Deconstructed Microfluidic Bone Marrow On-A-Chip to Study Normal and Malignant Hemopoietic Cell-Niche Interactions

Abstract

Human hematopoietic niches are complex specialized microenvironments that maintain and regulate hematopoietic stem and progenitor cells (HSPC). Thus far, most of the studies performed investigating alterations of HSPC-niche dynamic interactions are conducted in animal models. Herein, organ microengineering with microfluidics is combined to develop a human bone marrow (BM)-on-a-chip with an integrated recirculating perfusion system that consolidates a variety of important parameters such as 3D architecture, cell–cell/cell–matrix interactions, and circulation, allowing a better mimicry of in vivo conditions. The complex BM environment is deconvoluted to 4 major distinct, but integrated, tissue-engineered 3D niche constructs housed within a single, closed, recirculating microfluidic device system, and equipped with cell tracking technology. It is shown that this technology successfully enables the identification and quantification of preferential interactions—homing and retention—of circulating normal and malignant HSPC with distinct niches.

1. Introduction

Under normal conditions, hematopoietic stem/progenitor cells (HSPC) reside within specific bone marrow (BM) niches. These are comprised of different cell types located strategically to provide a myriad of signals and physical interactions that maintain HSPC and orchestrate hematopoiesis. In myeloid malignancies, the BM niche is remodeled by malignant cells, which displace HSPC, and create self-reinforcing malignant niches that drive disease progression, drug-resistance, and relapse. Studies using the mouse model have led to a fairly detailed understanding of HSPC and their dynamic, specialized microenvironment/niche.^[1] While human HSPC and their respective BM niches have been presumed to be similar to those of murine surrogates, there are few actual data to support this assumption, and these models have not always proven predictive of results in humans.^[2,3] Attaining a mechanistic understanding of the major determinants of HSPC function in the human BM has been difficult, due to challenges associated with visualizing and modeling the complex niche environment in humans. Given the important biological differences that exist between human and mouse HSPC,^[3] the development of tractable, physiologically relevant models of normal and malignant human BM niches would fill a critical gap, and is essential to fully understand the human hematopoietic system in health and disease. Such a system would also enable the development of therapies targeting deviant cells and/or signaling pathways.

In addition to murine model systems, traditional in vitro 2D cultures have also been the foundation of countless important scientific discoveries, but cells grown in 2D in traditional tissue culture dishes experience different surface topography, surface stiffness, cell–cell/cell–matrix interactions, and a 2D versus 3D architecture. As such, 2D culture systems fail to accurately recapitulate the in vivo microenvironment, and growth under 2D conditions can substantially alter the molecular and phenotypic properties of mammalian cells, producing experimental outcomes that may not be indicative of what happens in vivo.^[4] For example, we recently demonstrated that, when grown in 2D tissue culture dishes, metastatic colon carcinoma cells exhibited an epithelial morphology and expression profile, and it was only when they were transitioned into a 3D liver organoid environment that they adopted a mesenchymal and metastatic phenotype more reflective of their in vivo behavior.^[5,6] Therefore, 3D bioengineered platforms using human patient-derived cells can better mimic the cell–cell, cell-extracellular matrix (ECM), and mechanical interactions of in vivo tissue, and are thus more suitable for mechanistic research. Moreover, such platforms could one day even be deployed to improve personalized medicine approaches in the clinic.^[7] To date, only a handful of such models have been developed that recapitulate some aspects of the native bone and/or bone marrow microenvironment in vitro.^[8–11]

Herein, we combine organ microengineering^[12] with micro-fluidics,^[6,13,14] to develop a human bone marrow niche-on-a-chip (NOC) platform with an integrated recirculating perfusion system. This platform brings together a variety of important parameters that allow a better mimicry of in vivo conditions, including 3D architecture, cell–cell/cell–matrix interactions, circulation, and integration of multiple niches/tissues. In the present studies, we have deconvoluted the complexity of the BM niche to 4 major distinct bioengineered niche constructs that are housed within a single, closed, recirculating microfluidic device system, thus allowing direct manipulation and study of the impact of each individual niche on the homing and lodging/retention of circulating HSPC as well as the effects the distinct niches exert upon one another. Using this human NOC platform, we showed in real-time, and for 5 days, the preferential interactions of infused HSPC, lymphoma, and leukemic cells with periarterial (A), perisinusoidal (S), mesenchymal (M), or osteoblastic (O) niches (N). Our studies establish the feasibility of using microfluidic “on-a-chip” technology to recreate deconstructed representations of the various niches within the BM microenvironment, and they provide proof that this novel system can be used to study the interactions of normal and malignant HSPC with distinct cells of the niche.

2. Results

2.1. Scientific Design

The overall goal of these studies was to create an ex vivo, deconstructed bone marrow niche-on-a-chip platform with which to perform experiments to investigate the homing and engraftment/retention of normal and malignant human hematopoietic stem/progenitor cells to the different cellular niche constituents. To accomplish this, we sorted adult human bone marrow mononuclear cells (Figure 1a) into 3 populations: Sinusoidal endothelial cells, arterial endothelial cells, and mesenchymal stromal cells. We then differentiated a subset of MSCs into osteoblasts to create a fourth niche population (Figure 1b). Using a hyaluronic acid and gelatin-based hydrogel biofabrication technology (Figure 1c)–widely used in regenerative medicine and tissue engineering^[13,15,16]–the 4 niche cell populations were then encapsulated in this ECM-derived hydrogel, forming distinct 3D niche constructs (Figure 1d), each in its own chamber of a microfluidic device. This platform was then used to assess whether various healthy and malignant HSPC populations (Figure 1e), preferentially homed to specific niches following infusion into device (Figure 1f).

