The bone marrow provides the regulatory microenvironment or niche for the proliferation and survival of HSPCs [1
], which give rise to all blood and immune cells and repopulate bone marrow following stem cell transplantation. HPCs traffic between the blood circulation and the bone marrow, moving on and off niche sites [1
]. Osteoblasts, which reside in the bone marrow, are key participants in providing cues for HPC trafficking, proliferation and survival through the secretion of cell-signaling molecules, including SDF-1 [3
]. Osteoblasts also make intimate physical contact with HPCs, modulating their function in response to specific physiological conditions [4
]. While contact-dependent communication between HPCs and osteoblasts is critical for establishment and maintenance of progenitor cell proliferation and self-renewal [5
], the molecular pathways that govern this interaction are largely unknown. Moreover, the downstream events occurring at the HPC/osteoblast contact site remain uncharacterized, despite their major role in signaling and remodeling within the niche.
To investigate the morphological and functional characteristics of the HPC-osteoblast interface, we imaged a live-cell, co-culture system consisting of either primary human CD34+ cells or cells from the human KG1a progenitor cell line, co-cultured with the human SaOS2 osteoblastic cell line or primary human osteoblasts. HPCs (i.e., CD34+ and KG1a cells) displayed a rapidly changing morphology and were motile, often with a leading and lagging edge or uropod as previously described () (Supp. Mov. 1
]. Following movement toward and between osteoblasts, individual HPCs were observed making contact and adhering to osteoblasts ()(Supp. Mov. 2
). Given this HPC behavior fit previous descriptions, we began detailed examination of the cell contact interface and its dynamics.
Figure 1 Organization and maintenance of the HPC plasma membrane. HPC dynamics are revealed by time lapse confocal microscopy of (a) CD34+ cells co-cultured with osteoblasts stably transfected with GalT-YFP (green) or (b) KG1a cells labeled with PKH26 (red) and (more ...)
The surface distribution of different HPC plasma membrane components was investigated since this might be important for HPC interactions with osteoblasts. The very late antigen-4 (VLA-4) protein, which interacts with VCAM-1 [8
], was highly enriched within a specialized membrane domain of HPCs () regardless of contact with osteoblasts. A similar asymmetric distribution was seen for CD63 () and CD81 (), which are tetraspanins implicated in integrin regulation [11
]. The stem and progenitor cell marker, prominin 1, which localizes to highly curved membranes [13
], also showed the polarized pattern (), as did the fluorescent phosphatidyl ethanolamine analogue, N-Rh-PE (1,2-dipalmitoyl-sn-glycero-3-phophoethanolamine-N-[lissamine rhodamine B sulfonyl]) (), a cone-shaped lipid that inserts within the plasma membrane [15
]. N-Rh-PE labeling of a more enriched population of HSCs, human CD34+CD38- cells, showed a polarized distribution in 97% of the cells (n = 115 cells). Quantum dot (QDs) labeling of HPCs also showed clustering within the membrane domain (). This presumably occurred by QD clustering along the plasma membrane or potentially by a capping event, rather than by internalized delivery, since QDs could be removed by proteinase K digestion (not shown). The polarized localization of these markers was specific because other surface molecules, CD34 () and CD45 (), were distributed throughout the plasma membrane. Therefore, HPCs are highly polarized and contain a specialized membrane domain with specific protein and lipid components.
To better understand HPC surface polarity and ultimately the importance of asymmetry for events occurring at the HPC-osteoblast interface, cholesterol depletion was performed using the cholesterol-sequestering agent, methyl-β-cyclodextrin. After methyl-β-cyclodextrin treatment, molecules normally enriched in the specialized domain, including prominin1-GFP, CD63-Cherry and N-Rh-PE (), redistributed into punctate clusters across the plasma membrane (). Actin-destabilization using cytochalasin D treatment also disrupted the domain organization, resulting in prominin 1-GFP, CD63-Cherry and N-Rh-PE dispersing throughout the plasma membrane (). Hence, cholesterol and actin-based processes are important for the polarized domain localization of HPC cell surface components.
