Bone marrow–on–a–chip replicates hematopoietic niche physiology in vitro

Bone marrow formed in a cylindrical PDMS device implanted in a mouse can be surgically removed and cultured for a week in vitro without losing any of the hallmarks of in vivo bone marrow niches. Current in vitro hematopoiesis models fail to demonstrate the cellular diversity and complex functions of living bone marrow; hence, most translational studies relevant to the hematologic system are conducted in live animals. Here we describe a method for fabricating 'bone marrow–on–a–chip' that permits culture of living marrow with a functional hematopoietic niche in vitro by first engineering new bone in vivo, removing it whole and perfusing it with culture medium in a microfluidic device. The engineered bone marrow (eBM) retains hematopoietic stem and progenitor cells in normal in vivo–like proportions for at least 1 week in culture. eBM models organ-level marrow toxicity responses and protective effects of radiation countermeasure drugs, whereas conventional bone marrow culture methods do not. This biomimetic microdevice offers a new approach for analysis of drug responses and toxicities in bone marrow as well as for study of hematopoiesis and hematologic diseases in vitro.

current in vitro hematopoiesis models fail to demonstrate the cellular diversity and complex functions of living bone marrow; hence, most translational studies relevant to the hematologic system are conducted in live animals. here we describe a method for fabricating 'bone marrow-on-a-chip' that permits culture of living marrow with a functional hematopoietic niche in vitro by first engineering new bone in vivo, removing it whole and perfusing it with culture medium in a microfluidic device. the engineered bone marrow (eBm) retains hematopoietic stem and progenitor cells in normal in vivo-like proportions for at least 1 week in culture. eBm models organ-level marrow toxicity responses and protective effects of radiation countermeasure drugs, whereas conventional bone marrow culture methods do not. this biomimetic microdevice offers a new approach for analysis of drug responses and toxicities in bone marrow as well as for study of hematopoiesis and hematologic diseases in vitro.
The bone marrow microenvironment contains a complex set of cellular, chemical, structural and physical cues necessary to maintain the viability and function of the hematopoietic system [1][2][3][4][5] . This hematopoietic niche regulates hematopoietic stem cells (HSCs), facilitating a delicate balance between self-renewal and differentiation into progenitor cells that produce all mature blood cell types 4,5 . Engineering an artificial bone marrow that reconstitutes natural marrow structure and function, and that can be maintained in culture, could be a powerful platform to study hematopoiesis and test new therapeutics. It has proven difficult, however, to recreate the complex bone marrow microenvironment needed to support the formation and maintenance of a complete, functional hematopoietic niche in vitro [6][7][8][9] . Although various in vitro culture systems have been developed to maintain and expand HSCs and progenitor cells [6][7][8][9][10][11] , there is currently no method to recreate or study the intact bone marrow microenvironment in vitro. Therefore, studies on hematopoiesis commonly rely on animal models to ensure the presence of an intact bone marrow Bone marrow-on-a-chip replicates hematopoietic niche physiology in vitro Yu-suke Torisawa 1,7 , Catherine S Spina 1,2,7 , Tadanori Mammoto 3 , Akiko Mammoto 3 , James C Weaver 1 , Tracy Tat 3 , James J Collins 1,2,4,5 & Donald E Ingber 1,3,6 microenvironment that enables normal physiologic marrow responses [12][13][14][15] . Furthermore, although it has been reported that bone marrow can be engineered in vivo [16][17][18][19] , no method exists to culture engineered bone marrow in vitro. To bridge the functional gap between in vivo and in vitro systems, we developed a method to produce a bone marrow-on-a-chip culture system that contains artificial bone and living marrow. The bone with marrow is first generated in mice and then explanted whole and maintained in vitro within a microfluidic device.