Figure 1.

Overall NOC experimental summary. a) Human bone marrow is b) separated into 3 niche populations (sinusoidal endothelial, arterial endothelial, and mesenchymal) by magnetic sorting. A subset of the MSC population is differentiated to the osteoblastic lineage. c) Using an extracellular matrix-mimicking hydrogel comprised of thiolated hyaluronic acid, thiolated gelatin, and a polyethylene glycol diacrylate (PEGDA) crosslinker, d) individual niche populations are encapsulated in 3D niche constructs inside the f) NOC microfluidic device. Homing and lodging/retention studies e) are initiated by infusing either healthy HSPC or malignant leukemia or lymphoma cells, after which homing and lodging/retention of infused cells that have traveled through circulation to each of the niche constructs is quantified in an unbiased manner.

2.2. Bone Marrow Niche Cell Characterization

Following magnetic sorting of the cell populations used to create each BM niche, and differentiation of MSCs into osteoblasts, the phenotype of each of the resulting 4 niche cell populations was characterized in 2D culture using well-established markers for each niche/cell type.^[1,2,17] Specifically, immunofluorescence verified that the CD146+NG2⁺ arterial endothelial cells expressed high levels of Ephrin B2 and were devoid of EphB4, while the CD146+NG2^lo/- sinusoidal endothelial cells expressed high levels of EphB4 but lacked expression of Ephrin B2 (Figure S1a,b, Supporting Information). The MSC niche population expressed both CD44 and Stro-1, two widely accepted markers for MSC,^[18] while the differentiated osteoblastic niche cells showed calcium deposition, as evidenced by positive staining with Alizarin Red, while MSC did not stain with Alizarin Red (Figure S1c–e, Supporting Information).^[19] The distinct expression profile of each of these niche populations validated the successful isolation of the cell types required to recreate each of the four key bone marrow niches in vitro.

2.3. Microfluidic Niche-On-A-Chip Device Fabrication and Niche Construct Biofabrication In Situ

Devices were fabricated by soft lithography of a master mold with PDMS.^[20] The inlet and outlets of the negatively casted hemicylindrical microfluidics PDMS layer were punched, followed by irreversible bonding to a glass slide; to form a sealed fluidic device (Figure 2a). Importantly, the fabrication method was designed to create hemicylindrical channels, to avoid angular features that could potentially lead to trapping of circulating cells.

Figure 2.

Microfluidic chip device fabrication and niche construct integration. a) Microfluidic device fabrication is performed by bonding a molded PDMS layer in which channels and chamber features are defined by soft lithography to a glass slide. b) In situ 3D microconstruct formation workflow: Channels i) are filled with a mixture of photocurable hydrogel precursor, cells, and additional components (dark red, ii). A photomask (gray) is employed to define construct shape and location iii), and the remaining solution washed away with fresh PBS iv). c) Side and top view depictions of the niche constructs in the device chambers. d,e) Schematic and photograph, respectively, of operational recirculating multi-niche-on-a-chip systems. Scale bar: 1 cm. f) Four frames from a fluid dynamics computational model of the NOC using Flow3D software, in which a simulation of infusing HSPC into the device was performed by infusing 1000 roughly HSPC-sized particles. Time (T) is in seconds. Heatmap represents fluid velocity (m s⁻¹).

Following device fabrication, tissue constructs were biofabricated by photopatterning-based cell encapsulation^[5] (Figure 2b) using a hyaluronic acid (HA) and gelatin-based hydrogel, commercially available as HyStem, that has been employed extensively in tissue engineering and regenerative medicine applications including 3D culture,^[15] tumor modeling,^[21] bioprinting,^[22] and biofabrication of organoids for drug and toxicity screening.^[13,21,23] We recently modified this hydrogel system to provide faster gelation kinetics,^[24] allowing for tissue construct fabrication with significantly improved control over regions of hydrogel crosslinking, while maintaining established properties of the original non-photocrosslinkable hydrogel. In contrast to many existing biomaterials, this system is comprised of naturally derived materials that are native to the body, namely, hyaluronic acid, which is present in high concentrations in the ECM of the bone marrow.^[25] Moreover, this ECM hydrogel also supports the modular addition of additional factors, such as cytokines and growth factors, if desired, via heparin-modulated binding, thus facilitating local presentation to nearby cells.^{[13,15,22,26]} These capabilities and characteristics increase this hydrogel’s biomimetic properties compared to other photocrosslinkable hydrogels such as those based on PEGDA and methacrylated gelatin. Importantly, this hydrogel system supports on-demand, near-instantaneous photo-crosslinking of discrete constructs in situ, where encapsulation of cells in 3D only occurs upon UV light exposure and within those regions that are exposed. This capability is not supported by more traditional gel materials that exhibit slow crosslinking kinetics, such as collagen Type I and Matrigel, or fast, but more difficult to control kinetics, such as alginate, making these materials less effective for biofabrication.^[27]