Upon contact with an osteoblast (, 0h), an HPC (red) often remained attached to the same osteoblast (green) for more than 5 h ()(Supp. Mov. 2
). Confocal Z series imaging revealed that the HPC contact site was often mediated by the uropod, which resided in close association with the osteoblast plasma membrane (, XZ images) [9
]. Interestingly, the specialized membrane domain components, including prominin1-GFP (), N-Rh-PE (), VLA-4 (), and CD63-Cherry () were all highly enriched at the contact site or uropod, in contrast to CD45 (), which remained dispersed over the HPC surface. In QD labeled HPCs, the QDs also clustered at the contact site (). Time lapse imaging of N-Rh-PE labeled CD34+CD38- cells co-cultured with primary human osteoblasts further established the polarized N-Rh-PE domain as the site of cell-cell contact (Supp. Mov. 3
Figure 2 HPC/osteoblast contact occurs through a specialized HPC membrane domain. (a) Time-lapse microscopy of PKH26 labeled KG1a cells (red) co-cultured with osteoblasts transiently transfected with PH-Akt-GFP (green) reveals long-term adhesion of HPCs to osteoblasts. (more ...)
Scanning electron microscopy (SEM) of the contact interface revealed the presence of microvilli-like projections on the HPC membrane similar to those previously described () [16
]. The contact zone of the HPC seen in the SEM images comprised a region several microns in diameter, with the microvilli in this zone appearing to almost penetrate the osteoblast at the contact site (, see arrows). Nanotube-like structures projecting from many HPCs were also observed making contact with osteoblasts, but were distinct from the contact interface (). Thus, the HPC-osteoblast contact site is an elaborate, often microvilli-rich, region containing stem cell markers (i.e., prominin1) and adhesion-signaling molecules (i.e., VLA4, CD63 and CD81).
To study the contact site membrane dynamics, we imaged QD-labeled HPCs in contact with osteoblastic cells. Strikingly, QDs transferred from the HPC uropod into the cytoplasm of tubulin-YFP expressing osteoblasts () (Supp. Mov. 4
). The cytoplasmic localization of the transferred QDs was confirmed by Z series confocal imaging through the osteoblast, which showed that the QDs were in the same plane as the intercellular marker, tubulin-YFP, and not detected on the osteoblast surface (, zoom). This suggested that intercellular transfer was occurring at the contact site. Imaging of prominin 1-GFP-labeled HPCs in contact with osteoblasts revealed punctate structures enriched in the marker being transferred to the osteoblast over a 20 min contact period (). Unlike QDs, which reside peripherally on the surface of HPCs and so might be transferred to osteoblasts by a sticking and release process, prominin-1-GFP is embedded in the HPC bilayer and so can only be transferred along with HPC membrane. Co-culturing of CD63-cherry-labeled HPCs with tubulin-YFP osteoblasts also revealed uptake of CD63-cherry into the osteoblast () (Supp. Mov. 5
). Because both prominin 1-GFP and CD63-cherry proteins were engineered with the fluorescent proteins fused to the carboxy-terminus, which is located in the cytosol, their detection in the osteoblast cells after transfer from HPCs suggested that the entire protein was transferred, not a cleaved fragment. Intercellular transfer of N-Rh-PE into osteoblasts also occurred when N-Rh-PE-labeled HPCs contacted osteoblasts (). This indicated that both lipid and protein components from the HPC uropod were transferred to osteoblasts.
Figure 3 Intercellular transfer occurs between HPC and osteoblastic cells in a contact dependent manner. (a) KG1a cell labeled with QDs (red) and co-cultured with osteoblastic cells stably transfected with tubulin-YFP (green) for 2 h before live-cell imaging. (more ...)
Efficient intercellular transfer did not occur when HPCs were co-cultured with a non-niche cell type. For example, when N-Rh-PE labeled HPCs were co-cultured with HeLa cells for up to 3 h, only ~20% of the HeLa cells showed evidence of transfer (). By contrast, greater than 80% of osteoblastic cells contacting HPCs became labeled within 3 h of co-culturing (). Since no transferred HPC molecules were observed in neighboring osteoblasts not contacting a labeled HPC during 1 h of co-culture (), efficient molecular transfer only occurred between HPCs and osteoblasts that made direct cell contact. In addition, co-culturing of HPCs and osteoblasts separated by a 0.4 μm membrane filter revealed no molecular transfer from HPCs to osteoblasts during 3 h of co-culture (). This argued against transfer by bulk cellular release of exosomes, which are small membrane vesicles ~50-100 nm in diameter [17
]. Instead, the observed transfer appeared to be mediated by specific events occurring at the HPC-osteoblast contact site.