In vivo engineering of bone marrow
Tissue engineering methods have been used to induce the formation of new bone with a central marrow compartment in vivo [18][19][20][21] . To explore the possibility of engineering an artificial bone marrow that can be explanted whole, we microfabricated a poly(dimethylsiloxane) (PDMS) device with a central cylindrical cavity (1 mm high × 4 mm in diameter) with openings at both ends (Fig. 1a). We filled the hollow compartment with a type I collagen gel containing bone-inducing demineralized bone powder (DBP) and bone morphogenetic proteins (BMP2 and BMP4) [20][21][22] and implanted the device subcutaneously in the back of a mouse (Supplementary Fig. 1). Our goal was to engineer bone that would fill the cylindrical space within the implanted device so that it could be easily removed whole and inserted into a microfluidic system containing a similarly shaped chamber for in vitro culture (Fig. 1a,b). These initial studies resulted in the creation of new bone encasing a marrow compartment that formed within the PDMS device 4-8 weeks after subcutaneous implantation. Histological analysis revealed that the marrow was largely inhabited by adipocytes and that it exhibited a low level of hematopoietic cell contribution, even 8 weeks after implantation (Fig. 1c), as previously noted by others using similar tissue engineering approaches with bone-inducing materials [19][20][21] .
The presence of large numbers of adipocytes in bone marrow can inhibit hematopoiesis 23 . To reduce adipocyte content in the marrow, we sealed the top of the central cavity in the implanted device by adding a solid layer of PDMS to restrict access of cells or soluble factors from the overlying adipocyte-rich hypodermis to the bone-inducing materials while maintaining accessibility to the underlying muscle tissue through the lower opening ( Fig. 1a and Supplementary Fig. 1). Subcutaneous implantation of this improved PDMS device resulted in the formation of a cylindrical disk of white, bone-like tissue containing a central region of blood-filled marrow over a period of 8 weeks (Fig. 1b and Supplementary Fig. 2). Histological analysis confirmed the presence of a shell of cortical bone of relatively uniform thickness surrounding marrow that was dominated by hematopoietic cells and that contained few adipocytes (Fig. 1c). Comparison of histological sections of the eBM to sections from an intact femur confirmed that the morphology of the eBM was nearly identical to that of natural bone marrow (Fig. 1c).
Micro-computed tomographic (micro-CT) analysis of the eBM demonstrated that the newly formed cortical shell of bone also contained an ordered internal trabecular network that closely resembles the intricate architecture found in normal adult mouse vertebrae (Supplementary Fig. 3) and that is known to be supportive of HSCs 24 (Fig. 1d). Compositional analysis using energydispersive X-ray spectroscopy (EDS) also showed that the calcium and phosphorous content of the eBM were indistinguishable from that of natural trabecular bone (Supplementary Fig. 3).

characterization of engineered bone marrow
Interactions between CXCL12 expressed on the surfaces of various cell types in the bone marrow (such as osteoblasts 25 , perivascular endothelial and perivascular stromal cells 26 ) and its cognate receptor CXCR4 on the surfaces of HSCs and hematopoietic progenitor cells are critical for the recruitment, retention and maintenance of HSCs [26][27][28] . Immunohistochemical analysis confirmed that both of these key hematopoietic regulators were expressed in their normal positions in the eBM: CXCL12 localized to cells lining the inner surface of the bone and blood vessels, and CXCR4 was expressed by clusters of lymphoid cells in the endosteal and perivascular niches ( Fig. 2a-d). We also confirmed that key hematopoietic niche cells 3 including perivascular nestin + cells and leptin receptor + cells, as well as CD31 + vascular endothelial cells, resided in their normal positions (Supplementary Fig. 4).
To rigorously characterize the hematopoietic content of the engineered marrow, we harvested cells from the eBM immediately after surgical removal and analyzed them by flow cytometry. The cellular components of the marrow contained within the eBM were compared to hematopoietic populations isolated from femur bone marrow and peripheral blood from the same mice ( Fig. 2e,g). Devices harvested 4 and 8 weeks after implantation contained all blood cell types, including HSCs that are not recognized by a mixture of Lin antibodies that recognize mature, lineage-restricted blood cells (Lin − Sca1 + cKit + CD34 +/− , Lin − Sca1 + cKit + CD150 +/− CD48 −/+ ) and hematopoietic progenitor cells identified by four different marker sets (Lin − Sca1 + , Lin − cKit + , Lin − CD34 + , Lin − CD135 + ), as well as mature erythrocytes (Ter119 + ), lymphocytes (T cells, CD45 + CD3 + ; B cells, CD45 + CD19 + ) and myeloid cells (CD45 + Mac1 +/− Gr1 +/− ). The eBM harvested 4 weeks after implantation did not appear to be fully developed, as indicated by a lower proportion of HSCs and hematopoietic progenitor cells compared to that in normal marrow (Fig. 2f,h). However, cells harvested from the eBM 8 weeks after implantation exhibited a completely normal distribution of HSCs, hematopoietic progenitors and differentiated blood cells from all lineages that was nearly identical to that displayed by natural bone marrow (Fig. 2f,h and Supplementary Figs. 5 and 6).