Tissue construct photopatterning was performed in a stepwise fashion that we have previously described.^[5] Figure 2b shows a schematic of the NOC device and the biofabrication procedure with the four patterned niche constructs, each formed in an individually addressable fluidic chamber. Using this approach, in the sealed fluidic devices, the hydrogel-based arterial, sinusoidal, MSC, and osteoblast niche constructs were formed and then maintained under circulating flow (10 μL min⁻¹). Cells were incorporated into each niche constructs at 10 million cells mL⁻¹ of hydrogel. At volumes of ≈6 μL, this would result in ≈60 000 cells per individual construct. Figure 2c shows an enlarged top view of one of the niche constructs within the NOC and a cartoon schematic conceptually illustrating the tissue constructs under flow conditions, and suspended cells traveling with the recirculating media, enabling them to pass by and/or interact with each of the niche constructs.

2.4. Platform Operation and Modeling

A 3D rendering of the fabricated NOC devices is shown in Figure 2d, and a photograph of a device under operation is shown in Figure 2e. Devices were designed with a single inlet port in the center of the PDMS part of the device. The connected inlet channel then bifurcates several times, with each of the four resulting channels ending with a circular chamber, in which the niche construct resides. One important consideration was achieving equivalent fluid flow in each of the four arms of the device, to ensure that cells infused through the inlet and recirculating within the media were evenly distributed to each of the four niche constructs. This was first visualized by infusing dyed PBS, and confirming that the fluid was reaching each chamber at approximately the same time point (Movie S1, Supporting Information). To more rigorously evaluate this important issue, we created a fluid dynamics computational model of the NOC using Flow3D software, in which a simulation of infusing cells into the device was performed by infusing 1000 roughly HSPC-sized particles. Figure 2f shows 4 images that are screenshots of the video (Movie S2, Supporting Information) of this simulation. Importantly, the simulation revealed a heatmap of the fluid flow rates through the virtual device, which indicated that fluid flow is approximately equal in each parallel region of the device. Moreover, arrows point to the location of the groups of infused particles, which are also present in approximately parallel locations within the chip at each time point. These particle numbers were then quantified (Figure S2, Supporting Information) using a custom-made MATLAB script (Code File S1 in the Supporting Information). This analysis showed roughly equivalent numbers of infused particles were present in each bifurcation (Table S1, Supporting Information). In addition, upon actual initiation of cell-based NOC device studies, the physical locations of each niche type in the devices were rotated to avoid the possibility of introducing any position-dependent bias to the observed homing and lodging/retention.

2.5. On-Chip Characterization of 3D Niche Constructs

Initially, NOC devices containing each of the 4 niche construct types were maintained under 10 μL min⁻¹ flow to assess cell viability and the stability of the platform. The flow rate at which circulating cells are transported through the device begins at 10 μL min⁻¹. This was initially chosen to prevent a buildup of pressure within the device, which could damage the device if too high. However, when analyzing the flow rates throughout the device, as the channels bifurcate and flow reaches the construct chambers, the significantly increased cross-sectional area means a significantly decreased velocity that mimics that in the bone marrow capillary bed.^[28] Devices were maintained under these conditions for 8 days, after which staining was performed directly on chip, using the antibodies to niche-specific markers and the staining procedures described above. In parallel, viability was determined by LIVE/DEAD staining. To image the fluorophores on-chip, a macro-confocal microscope was utilized to accommodate the 3D nature of the constructs. As can be seen in Figure 3, all niche types continued to express appropriate cell-specific markers within the 3D niche constructs, with arterial constructs staining for Ephrin B2 (Figure 3a), sinusoidal constructs staining for EphB4 (Figure 3c), mesenchymal constructs staining for CD44 (Figure 3d), and osteoblastic constructs staining positive for Alizarin Red, confirming calcium deposition (Figure 3b). Continued expression of these markers by each niche cell type was also confirmed on day 3, establishing the biological stability of the cells within this novel platform (Figure S3, Supporting Information). Importantly, LIVE/DEAD staining over this same time period demonstrated overwhelmingly high viability, with very few dead cells seen in any of the 4 niche construct types (Figure 3e). Interestingly, in many of the osteoblastic niche constructs, with time in culture, we observed fairly marked remodeling. Specifically, as the cells differentiated and deposited calcium, they also contracted to form a tighter, more compact 3D structure, much like native bone tissue (Figure S4, Supporting Information).

Figure 3.

Niche cell biomarkers in 3D niche constructs in NOC devices on Day 8 following construct biofabrication. a) Arterial niche constructs stained for Ephrin B2 (B2); b) Osteoblastic niche constructs stained with Alizarin Red for calcium deposits; c) Sinusoidal niche constructs stained for EphB4 (B4); and d) mesenchymal (MSC) niche constructs stained for CD44. Panels are organized as: i) DAPI, ii) indicated stain, and iii) merged image; e) LIVE/DEAD viability staining of each niche construct type at day 8. Green-calcein AM-stained viable cells; Red-ethidium homodimer-stained dead cells. Scale bars: 100 μm.

It should be noted here that 8 days for the total duration of the study (3 + 5 days after infusion of HSPC (described below), was chosen for practical reasons. Due to the requirement of closed loop circulation for the 5 days post HSPC infusion, spent media was not replaced with fresh media. After day 8 (or after 5 days in closed loop circulation), we began to see deterioration of niche construct viabilities (Figure S5, Supporting Information), which drove our time point choices.

