tissue densities and culture control

Perfused 3D BM organoid models utilize a perfusion chamber or bioreactor for circulation of media through a continuous peristaltic pump(76, 77). Normal or abnormal hematopoietic cells can either be inoculated through a seeding port into the scaffold itself whilst the system undergoes continuous perfusion enabling more robust stochastic formation of cell niches, or scaffolds are combined with defined cell types prior to insertion into the perfusion device(78). Perfused models ensure better nutrient distribution, higher mass transfer and more physiologic gradients, thus enabling longer culture times and improved metabolite turnover preventing metabolic shocks due to waste accumulation or artificial metabolism(79–81). High tissue-like cell densities can be achieved with more complexity and cell niche diversity, mimicking natural cell development(82, 83). It is important to note that tissue density can alter the dynamics of the system resulting in different organoid function, affecting cytokine profiles, cell-cell interactions and maturation profiles(84–86). Likewise, parameters such as nutrient concentration, shear stress, oxygen level and soluble factor administration can be tuned throughout the culture, allowing long-term dynamic analysis of cell populations, modeling unique features such as cytokine and oxygen gradients as a function of cell type and differentiation kinetics(77, 79, 82, 85, 86). Unfortunately, the high cost of operation, fabrication and assembly challenges inherent in current 3D bioreactors prevents high-throughput research using these models(82).

To date, two types of 3D perfused organoids have been produced: hollow fiber bioreactors that perfuse through the 3D scaffold, and perfusion chambers that perfuse media around the scaffold. Human erythropoiesis was modeled in a perfused 3D hollow-fiber bioreactor using PU coated with collagen I and seeded with unselected CBMNCs, without feeder cells, serum or cytokines apart from physiological concentrations of erythropoietin and SCF(84, 85). The platform supported long-term (over 4 weeks) multi-lineal hematopoiesis, producing diverse microenvironments within the same bioreactor that changed over time and supported the required conditions for erythropoiesis through to enucleation, forming diverse cell niches with HSPCs, osteoblasts and stromal cells achieving cell densities similar to that of human BM. Two dynamic culture phases were identified: an adaptive phase (D0-D14) wherein the microenvironment was formed, including endogenous production of cytokines and ECM, and a functional phase (D15-D28) with establishment of multi-lineal hematopoiesis with continuous production and harvest of enucleated erythrocytes via in situ formation of erythroblastic islands, mimicking the BM cell manufacturing function.

Another hollow-fiber reactor with an hydroxyapatite scaffold was used to culture BM mononuclear cells in 5% human serum(87). Niches were formed predominantly by autologous MSCs and, in some, osteoblasts, with few hematopoietic cells attached. Endothelial cells and erythrocytes were present throughout the 42 days of culture. Despite low numbers, the bioreactor microenvironment maintained metabolically active cells (consuming glucose and producing lactate) and supported higher colony forming unit capacity of hematopoietic progenitors compared with 2D cultures. In contrast, a hydroxyapatite scaffold placed in a perfusion chamber was used to culture cord blood derived HSPCs(88). The scaffold was preconditioned for 3 weeks with allogeneic human MSCs cultured in osteogenic media allowing osteoblastic differentiation, ECM deposition, soluble factor secretion and niche pre-formation before adding the allogeneic CD34⁺-selected human HSPCs and culturing for an extra week in media supplemented by SCF 10ng/mL, FLT3L 10ng/mL, and TPO 10 ng/mL. The MSCs created an endosteal-like supportive microenvironment enabling multipotent and lymphoid progenitor maintenance and proliferation with in situ cytokine production by both HSPCs and MSCs. Myeloid and lymphoid lineage reconstitution was observed in myeloablated mice engrafted with the cultured cells, suggesting preservation of stem cell properties. Hematopoietic niches could be customized by viral transduction of feeder MSCs in order to alter cytokine secretion levels. In parallel to the endosteal niche 3D scaffold created by pre-seeding with MSCs differentiated into osteoblasts, human adipose tissue-derived stromal-vascular cells were pre-seeded onto another scaffold to simulate the vascular microenvironment(89). Primary human AML and myeloproliferative neoplasm cells could be cultured over 3 weeks in the perfused organoids, supplemented with SCF 10ng/mL, FLT3L 10ng/mL and TPO 10ng/mL. Engraftment in the scaffolds in some cases preserved the original genotype, but in others was skewed towards subclones, possibly due to the cytokines used during expansion phases, and is indicative of the challenges in maintaining the fidelity of the myeloid neoplasm for study in any personalized BM organoid. As in other 3D models, the AML cells were more chemo-resistant when compared with those cultured in feeder-free scaffold controls, highlighting the protective effects of the multicellular niches. Unfortunately, as the microenvironments were prescribed and selected cell populations were used for scaffold seeding, lymphoid and other microenvironment populations were not observed. A similar perfusion system with HSPC-allogeneic MSCs co-cultured in a porous PEG hydrogel for 21 days supplemented with TPO, SCF, Flt-3L and IL-3 maintained more CD34⁺ cells and better mimicked responses to 5-fluorouracil compared with static controls, again illustrating the chemo-protective niche environment(90).