In summary, our modified strategy for eBM produced a cylindrical disk of cortical and trabecular bone (Supplementary Fig. 3) containing marrow with a hematopoietic cell composition nearly identical to that of natural bone marrow. The presence of key cellular and molecular components of the hematopoietic niche suggests that the cellular content of the eBM closely resembles the natural bone environment.  npg

In vitro culture of engineered bone marrow
To determine whether the eBM could maintain a functional hematopoietic system in vitro, we surgically removed the eBM formed 8 weeks after implantation from the mouse, punctured it in multiple places with a surgical needle to permit fluid access and cultured it in another clear PDMS microfluidic device containing a similarly shaped cylindrical central chamber that is separated from overlying and underlying microfluidic channels by porous membranes (Fig. 1a,b). To maintain the cellular viability of the eBM, we perfused culture medium through the top and bottom channels using a syringe pump at an optimal rate (1 µl/min) ( Supplementary  Fig. 7) once the eBM was inserted into the central chamber and the surrounding porous membranes and microchannel layers were attached. The eBM was cultured in vitro for 4 or 7 d within the bone marrow-chip microsystem (Fig. 1b), which covers a time period that is commonly used to test for drug efficacies and toxicities in vitro 29,30 . The cultured bone and marrow retained their morphology during this time, including the distribution of CXCL12expressing stromal cells (Supplementary Fig. 8). Stroma-supported culture systems represent the current benchmark for maintaining survival of HSCs and hematopoietic progenitor cells in vitro 7,31 .
Thus, we used flow cytometric analysis to compare the hematopoietic cellular composition of the cultured bone marrow-on-a-chip to that of marrow isolated from mouse femur cultured for the same amount of time on a stromal 'feeder' cell layer ( Supplementary  Fig. 9). Because past work has shown that the addition of cytokines is required to maintain or expand HSCs and their progenitors 6,7 , and because serum can suppress the marrow-reconstituting activity of HSCs 32 , the stroma-supported cultures were maintained in serum-free medium supplemented with cytokines (mSCF, mIL-11, mFLt-3 ligand and hLDL) that have been shown by others to more efficiently maintain and expand both HSCs and hematopoietic progenitor cell populations in vitro 33 . Our analysis revealed that there was no significant difference in cell viability after 4 or 7 d of culture in the microfluidic eBM device compared to the static stroma-supported culture (Supplementary Fig. 10). However, bone marrow cultured on stroma exhibited a significant decrease (P < 0.0005) in the number of long-term HSCs (Lin − CD150 + CD48 − cells) and a concomitant increase (P < 0.0005) in hematopoietic progenitor cells (Lin − CD34 + , Lin − Sca1 + , Lin − cKit + ) relative to cells freshly isolated from natural mouse bone marrow (Fig. 3a,b)  npg self-renewal and multilineage potential, appeared to be differentiating into more specialized progenitor cells in the static stromasupported culture system, as previously reported [6][7][8][9] . In contrast, the number and distribution of HSCs and hematopoietic progenitor cells in the eBM cultured for up to 7 d on-chip were maintained in similar proportions to those of freshly harvested bone marrow (Fig. 3a). The bone marrow-on-a-chip enabled maintenance of a significantly higher proportion of long-term HSCs while more effectively maintaining the distribution of mature blood cells compared to the stroma-supported cultures (Fig. 3b,c). Interestingly, although the proportions of hematopoietic cells were retained over this culture period, there was no significant difference in the number or viability of cells cultured on-chip for 7 d compared to 4 d; hence, the HSCs and hematopoietic progenitor cells appeared to remain relatively quiescent in the marrow-on-a-chip microdevice. Moreover, although addition of exogenous (and expensive) cytokines, including mSCF, mIL-11, mFLt-3 ligand and hLDL, are critical for maintenance of these cell populations in conventional stroma-supported cultures 6,7,33 , their removal from culture medium had little effect on the distribution of HSCs and hematopoietic progenitors in the cultured eBM (Fig. 3d). Thus, the eBM contained a functional hematopoietic niche that behaved in an autonomous fashion to support the continued survival of these critical blood-forming stem and progenitor cells in vitro. Blood cell populations could be maintained in normal proportions for at least 1 week under microfluidic flow in vitro, even in the absence of exogenous cytokines.