To investigate the importance of the specialized membrane domain or uropod in intercellular transfer, this region was disrupted using methyl-β-cyclodextrin or cytochalasin D treatment, as described before. Transfer between a contacting HPC and osteoblast was assessed by live cell imaging of N-Rh-PE-labeled HPCs pretreated with these reagents and then co-cultured with osteoblasts for 1 h. A significantly reduced level of intercellular transfer between contacting cells was observed () indicating that efficient intercellular transfer requires contact of the specialized HPC uropod domain with osteoblasts.
HPC uropod domain components were transferred into punctate compartments inside the osteoblast without being noticeably diluted (see for prominin 1-GFP). This suggested that intercellular transfer did not involve fusion of the HPC uropod with the osteoblast plasma membrane, as transferred components would disperse into the plasma membrane. Instead, transfer appeared to occur by a cytophagocytosis event, in which portions of the HPC uropod domain were engulfed by the osteoblast. This was most convincingly shown in co-cultures of primary CD34+ cells and osteoblasts. There, an entire domain enriched in N-Rh-PE of the primary CD34+ cell could be seen being transferred to the osteoblast over a short transfer period leaving the cell depleted of N-Rh-PE () (Supp. Mov. 6
). The N-Rh-PE domain transfer was also detected from the more primitive CD34+CD38- cells to primary human osteoblasts (). Addition of dynasore, an inhibitor of the endocytic effector protein, dynamin [18
], to the co-culture system significantly reduced the amount of uptake of HPC components at the contact site (), suggesting the intercellular transfer mechanism was dynamin dependent.
We next followed the fate of transferred material taken up by the osteoblast. After 30 min of uptake, HPC components were seen in various osteoblast endocytic compartments, including Rab5- and Rab7-positive structures (). In addition, HPC components were observed in endocytic structures positive for the 2xFYVE domain [20
] (). Imaging of single osteoblasts that had taken up prominin 1-GFP from a contacting HPC () showed that the transferred molecules remained in distinct intracellular structures within the osteoblast for up to 12 h post-transfer. In contrast, endocytosed EGF and the soluble amyloid beta protein, which are delivered to lysosomes for degradation [21
], were completely destroyed by the osteoblast during a similar time period (). These results suggested that while transferred molecules could be degraded, a proportion of these molecules avoided destruction within osteoblast lysosomes and were delivered through endocytic intermediates to a longer-lived intracellular compartment.
Figure 4 Transferred molecules are detected within various endocytic compartments of the osteoblasts. Live cell microscopy of N-Rh-PE (red) labeled KG1a cells co-cultured with osteoblastic cells transiently transfected with (a) Rab 5-GFP (green) or (b) Rab 7-GFP (more ...)
To determine the significance of transferred molecule localization for cell communication between HPC and osteoblast, we further characterized the endocytic structures containing transferred molecules. In particular, we focused on the endocytic structures that were positive for the 2xFYVE domain-GFP, since many FYVE domain-containing proteins are involved in signal transduction [22
]. SARA (Smad Anchor for Receptor Activation) is a FYVE domain-containing protein localized to endosomes, which correspond to a signaling compartment specialized for the propagation of extracellular signals such as TGFβ signaling [23
]. Upon TGFβ receptor activation, Smads are translocated to the nucleus with the help of the cofactor, SARA, resulting in gene activation. Osteoblasts fixed following N-Rh-PE transfer and stained using SARA-specific antibodies displayed a significant co-localization of N-Rh-PE with SARA-labeled endosomes (). The histogram of the weighted co-localization coefficients () revealed a similar level of N-Rh-PE co-localization with SARA as with Rab 7, although SARA and Rab 7 appeared to represent two different populations of endosomes. These data showed that molecules transferred from the HPC membrane domain can be delivered to a SARA-positive signaling endosome within osteoblasts.