Murine-Scaffold Hybrid Models

In order to overcome the limitations of microenvironmental signals inherently lacking in ex vivo systems, human BM organoids can be implanted subcutaneously into mice to produce a model that is easily accessible and has more human cellular design elements than that of the PDX mouse rendering it a relevant hybrid approach for the study of hematopoiesis and disease. Following implantation, the ossicle is colonized by adipocytes, endothelial and hematopoietic cells with vascular supply from the mouse to recreate the microenvironment necessary for human HSPCs or leukemic cells to proliferate and differentiate in long term studies. Such scaffolds create a humanized microenvironment within the animal capable of supporting cell expansion, retaining similar physiology and genetic characteristics to the human tumor, even in the support of extra-medullary hematopoiesis. Several immunodeficient mouse strains can harbor cell-seeded scaffolds including the most popular, NSG and NSGS, though these models also use selected cells and feeder layers (such as mesenchymal stem cells) to set-up the scaffold microenvironment(15, 16).

Implantable hydrogels with humanized microenvironments established prior to mouse implantation have also been used in hybrid models. In one model, a polyacrylamide hydrogel was used to culture allogeneic human stromal cells for 3 days to form a human BM microenvironment, followed by implantation into the mouse and direct scaffold injection of normal or leukemic CD34⁺ cells 4 weeks later(91). The result was a vascularized scaffold that contained both human and mouse HSPC-supportive niches that could be used to study normal and leukemic hematopoiesis and cytokine production. Similarly, endothelial colony-forming feeder layer cells were added to human MSCs and seeded onto a Matrigel scaffold(92). As a result, an extramedullary bone structure was formed, exhibiting hypoxic areas, and enabling HSPC homing, revealing differences due to niche hypoxia in normal and malignant states. In a different hybrid model, a calcium phosphate-coated polycaprolactone scaffold was used to culture human MSCs in osteogenic conditions to generate a bone-like microenvironment that exhibited a structure similar to that of native tissue which derived vascular supply from the murine host(93). After injection of CD34⁺ cells, the humanized scaffold microenvironment allowed HSPC homing and engraftment, and the maintenance of different hematopoietic clusters including myeloid, lymphoid and multipotent progenitors. Dehydrated gelatin-based gelfoam has also been used to culture cord blood HSPCs supported by osteogenic-differentiated human MSCs followed by mouse ectopic implantation(94). The implanted ossicle contained a humanized microenvironment amenable to HSPC engraftment with little spread to the host BM, exhibited adipogenic and continued osteogenic differentiation and even HSPC homing when cells were delivered intravenously.

Such hybrid models have enabled study of patient samples that could not be engrafted in traditional PDX models(94) or which traditionally have been more challenging to study in vivo, such as acute promyelocytic leukemia and myelofibrosis. Though many of these models have used Matrigel, they do generate ossicles that are capable of maintaining both myeloid and lymphoid leukemic cells while preserving clonal heterogeneity better than when cells are directly engrafted into mice(95). Similarly, a ceramic scaffold coated with MSCs successfully engrafted AML cells, preserving phenotype and stemness in a microenvironment that included murine vascularization, and differentiation of adipocytes with generation of bone tissue(96). PU scaffolds have also been used to co-culture primary AML samples with allogeneic MSC feeder cells, with stromal differentiation towards adipogenic and osteoblastic lineages and vascularization that formed a humanized BM-like microenvironment(97). The latter supported quiescent AML niches and cell-cell niche interactions were observed through formation of pseudopods between hematopoietic cells and MSCs and via SDF-1 secretion.

Despite the benefits of a humanized hematopoietic microenvironment that can be generated or designed, the mouse-scaffold hybrid models have limitations, similar to those encountered in PDX mice: low engraftment rate, long time to engraftment, requirement for irradiation which damages the BM stroma and undermines studies on cell-stroma interactions, and the lack of immune cells in the microenvironment due to the characteristics of the immunodeficient mouse and lymphodepletion of the human sample required for the graft to take(68), though some degree of chimerism has been reported(15, 26). Their main limitation is that the tumor microenvironment is manufactured either in vivo by implanting the scaffold into a mouse, or in vitro by engineering blood vessels with endothelial and mesenchymal cells, both with the addition of exogenous soluble factors, rather than allowed to form spontaneously within the organoid, and is prohibitive to the development of truly AML-specific or patient-specific personalized organoid systems that can recapitulate native microenvironments with fidelity to the original source. This aspect of personalization is critical to the development of an AML organoid, since structural niche organization is heterogeneous and varies from patient to patient impacting patient-specific drug responses and disease kinetics.