To confirm that the HSCs and hematopoietic progenitor cells retained in the cultured bone marrow-on-a-chip remained truly functional, we evaluated their self-renewal and differentiation capabilities by testing engraftment and hematopoietic reconstitution potential following transplantation into lethally irradiated, syngeneic recipient mice. Cells within the marrow compartments of eBMs that were formed in GFP-expressing animals and cultured on-chip for 4 d were harvested and transplanted into γ-irradiated mice; results were compared to those from irradiated mice transplanted with cells from freshly harvested bone marrow from mouse femur. Total engraftment was assessed in the peripheral blood of recipient mice 6 and 16 weeks after transplantation to confirm the presence of functional short-and long-term HSCs, respectively. Cells harvested from the eBM after 4 d in culture on-chip successfully engrafted the mice at a similar rate to that of freshly isolated, uncultured bone marrow (Fig. 3e) and repopulated all differentiated blood cell lineages (Fig. 3f), showing 70% and 85% engraftment by 6 and 16 weeks after transplantation, respectively. These data confirmed that the hematopoietic compartment of the eBM retained fully functional, self-renewing, multipotent HSCs after it was cultured in the microfluidic bone marrow chip for 4 d in vitro.

In vitro model for radiation toxicity
The functionality and organ-level responsiveness of the bone marrow-on-a-chip were tested by exposing the eBM to varying doses of γ-radiation to determine whether this method could be used as an in vitro model for radiation toxicity, which currently can only be studied in live animals. Live mice, eBMs cultured on-chip and marrow cells maintained in stromasupported culture were exposed to 1-and 4-Gy doses of γ-radiation, npg which have been shown to produce marrow toxicity in mice 34 , and were maintained in culture. In measurements made 4 d after radiation exposure, we detected a statistically significant, radiation dose-dependent decrease in the proportion of HSCs, hematopoietic progenitors, lymphoid cells and myeloid cells (Fig. 4a-e), which closely mimics what is observed in the bone marrow of live irradiated mice. Interestingly, the proportion of HSCs (Lin − Sca1 + cKit + ) or progenitors (Lin − CD34 + ) observed in eBM after exposure to 1-and 4-Gy doses of γ-radiation were nearly identical to the proportions measured in whole marrow from live mice that underwent similar irradiation. In contrast, the proportion of HSCs and progenitors were significantly lower (P < 0.05) in the stroma-supported culture after 4-Gy irradiation compared to after 1-Gy irradiation. Various types of marrow cells (HSCs, progenitors, lymphoid cells and myeloid cells) cultured on stroma also exhibited suppressed responses, and all were significantly more resistant (P < 0.05) to the effects of radiation toxicity (Fig. 4a-e and Supplementary Fig. 11).
To further evaluate the functional relevance and power of our system, we tested the effects of administering granulocyte colonystimulating factor (G-CSF), which has been shown to accelerate recovery and prevent potentially lethal bone marrow failure following radiation exposure in vivo 35 . When G-CSF was added to the eBM cultured on-chip 1 d after exposure to γ-radiation, samples analyzed 3 d later demonstrated a significant increase in the total number of HSCs (Lin − Sca1 + cKit + ) and hematopoietic progenitor cells (Lin − cKit + , Lin − CD34 + ) compared to untreated bone marrow chips that were similarly irradiated (Fig. 4f). These findings suggest that G-CSF induced proliferation of HSCs and hematopoietic progenitor cells in the bone marrow chip in vitro, as previously reported in vivo 35 . These data clearly demonstrate that the bone marrow-on-a-chip faithfully mimicked the natural physiological response of living bone marrow to clinically relevant doses of γ-radiation and to a validated radiation countermeasure drug (G-CSF), whereas conventional stromasupported cultures do not.