2.6. Healthy and Malignant Human HSPC Exhibit Distinct and Selective Homing to Specific Niches on NOC Devices

For all homing studies, the locations of the 4 distinct niche constructs were randomized to account for any potential bias of circulating cells based on the device geometry. Three days after establishing the NOC devices, normal healthy adult human bone marrow-derived CD34+ cells, human lymphoma cells (U937), or human acute myelogenous leukemia (AML) cells (MOLM13) were fluorescently labeled with DiI fluorescent membrane dye (red), and infused into the device via the input port, and the cells were allowed to circulate for 5 days as the microperistaltic pump recirculated the media. DiI-labeling enabled facile tracking of the cells within the NOC devices via direct visualization of red fluorescence using an on-chip micro-camera system we adapted from our previous organ-on-a-chip studies^[6,21] to be compatible with the NOC platform (Figure 4a). These camera systems consisted of an LED light source and a lens that would sit on one side of the NOC, paired with a filter and the camera on the other side of the NOC (Figure 4b). This platform could be employed to not only take individual snapshots of the constructs, but also to capture videos in which one could observe, in real-time, labeled cells circulating through the system and: 1) lodging within specific constructs (Figure 4c); 2) passing by the constructs without directly interacting (Figure 4di–iii); or 3) transiently attaching to the construct, only to then be released and re-enter the circulation (Figure 4div–vi). Movie S3 (Supporting Information) is a representative example of one such video that was captured.

Figure 4.

Tracking of fluorescently labeled normal/healthy, leukemic, or lymphoma HSPC using an onboard fluorescent camera system to visualize HSPC homing, lodging/retention, or passing by the 3D niche constructs. a) Working NOC device that is then monitored using b) a custom-built fluorescent camera system comprised of the NOC sandwiched between an LED, lens, filter, and camera, shown in operation in the right panel. This system allows (c,d) real-time visualization of infused labeled cells in the NOC system. c) A U937 cell is indicated (white arrows) moving toward an arterial niche construct i–iv) and remaining in place after contact v–vi). d) A different U937 cell is indicated (white arrows) that passes around the arterial construct i–iii), never making contact, while in panel iv), a second U937 cell is observed detaching from the construct and re-entering “circulation” v–vi). Both sequences occur in ≈5–10 s time.

The most informative approach to quantitate the degree of homing (early migration/attachment to niche) and lodging/retention (later, stable colonization of niche) of each HSPC population within each of the niche constructs proved to be capturing images both in the fluorescent channel and by light microscopy, and then superimposing these images to precisely visualize the localization of the labeled HSPC within the niche constructs at any given point in time. An in-depth example of the comprehensive data that can be obtained with this approach is shown in Figure 5, which provides a detailed overview of the results of the homing studies. To obtain precise quantitation of labeled HSPC in each niche construct from these superimposed images, we also developed a novel methodology utilizing ImageJ software, which is described in Figure S6 in the Supporting Information. In brief, following biofabrication of each niche construct, a digital mask was created that corresponded to the circumference of that particular construct, thereby defining the region of the construct in which lodged/retained HSPC could be detected (Figure S6a, Supporting Information). Next, the fluorescent and light microscopy superimposed image was created (Figure S6b,c, Supporting Information) at each subsequent time point during the study. The digital mask from the initial timepoint was then applied to these overlays, after which the cells outside the mask were erased, and the red cells were transformed to white. Last, black/white thresholding was employed to turn the white regions into black, and a script was used to break large features into cell-sized objects for quantification (Figure S6d, Supporting Information). As is detailed in the following section, these analyses revealed that each of the 3 populations of HSPC infused–healthy CD34+ cells, lymphoma cells (U937), and leukemia cells (MOLM13)–exhibited a marked preference for homing to and lodging within particular niche constructs.

Figure 5.

Quantification of initial homing of HSPC to niches after 24 h. Cell counts at each niche for: a) normal/healthy CD34+ HSPC, b) U937 lymphoma cells, and c) MOLM13 leukemia cells.

After each HSPC population had been infused and allowed to recirculate within the NOC device for 24 h, the first studies to quantify the co-localization of labeled HSPC with each of the 4 niche constructs (homing) were performed. Figure 5 depicts these initial homing trends. CD34+ HSPC homed with approximately equivalent proclivity to MSC, sinusoidal, and osteoblastic niches, with very few cells homing to the arterial niche (Figure 5a). This is in contrast to the U937 lymphoma cells which exhibited an increased trend in homing to the arterial and osteoblastic niches (Figure 5b), while the MOLM13 leukemia cells tended to home at higher numbers to the osteoblastic niche (Figure 5c). It should be noted that, at this early timepoint, these data exhibited trends, but did not achieve statistical significance. However, as the studies progressed over time and subsequent lodging/colonization of the niches took place, increased numbers of normal/healthy CD34+ HSPC were detected in the osteoblastic niche, while U937 and MOLM lodging/retention preferences paralleled that of early homing, and these differences achieved statistical significance.