Microfluidic devices in hematopoiesis research

Marrow-on-a-chip and microfluidic devices have been used as short-term culture models for normal hematopoiesis, AML and other hematological malignancies(98). These devices culture cells in a continuously perfused micrometer-sized chamber, which tries to model physiological functions of BM, and is usually comprised of a porous membrane, scaffold or hydrogel and biofunctionalized with ECM or adhesion molecules and soluble factors to create a more physiological 3D environment that mimics the BM. Nutrient perfusion allows the formation of gradients, adds shear stress and improves metabolite turnover during culture time(99). Microfluidics have controllable parameters such as concentration gradients, dynamic stress or shear forces that can be manipulated by the operator, allowing the study of different microenvironment conditions, and have been applied as preclinical models to study new drug and toxic effects which may be anticipated in BM while reducing the number of animals for such experiments(100, 101).

One of the advantages of the engineered microfluidics device is that it enables the recreation of selected BM compartments to study under physiological flow conditions. Fibrin hydrogel has been used to create a marrow-on-a-chip model seeded with HSPCs to recreate the vascular and endosteal niches in two separate but interconnected chambers: (1) co-culture of endothelial and stromal cells, (2) co-culture of fetal osteoblasts and endothelial cells(102). The system supported HSPC proliferation and differentiation towards neutrophils for 2 weeks. A similar fibrin hydrogel was used to mimic the vascular microenvironment by co-culturing CD34⁺ cells and MSCs in microfluidic device channels connected by a porous membrane to an endothelial cell-containing parallel channel(103). Formed niches could support myelo-erythroid proliferation over a 4-week period. A similar co-culture set-up in a hydroxyapatite coated zirconium oxide scaffold was used in a microfluidics device to support the differentiation of granulocyte, erythrocyte, macrophage, and megakaryocyte colony formation over a 4-week period(104). Cell proliferation was favored by a microenvironment tailored by MSC-derived adhesion molecules (fibronectin, osteopontin, ICAM1) and soluble factors (SCF, Ang-1, VEGF). Capitalizing on the natural composition of human BM, demineralized bone matrix coated with collagen I has also been used in MSC-AML cell line co-cultures in a microfluidic device(105). The resultant microenvironment not only retained the original leukemic phenotype but also formed niches and protected cells from chemotherapy through BCL-2 activation. Similarly, acute lymphoblastic leukemia cell lines cultured in a device using a collagen matrix to encapsulate osteoblasts and stroma cells were able to migrate and form cell-cell interactions in niches and conferred chemoresistance(106).

BM-on-a-chip can also be applied following removal of the scaffold from a hybrid model to create hematopoietic niches in both vascular and endosteal microenvironments(107). A bone-like tissue scaffold was engineered and filled with marrow using a cylindrical poly(dimethysiloxane) (PDMS) scaffold functionalized with collagen I and bone-inducing factors and allowed to engraft in the subcutaneous tissue of a mouse for 8 weeks, creating a BM-like microenvironment containing HSPCs, myeloid, lymphoid and erythroid populations. The extracted in vivo engineered scaffold was placed in a microfluidic device and exhibited multi-lineal output for at least one week. To study cell trafficking in the BM vascular compartment, device microchannels were collagen coated and multi-cultured with endothelial, stromal cells and fibroblasts then perfused with monocytes, CD34⁺ HSPCs and AML cells(108). The manufactured vascular microenvironment recapitulated cell-cell interactions and permitted the study of adhesion and migration of hematopoietic and non-hematopoietic cells. Unlike most 3D static models, cell niches are not spontaneously formed but artificially engineered and generated by a limited number of cells in each device(109). Consequently, microfluidic niches do not preserve the original niche architecture, complexity or composition, have small microchannel dimensions wherein surface effects predominate over volume effects which limits spatial interactions and oversimplifies niche interactions(110).

Conclusions

Great strides have been made in generating different 3D organoid models which can recapitulate the BM organ structure and function more faithfully than pioneering 2D Dexter-like culture systems used to date. Though there are currently two different strategies in how best to create 3D BM organoids, be it with designer-created microenvironments or by promoting autonomously formed self-organized niches, all models share the same aim: to accurately recreate a biomimicry of the BM microenvironment in vitro that represents in vivo cell interactions. Fidelity of the microenvironment is key in ensuring a robust BM organoid to study both normal and abnormal hematopioesis and, though cell source matters, ultimately the biomaterial used, biofunctionalization of the scaffold and operating conditions of the culture itself are critical components affecting cell fates. More emphasis on the inherent innate and adaptive immune populations and their maintenance in heterogeneous culture systems is required for improved understanding of normal or abnormal BM and successful application of therapy. The identification of key supplemental factors not endogenously produced in the BM, such as sympathetic nervous signals, hormonal and extramedullary cytokines (e.g., TPO and erythropoietin) in physiological concentrations will require special consideration. Improved 3D printing technologies and novel biofunctionalized biomaterials will likely lead to further advances in the field along with standardized platforms for perfusion which is key to the maintenance of tissue-like densities, complex heterogenous microenvironments with inclusion of lymphoid populations, and endogenous self-organizing/self-sustaining dynamic cultures to aid in the development and testing of novel personalized precision therapeutics, including in immunotherapy.