discussion
Our bone marrow-on-a-chip fabrication strategy provides a proof of concept for the creation of an organ-on-chip device that reconstitutes and sustains an intact, functional, living bone marrow when cultured in vitro. This strategy differs substantially from conventional tissue engineering approaches in which materials or living cells are implanted in vivo without geometric constraint and without any intent of removing the newly formed organ and maintaining its viability ex vivo. Although we regenerated the complex structural, physical and cellular microenvironment of whole bone marrow by employing in vivo tissue engineering techniques, we then leveraged microfluidic strategies to deliver nutrients, chemicals and other soluble signals in a fashion that supports the continued viability and function of this engineered organ in vitro. This also differs from most organ-on-chip methods that use microengineering and microfluidics approaches to model tissue architecture, cell-cell relationships, chemical gradients and the mechanical microenvironment and then populate the devices with cultured cell lines or isolated stem cells 36 .
Our in vivo engineering approach enabled us to reconstitute hematopoietic niche physiology and restore complex tissuelevel functions of natural bone marrow. The eBM autonomously npg produces the factors necessary to support the maintenance and function of the hematopoietic system in vitro, which offers a major practical advantage over existing culture systems in that expensive growth supplements can be removed from the culture medium or greatly reduced. Another advantage is that the bone marrow-on-a-chip supports HSCs and progenitor cells in normal in vivo-like proportions relative to the other hematopoietic cell populations and maintains their spatial positions within a fully formed three-dimensional bone marrow niche in vitro. These features of the bone marrow-on-a-chip are likely key to its ability to preserve complex functionalities of the whole organ that cannot be replicated by conventional stroma-supported cultures. Notably, the use of microfluidics also enables analysis of responses under flow, which is important for both the regulation of marrow physiology 5,6,10,37,38 and the study of pharmacokinetic and pharmacodynamic behaviors of drugs that are critical for evaluation of their clinical behavior. The eBM cultured on-chip mimics complex tissue-level responses to radiation toxicity normally observed only in vivo and to a therapeutic countermeasure agent (G-CSF) that is known to accelerate recovery from radiation-induced toxicity in patients 39 . Thus, this biomimetic microsystem could serve as a valuable in vitro replacement for whole animals in the testing and development of drugs and other medical countermeasures that might protect against radiation poisoning in the future. The completeness of our organ mimic permits us to recapitulate the physiologic responses of the whole hematopoietic niche to clinically relevant cues (such as cytokines, drugs and radiation), whereas conventional cell cultures do not. This finding underscores the novelty of maintaining functional marrow containing multiple components of the hematopoietic niche in vitro rather than merely culturing particular hematopoietic cell types.
The bone marrow-on-a-chip provides an interesting alternative to animal models because it offers the ability to manipulate individual hematopoietic cell populations (such as genetically or using drugs), or to insert other cell types (such as tumor cells) in vitro, before analyzing the response of the intact marrow to relevant clinical challenges, including radiation or pharmaceuticals. It also might be possible to generate human bone marrow models: for example, an eBM could be engineered in immunocompromised mice (such as NOD.Cg-Prkdc scid Il2rg tm1Wjl /Sz; NSG) that have their endogenous marrow cells replaced with human hematopoietic cells.
The ability to produce trabecular bone with architectural and compositional properties similar to those of natural bone offers a way to produce bones of predefined size and shape, and it could represent a new method for the study of bone biology, remodeling and pathophysiology in vitro. Therefore, the bone marrow-on-a-chip is a powerful method to accelerate discovery and development in a wide range of biomedical fields ranging from hematology, oncology and drug discovery to tissue engineering.

methods
Methods and any associated references are available in the online version of the paper. bonded together using a plasma etcher in air for 30 s. To allow introduction of solution into the channels, we punched access holes through the top channel layer with a 2-mm biopsy punch, and the inlets and the outlets were connected with tubes (i.d. = 1/32 inch). The microfluidic device was oxidized using a plasma etcher in air for 10 min to make the PDMS surface hydrophilic. The eBM was inserted into the central chamber (which was bonded to the bottom channel layer) before attachment of the top channel layer. The microfluidic device was placed between two acrylic plates (30 mm in diameter) made by a laser cutter and immobilized using screws (Fig. 1b). To maintain cellular viability of the eBM, we perfused culture medium (SFEM basal medium, StemCell Technologies) containing cytokines 32,33 (50 ng/mL mouse SCF, 100 ng/mL mouse IL-11, 100 ng/mL mouse FLt-3, and 20 µg/mL human LDL, StemCell Technologies) through the top and bottom channels (1 µL/min, 0.005 dyn/cm 2 ) using a syringe pump (BS-8000, Braintree Scientific).