Evaluation of lodging/retention at day 5 showed that CD34+ HSPC lodged, to some degree, within the mesenchymal and sinusoidal niche constructs (Figure 6a,b), and while not statistically significant, there was a trend to preferentially lodge in the osteoblastic niche constructs (Figure 6c). Assessment over time showed that very few CD34+ HSCs homed to the arterial niche constructs, and even though small numbers of HSPC appeared to initially home within this niche, they were subsequently released (Figure 6d). Relative quantification of the lodging/retention of the normal/healthy HSPC within each niche construct at day 5 is summarized in Figure 6e. U937 lymphoma cells lodged within the osteoblastic niche constructs (Figure 6h) and to some degree within the mesenchymal and sinusoidal niche constructs (Figure 6f,g). However, in marked contrast to the normal HSPC, the lymphoma cells lodged/colonized the arterial niche constructs with the highest frequency (Figure 6i). Relative quantification of the lodging/retention of the U937 cells within each niche construct at day 5 is summarized in Figure 6j. The last HSPC population examined, the leukemic (MOLM13) cells lodged with greatest frequency within the osteoblastic niche (Figure 6m), followed closely by the arterial niche (Figure 6n), and lodged at lower levels within the mesenchymal and sinusoidal niche constructs (Figure 6k,l). A summary of the relative quantification of lodging/retention/colonization of each niche type by the MOLM13 cells at day 5 appears in Figure 6o.

Figure 6.

Niche-on-a-chip lodging/retention experiments using CD34+ cells from normal/healthy adult donors (CD34+ HSPC), CD34- cell line derived from a lymphoma patient (U937), CD34+ cells derived from an AML patient (MOLM13), which show distinct niche lodging/retention preferences between cell types. a–e) Normal/healthy CD34+ HSPC located preferentially within the ON, exhibited moderate lodging/retention within the MN and SN, and were only rarely found in the AN. f–j) The CD34- lymphoma cell line (U937) exhibited a marked predilection for the AN, followed by the ON. k–o) CD34+ cells derived from an acute monocytic leukemia patient (MOLM13) engrafted/lodged primarily in the ON and AN, with some lodging/retention in the MN and SN. a–n) show representative lodging/retention images with fluorescent images in the red channel highlighting the infused cell types overlaid on light microscopy images of each niche. Images were taken on day 1, day 3, and day 5 following infusion of the cells into each NOC device. e,j,o) provide quantified average cell numbers of lodged/retained CD34+ HSPC, U937 lymphoma cells, and MOLM13 leukemia cells, respectively, in each niche. Red highlighted borders indicate regions of common lodging/retention. One-way ANOVA (n = 10) **P < 0.005,***P < 0.001. Scale bar: 250 μm.

3. Discussion

In the present studies, we developed a novel human bone marrow niche-on-a-chip platform to model the in vivo complexity of the native human BM. Thus, we disentangled the BM microenvironment by breaking it down into 3D constructs representing the 4 major hematopoietic niches (N) that exist within the BM, namely, the periarterial (A), perisinusoidal (S), mesenchymal (M), and osteoblastic (O). Importantly, these 4 distinct NOCs are housed within a single, closed, recirculating microfluidic device, in which microchannels allow the continuous physiologic flow of human HSPC through each of the specific niches, reproducing, in effect, a primitive circulatory system. These NOCs contain 3D tissue constructs, sometimes referred to as organoids, fabricated at scales (200–400 μm) that are well below the diffusion limit for nutrients, oxygen, and small molecules,^[29] allowing preservation of high viability for at least 8 days, and supporting the continued expression of appropriate phenotypic/functional markers by each distinct NOC throughout this time.

Despite the significant advancement and increased attention to organ-on-a-chip research in recent years, there have only been a small number of published examples of attempts to produce an on-chip system that captures the biology of the bone marrow.^[8–11] Importantly, these differ significantly from our platform and its intended objectives. For example, two recent studies have employed bone-based tissue chips to study cancer cell migration from vasculature into bone tissue.^[8] In another study, Torisawa et al., engineered a bone marrow-on-a-chip device by implanting a scaffold into a mouse and allowing it to be populated by murine cells. This construct was then removed and placed in a microfluidic device that was then shown to support the maintenance of primitive hematopoietic cells.^[11] Similarly, a study by Siebert and colleagues^[9] described a 3D microfluidic system to model the bone marrow, but the biology of the system was significantly simplified by only including a single niche cell type (MSC) to test the ability of the device to support primitive HSPC. These are all important studies that add to the field of microphysiological systems in the context of hematopoiesis and the bone marrow niche, yet they all employed vastly different bioengineering approaches and explored completely different biological questions than our study.

Throughout the duration of our NOC studies, we showed that this new system has the power and resolution to make possible the direct visualization, tracking, and quantitation, in real-time, of the interactions that occur between fluorescently-labeled normal human BM-derived HSPC, leukemic cells (MOLM13), and lymphoma cells (U937), and each of the 4 distinct NOCs. The microfluidic devices were designed and tested to provide an equal probability for interaction of these three human hematopoietic cell populations with the four NOC constructs, yet each population exhibited a marked predilection for homing to and subsequently lodging within specific NOC constructs. Importantly, these preferences differed between normal/healthy and malignant human HSPC, and also differed depending upon the type of malignancy, such that HSPC lines from leukemia and lymphoma patients exhibited distinct patterns of NOC homing and lodging/retention.