Figure 5 Intercellular transfer correlates with decreased Smad 2/3 activation and an increased production of SDF-1 by the osteoblast. (a) N-Rh-PE (red) labeled KG1a cells were co-cultured for one hour with osteoblasts, which were then fixed and immunostained for (more ...)
SARA endosomes are known to signal by virtue of Smad activation [25
]. Therefore, Smad signaling was tested in the presence of transferred molecules in the osteoblasts. Osteoblasts without a transfer event, or osteoblasts cultured without HPCs, displayed a high nuclear localization of Smad2/3, suggesting activated Smad signaling (). In contrast, osteoblasts with detectable N-Rh-PE transfer primarily had a cytoplasmic localization of Smad2/3 (), consistent with a reduction in activated Smad2/3 signaling. Quantification of the nuclear-cytoplasmic intensity ratio of Smad2/3 confirmed that Smad nuclear localization, and therefore TGFβ signaling, was diminished in osteoblasts following a N-Rh-PE transfer event ().
Previous studies have suggested that TGFβ1 signaling can be inhibitory to osteoblast maintenance of HPCs by downregulating a specific chemokine, SDF-1 [27
], produced by osteoblasts. SDF-1, also referred to as CXCL12, is known to modulate progenitor cell migration and adhesion [3
]. Immunofluorescence showed that approximately 30% of the osteoblasts in culture expressed detectable SDF-1 (). This finding is consistent with the osteoblasts having activated TGFβ1 signaling and the observed high levels of nuclear Smad2/3 (). We hypothesized that if the intercellular transfer event is reducing TGFβ1 signaling, we may observe a subsequent increase in SDF-1 expressing osteoblasts specifically following intercellular transfer. After 1 h of co-culture, 45% of the osteoblasts containing transferred material expressed SDF-1 (). shows an example of an SDF-1 expressing osteoblast (green) after uptake of N-Rh-PE (red) from an HPC. When the time of co-culture was increased to 5 h, almost 75% of the osteoblasts with a transfer event were expressing SDF-1 (). Therefore, after prolonged co-culture, osteoblasts positive for transferred N-Rh-PE were also positive for SDF-1 expression.
To discriminate between induction of SDF-1 expression by the transfer event or by continued KG1a cell contact, we washed the KG1a cells from the osteoblastic monolayer after 1 h of co-culture. After HPC removal, the osteoblasts were cultured an additional 4 h to evaluate SDF-1 expression. Once again, approximately 75% of osteoblasts with a transfer event were positive for SDF-1 expression (), therefore prolonged HPC contact with osteoblasts does not lead to the increased SDF-1 detected. To address the possibility that HPC contact with osteoblasts rather than transfer resulted in the increased SDF-1, we treated the HPC-osteoblast co-cultures for 1 h with 10 mM methyl β cyclodextrin (MβCD), 2 μm cytocholasin D (Cyto D), or 80 μm Dynasore, which we showed in reduced N-Rh-PE transfer. Drug treatment, which allows for osteoblast contact but reduces transfer, resulted in a decreased percent of SDF-1 expressing osteoblasts when compared to control co-cultures. These data suggest that intercellular transfer, rather than cell contact, leads to changes in protein expression within osteoblastic cells, more specifically to an increased production of SDF-1.
In this paper, we identify events following HPC-osteoblast contact that may play important roles in the signaling and remodeling events within the bone marrow niche. Specifically, we show that portions of the HPC uropod membrane at the contact site are actively endocytosed by osteoblasts and delivered to SARA-positive, signaling endosomes. In response to intercellular transfer, osteoblasts exhibit decreased Smad signaling and greater production of the bone marrow chemokine, SDF-1. This may occur because the transferred material within SARA-endosomes sequesters SARA away from its general co-factor function in Smad activation, allowing the osteoblast to produce more SDF-1. These events could greatly influence the overall trafficking, proliferation and survival of hematopoietic stem-progenitor cells in the bone marrow niche by permitting signaling events to be initiated and propagated at the single cell level. Furthermore, the intercellular transfer and downstream endosomal signaling described in this study could be significant for cell-cell communication not only in the bone marrow, but also in other intimate cellular microenvironments [30