Stroma-supported bone marrow cell culture. Bone marrow stromal cells were harvested from 8-to 12-week-old C57BL/6 mice, resuspended in DMEM medium (Gibco) containing 20% FBS (Gibco), GlutaMAX (Gibco) and 100 units/mL penicillinstreptomycin (Gibco), and were cultured in the same medium on tissue culture plates (Falcon), with the medium changed every other day to create stromal feeder layers. After the adherent monolayer became established (about 3 weeks), the cells were irradiated with 12 Gy. Bone marrow cells harvested from femurs of C57BL/6-Tg(UBC-GFP)30Scha/J mice were cultured on this bone marrow stromal cell layer using the same culture medium used in the microfluidic culture.
Bone marrow transplantation. Bone marrow transplantation was performed on 8-week-old C57BL/6 mice exposed to two doses of radiation measuring 6 Gy separated by 2-3 h. Because these mice were purchased from a vendor and randomized upon arrival, a randomization procedure was not conducted. The bone marrow cells were harvested from femurs of C57BL/6-Tg(UBC-GFP)30Scha/J mice or from eBM produced in similar GFP-labeled mice after 4 d of microfluidic culture. 2.5 × 10 5 bone marrow cells were delivered by intravenous (i.v.) tail-vein injection within 12 h of lethal irradiation. Engraftment was measured 6 weeks and 16 weeks after transplant using retro-orbital bleeds and flow cytometric analyses.
Micro-computed tomographic (micro-CT) analysis. eBM harvested from mice 4 and 8 weeks after device implantation were fixed for 48 h in 4% paraformaldehyde and stored in 70% ethanol at 4 °C. Vertebrae harvested from the same mice immediately after device removal were handled similarly. Both the eBM and vertebrae were imaged (in 70% ethanol) with an XRA-002 X-Tek MicroCT system. X-ray transmission images were acquired at 55 kV and 200 µA, and the 3D reconstructions were performed using CT-Pro (Nikon Metrology); surface renderings were generated using VGStudio Max.
Compositional backscattered scanning electron (BSE) micrographs and elemental mapping. eBM and vertebrae harvested and fixed as described for micro-CT were serially dehydrated into 100% ethanol and then embedded in Spurr's resin and sectioned at the desired imaging plane using a slow-speed diamond saw. The resulting sections were polished with silicon carbide papers down to P1200, sputter coated with gold and examined using a Tescan Vega-3 scanning electron microscope equipped with a Bruker X-Flash 530 energy-dispersive spectrometer (EDS). All EDS spectra and elemental maps were acquired at 20-keV accelerator voltage. For calculating elemental composition of both the sectioned implant and vertebra samples, ten point spectra from the surface of each sample were acquired, and the percent phosphorous and calcium content was determined by averaging the obtained values ± s.e.m. g-radiation. Freshly harvested eBM made in C57BL/6-Tg(UBC-GFP)30Scha/J (Jackson Laboratories) mice, 1 × 10 7 mouse femur bone marrow cells maintained in stroma-supported culture and 8-week-old C57BL/6-Tg(UBC-GFP)30Scha/J were exposed to one dose of γ-irradiation (Cs-137) at 1 Gy or 4 Gy. 96 h after irradiation, marrow from the eBM cultured on-chip, bone marrow in stroma-supported culture and bone marrow from the femurs of live mice were collected for flow cytometric analysis. 500 U/mL granulocyte colony-stimulating factor (G-CSF, Sigma-Aldrich) was added in the culture medium containing cytokines 24 h after exposure to γ-irradiation. After 72 h in culture on-chip with G-CSF, marrow from the eBM was collected for flow cytometric analysis.
Statistics. Sample size for in vitro and in vivo experiments was determined on the basis of a minimum of n = 3 biological replicates. Statistical differences were analyzed by Student's t-test. All statistical evaluation was conducted using a two-tailed t-test, assuming independent samples of normal distribution with equal variance. P < 0.05 was deemed statistically significant; all error bars indicate s.e.m. npg