In addition, by employing optically-clear devices and an onboard camera, we also showed the ability of this system to discriminate distinct phases of the interactive process between the various niches and the infused human HSPC, as we were able to clearly visualize the early stages of homing in which the flowing HSPC tethered and rolled along the specific niches, as well as their subsequent establishment of firm adhesive interactions with the components of the niche to permanently lodge for the duration of the 5-day study. We also observed brief interactions in which the flowing HSPC transiently tethered and rolled along a niche, only to then detach and re-enter the “circulation,” and ultimately lodge within/colonize a different niche. These observations were further supported by quantitation at day 1 following cell infusion where cell–niche interactions could be considered “homing”-based, versus quantitation at day 5, by which time, cells could be considered to have truly lodged within the niches. In some cases (CD34+ HSPC), early homing involved predilection to interact with both mesenchymal and osteoblastic niches, but over time shifted to preferential lodging/retention in the osteoblastic niches. In contrast, MOLM13 leukemia cells and U937 lymphoma cells had similar homing (early, day 1) and lodging/retention (late, day 5) niche preferences, with these preferences becoming even more pronounced at later times.

A critical aspect to validate the biological relevance of the NOC device/system described herein is its parity to the in vivo setting. Unfortunately, in vivo data of this type are not available for the human system, making it impossible for us to compare the homing/lodgment patterns we observed on the NOC devices to what occurs in human patients receiving healthy HSC transplants, or human patients with leukemias/lymphoma. However, multiple elegant experiments have been done in murine model systems in which healthy or leukemic human cells/cell lines have been infused and imaged using high resolution MRI and both real-time and “temporal snapshot” confocal microscopy to track the homing of these infused human hematopoietic cells.^[30,31] These studies have collectively shown that: 1) following infusion into mice, normal/healthy human HSC home to both the perivascular (sinusoidal endothelium and/or MSC/pericytes) and osteoblastic/endosteal niches/regions; 2) HSC do not home to nor engraft the arterial niches to any significant degree following infusion; and 3) acute myelogenous leukemia cells (the same type of leukemia as the MOLT-13 cells we employed herein as a model for leukemia) home almost exclusively to the endosteal (osteoblastic) niches/zones following infusion into mice. Since these patterns parallel those that we observed with our novel NOC platform, we are hopeful that the results we have obtained with this new in vitro system are likely to be predictive of aspects of the in vivo setting.

An additional consideration that bears mention is the utility of this 3D in vitro system versus traditional 2D culture approaches. One could argue that, in a more straightforward experiment, a 2D control might be an appropriate control and a desired component of the study. However, we have published extensively in a variety of tissue types and pathologies (liver, cardiac, pancreas, cancer, etc.) showing that 3D in vitro models functionally mimic in vivo tissue more accurately than their 2D counterparts.^{[6,13,15,23,32]} Since the marrow microenvironment in vivo exists in three dimensions, like all other tissues, we do not feel that including a 2D analog of our 3D system in our experiments would be of significant value nor add to the validity of the data being presented.

We thus feel that the data provided herein provide compelling evidence of not only the feasibility, but also the validity, of using microfluidics “niche-on-a-chip” technology as a tool to study the interactions of distinctive HSPC with different niches, and ultimately to dissect, in real-time, the molecular mechanism(s) underlying the fate-dependent niche interactions of normal and malignant human HSPC subpopulations, and to delineate the HSPC-niche signaling pathways by which the different niches regulate human HSPC function, under conditions of both health and disease.

Looking specifically at myeloid malignancies, the BM niche is known to be profoundly remodeled by malignant cells, which displace resident HSPC, and create self-reinforcing malignant niches that drive disease progression and chemoresistance/relapse.^[31,33] Therefore, the NOC platform we describe herein can serve as a valuable tool for preclinical testing of drug efficacy/toxicity, and, since all cells in the NOC are human, should more accurately predict clinical response. We are currently using this system to delineate the signaling pathways responsible for the observed preferential HSPC/niche cell interactions, with the ultimate goal of using this knowledge to develop more effective treatments for hematological malignancies and enhance engraftment following HSPC transplant. Specifically, defining the differences within normal versus malignant niches that alter the interactions with HSPC could lead to novel therapies targeting the deviant cells and/or signaling in the malignant niche; a significant advance over current drugs to disease-specific driver mutations. Furthermore, repopulating the NOCs with patient-derived cells could ultimately pave the way for safer and more effective personalized medicine approaches in the clinic.

4. Experimental Section

Study Design:

The objective of these studies was to create an ex vivo, deconstructed bone marrow niche-on-a-chip platform with which to perform experiments to investigate the homing and engraftment/retention of normal and malignant human hematopoietic stem/progenitor cells to the different cellular niche constituents. All cells employed in these studies were purchased commercially. All data presented is based on studies in triplicate or higher.

Cell Sorting and Differentiation:

Frozen human bone marrow mononuclear cells (BMNC) were obtained from Stem Cell Technologies (Lot # 1507160920, harvested from a 28 year old male Caucasian, 77 kg, 170 cm, nonsmoker). Thawed BMNC were divided into two aliquots for Stro-1 and CD146⁺ isolation using a MiniMACS Separator kit (Miltenyi Biotec, Auburn, CA) according to manufacturer’s instructions. Briefly, one aliquot of BMNC were labeled with IgM anti-human Stro-1 and incubated with anti-rat IgM microbeads. The other fraction of BMNC was labeled with a phycoerythrin (PE)-conjugated antibody to CD146 (Becton Dickinson ImmunoSystems, San Jose, CA), followed by incubation with anti-PE Multisort microbeads (Miltenyi Biotec), as previously described.^[18] The Stro-1+ and CD146+ populations were then obtained by magnetic sorting using an MS magnetic column (Miltenyi Biotec), and the Stro-1+ mesenchymal stromal cells (MSC) were directly cultured in mesenchymal cell growth medium (MSCGM; Lonza, Walkersville, MD) in fibronectin-coated flasks.

The CD146⁺ fraction of BMNC was then further sorted into NG2 positive (arterial endothelial cells; AEC) or negative (sinusoidal endothelial cells; SEC) populations by incubating the cells with an allophycocyanin (APC)-conjugated anti-NG2 antibody (Novus Biologicals, Littleton, CO), followed by anti-APC microbeads and MS magnetic column sorting. The CD146+NG2⁺ and CD146+NG2⁻ cells were each cultured in MSCGM (Lonza) in fibronectin-coated flasks.

To form the osteoblastic niche, an aliquot of the Stro-1+ MSC were induced to undergo osteogenic differentiation using the StemPro Osteogenesis Differentiation Kit, in accordance with the manufacturer’s directions (Life Technologies, Grand Island, NY). In brief, at 50% confluence, MSCGM media was removed from the MSC and exchanged for freshly prepared osteocyte/chondrocyte differentiation basal media with supplements. Partially differentiated osteoblasts were then harvested at day 3–4 and used for chip integration.

3D Construct Biofabrication:

The stromal constructs were fabricated “off-chip” by encapsulating the Arterial, Sinusoidal, Osteoblast, or mesenchymal (MSC) niche cells in thiol-modified hyaluronic acid-based hydrogel (ESI-BIO, Alameda, USA), in separate wells. ECM-based HA/gelatin hydrogels were formed using HyStem-HP (ESI-BIO, Alameda, CA). The thiolated HA component (Heprasil) and the thiolated gelatin component (Gelin-S) were dissolved in water containing 0.5% w/v of the photoinitiator 4-(2-hydroxyethoxy)phenyl-(2-propyl)ketone (Sigma St. Louis, MO) to make 1% w/v solutions. The polyethylene glycol diacrylate crosslinker (Extralink, ESI-BIO) was dissolved in the photoinitiator solution to make a 2% w/v solution. Heprasil, Gelin-S, and Extralink were then mixed in a 2:2:1 ratio by volume. The resulting solution used to resuspend cells at a cell density of 10 million cells mL-1. A 6 μL volume of the hydrogel precursor/cell mixture was placed on top of a 6 mm PDMS puck and exposed to UV light for 0.25 s (365 nm, 7 W cm⁻²) to crosslink, followed by washing with PBS, and addition of serum-free QBSF-60 Stem Cell Medium (Quality Biological, Gaithersburg, MD). At volumes of ≈6 μL, this would result in ≈60 000 cells per individual niche construct.

Microfluidic Platform Fabrication:

The microfluidic platform was built using PDMS (10:1 Sylgard 184 silicone elastomer and curing agent respectively, Dow Corning) and a glass slide (VWR, Radnor, PA); both treated and irreversibly bounded with O₂ plasma. A convex hemicylindrical microfluidic channel (120 μm height) was fabricated using a modified previously reported standard photolithography technique. Briefly, a primary SU-8 2050 (MicroChem, Westborough, MA) wafer was generated by standard lithography. A primary PDMS device was generated by molding the square channel then the channels were sealed with a 40 μm PDMS membrane. The membrane in the channels was deformed into the channels by a vacuum negative pressure system, SU-8 was placed into them and also spin-coated evenly on a glass slide. Both SU-8 covered surfaces were bind and UV-cured from the top. A final PDMS channel mold was generated for the microfluidic layer.

Individual Niche Construct Biofabrication On-Chip:

The in situ patterned stromal niche constructs were based on a modified reported technique. Chambers were filled with Arterial, Sinusoidal, mesenchymal (MSC), or Osteoblast cells encapsulated in thiol-modified hyaluronic acid based hydrogel (ESI-BIO, Alameda, USA), in their respective chamber. A photomask with four equidistant 500 μm diameter circles was aligned in the center of the chambers and exposed to UV light for 0.25 s (7 W cm⁻²) to crosslink, washed with PBS to remove un-crosslinked solution, and placed in serum-free QBSF-60 Stem Cell Medium (Quality Biological, Gaithersburg, MD) (Figure 2b).

Assessment and Characterization of Niches:

To confirm that the individual niches exhibited the appropriate phenotype, both when cultured in 2D and within the 3D constructs, immunofluorescent staining was performed at days 3, 7, and 14. Briefly, cells or tissue constructs were fixed in 4% paraformaldehyde at room temperature and maintained at 4° in PBS until processed. Cells were cultured and stained on coverslips, and off-chip constructs were processed by standard paraffin-embedding; 5 μm sections were cut using a microtome (Leica Microsystems Inc., Buffalo Grove, IL). Niche construct sections were deparaffinized and rehydrated with decreasing concentrations of ethanol. All samples were permeabilized in 0.1% Triton X100 for 5 min. Heat-induced epitope retrieval was performed using an automated Dako PT Link system (Agilent, Santa Clara, CA) with sodium citrate buffer, pH 6.0. Nonspecific binding was blocked using Protein Block (Abcam, Cambridge, MA) for 1 h at room temperature. Slides were incubated overnight at 4 °C in a humidified chamber with the following primary antibodies, diluted in Antibody Diluent Reagent Solution (Life Technologies; Thermo Fisher Scientific): rabbit anti-human Ephrin B2 (NBP1–84830; Novus Biologicals, Littleton, CO), mouse anti-human EphB4 (37–1800; Thermo Fisher Scientific, Waltham, MA), sheep antimouse/rat/porcine/equine CD44 (AF6127; R&D Systems, Minneapolis, MN), and mouse anti-human Stro-1 (MAB1038; R&D Systems, Minneapolis, MN). Following washing, sections were incubated with Alexa Fluor 488- or 594- or 647- conjugated secondary antibodies (Life Technologies; Thermo Fisher Scientific) for 1 h at room temperature in a humidified chamber. Following washing, samples were counterstained with 4',6-diamidino-2-phenylindole (DAPI; Life Technologies; Thermo Fisher Scientific) for 5 min, washed, and cover-slipped with ProLong Gold anti-fade mounting medium (Invitrogen; Thermo Fisher Scientific). Controls included identical slides stained in parallel, in which the primary antibody was either absent or was replaced by a nonspecific isotype-matched primary antibody conjugated to the respective Alexa-Fluor.

To determine the presence of calcium deposition, a group of slides were stained with 40 × 10⁻³ M Alizarin Red S (ARS; Sigma-Aldrich, St. Louis, MO), pH 4.1 for 5 min. Full osteoblastic commitment of the differentiated MSC was assessed by treatment with tetracycline, observing the resultant fluorescence staining of hydroxyapatite. In brief, 40 μgmL⁻¹ of tetracycline (Sigma-Aldrich, St. Louis, MO) was prepared in MSCGM culture medium or StemPro Osteogenesis Differentiation Kit medium, and the cells/constructs to be tested were then incubated with this solution for 72h, after which they were scored/imaged for fluorescence.

All light and fluorescence microscopy images were captured using an Olympus BX63 microscope (Olympus, Center Valley, PA) equipped with an X-Cite 120 LED Boost fluorescence lamp (Excelitas Technologies, Waltham, MA) and an Olympus DP80 dual camera (Olympus).

Fluidic Dynamics and Cellular Infusion:

To ensure homogeneous distribution of the infused cells into the four downstream chambers, transient 3D simulations were performed using computational fluid dynamics software, (FLOW3D v11.2, Flow Sciences, Santa Fe, NM). Mass-density particle species were assigned with random rate of generation with fully coupled particle-fluid interaction. The inlet source fluid flow rate was specified with four outlet sinks to simulate a continuous loop system. Particle counts and lifespan were tracked using measuring flux planes and sample volumes within the software for quantification and visual analysis.

After fabrication, niche constructs were allowed to equilibrate/stabilize for three days, after which, a closed loop system was set up in a four-channel precision micro peristaltic pump (Elemental Scientific, Omaha, NE). Previous to infusion, U937, MOLM13 (both from ATCC, Manassas, VA), and healthy CD34⁺ cells (StemCell Technologies, Vancouver, BC, Canada) were independently labeled with the Qtracker 605 Cell Labeling Kit (Life Technologies, Carlsbad, CA) following the manufacturer’s instructions. In short, a 10 × 10⁻⁹ M labeling solution was prepared from component A and B. The solution was diluted in 0.2 mL of fresh complete growth medium. The cells were incubated in the solution for 45 min and washed twice with media; 1 × 10⁶ cells mL⁻¹ in QBSF-60 were put into a fresh reservoir and bundled into the system. On the day of infusion, a close loop system was set up, and the labeled U937, MOLM13, and healthy CD34⁺ cells were independently perfused into the NOC device at a flow rate of 9.5 μL min⁻¹ (Figure 2b). Upon infusion, cells enter a main channel, and then evenly disperse into the four equidistant chambers, each of which houses a specific stromal niche (Figure 2b,d).

Unbiased Quantification of Engrafted Cells:

Brightfield (BF)/Qtracker stacked images in TIFF format were captured and processed in ImageJ in the following manner: 1) Using oval or elliptical tools, the stromal construct area was selected. 2) Using Image ¦ Duplicate, a new image was generated, and the surrounding area of the construct was cleared (Clear outside). This generated “mask” was utilized on each remaining correspondent stromal construct. 3) Using Image ¦ Color ¦ Split Channels, a new image with the red labeled signals was separated from the composite BF/Qtracker image. 4) Using Image ¦ Adjust ¦ Threshold, the red signal’s borders were defined and colored black with a white background. 5) Using Analyze ¦ Analyze Particles, a constant size inclusion filter (25–225 μm²) with a circularity of 0.0–1.0 was applied to ensure that debris, noise, and bright stromal cells were excluded.

Statistical Analysis:

Statistical analyses were performed using statistical analysis software (GraphPad Prism 7, GraphPad Software Inc., USA). One-way ANOVA was employed for multiple comparisons. P < 0.01 or less was considered statistically significant. Data are presented as means plus/minus standard deviation, and all experiments were performed with n ≥ 4.