Megakaryocyte emperipolesis mediates membrane transfer from intracytoplasmic neutrophils to platelets

Bone marrow megakaryocytes engulf neutrophils in a phenomenon termed emperipolesis. We show here that emperipolesis is a dynamic process mediated actively by both lineages, in part through the β2-integrin/ICAM-1/ezrin pathway. Tethered neutrophils enter in membrane-bound vesicles before penetrating into the megakaryocyte cytoplasm. Intracytoplasmic neutrophils develop membrane contiguity with the demarcation membrane system, thereby transferring membrane to the megakaryocyte and to daughter platelets. This phenomenon occurs in otherwise unmanipulated murine marrow in vivo, resulting in circulating platelets that bear membrane from non-megakaryocytic hematopoietic donors. Transit through megakaryocytes can be completed as rapidly as minutes, after which neutrophils egress intact. Emperipolesis is amplified in models of murine inflammation associated with platelet overproduction, contributing to platelet production in vitro and in vivo. These findings identify emperipolesis as a new cell-in-cell interaction that enables neutrophils and potentially other cells passing through the megakaryocyte cytoplasm to modulate the production and membrane content of platelets.


Introduction
Megakaryocytes (MKs) are the cellular source of platelets. Derived from hematopoietic stem cells, developing MKs undergo multiple rounds of endomitosis to become highly-polyploid cells averaging 20 to 100 mm in size (Levine et al., 1982;Machlus and Italiano, 2013). Mature MKs develop a complex network of intracytoplasmic membrane, termed the demarcation membrane system (DMS), that provides a membrane reservoir to enable platelet generation (Schulze et al., 2006). MKs then protrude pseudopodial extensions of this membrane via the marrow sinusoids into the bloodstream,

Results
In vitro modeling of emperipolesis reveals a rapid multi-stage process Whole-mount 3-dimensional (3D) immunofluorescence imaging of healthy C57Bl/6 murine marrow revealed that~6% of MKs contain at least one neutrophil, and occasionally other bone marrow cells ( Figure 1A and Video 1). Emperipolesis was similarly evident upon confocal imaging of unmanipulated human marrow ( Figure 1B). To model this process, we incubated cultured murine or human MKs with fresh bone marrow cells or peripheral blood neutrophils, respectively ( Figure 1C and D). Murine MKs, derived either from bone marrow or fetal liver cells, were efficient at emperipolesis (~20-40% of MKs). Neutrophils were by far the most common participants, although B220+ B cells, CD115+ monocytes, and occasional CD3+ T cells and NK1.1+ NK cells were also observed within MKs (Figure 1-figure supplement 1A). Emperipolesis was less efficient in human cultured MKs (2-5% of MKs), which are typically smaller than murine MKs, and was observed in MKs cultured from marrow CD34+ cells but not from the even smaller MKs derived from cord blood CD34+ cells ( Figure 1D and not shown). We elected to continue our mechanistic studies in murine MKs, principally cultured from marrow.

Neutrophils engaged in emperipolesis penetrate into the MK cytoplasm
Confocal microscopy revealed four distinct steps. First, neutrophils become adherent to the MK surface ( Figure 2A and Video 2 and 3), including to membrane protrusions we term MK tethers bound vacuoles, hereafter termed emperisomes, bearing the MK surface marker CD41+ (Figure 2A and Video 3). Third, the emperisome undergoes transformation such that CD41 is no longer evident surrounding the neutrophil (Figure 2A and Video 4). While most MKs engaged in emperipolesis contained only one or two neutrophils, some resembled 'reservoirs' containing dozens of neutrophils in stages 2 and 3 ( Figure 1-figure supplement 1C), an appearance recognized in human marrow as well (Cashell and Buss, 1992;Larsen, 1970;Monteferrario et al., 2014;Thiele et al., 1984) ( Figure 1-figure supplement 1D). Fourth, neutrophils exited MKs, returning to the extracellular milieu as viable motile cells ( To better understand the stages of emperipolesis, we employed electron microscopy (EM). After neutrophil uptake into the emperisome ( Figure 2B), the vacuolar space between neutrophil and MK was resorbed such that neutrophil and MK membranes became closely apposed, resulting in a structure composed of two membrane leaflets surrounding the neutrophil ( Figure 2C and D). This structure was often associated with the appearance of neutrophilic protrusions deeper into the host MK ( Figure 2C). Areas in which the membranes approximated very closely, becoming indistinct for short stretches, were sometimes observed ( Figure 2D). Subsequently, only a single bilipid membrane came to separate the neutrophil cytoplasm from the MK cytoplasm, a finding that echoed the loss of CD41 staining observed by immunofluorescence, confirming dissolution of the emperisome and thereby translocation of the neutrophil to an intracytoplasmic location ( Figure 2E). Whereas CD18 and Ly6G but not CD41 were preserved (Figure 2A above), this remaining membrane is most likely primarily of neutrophil origin.

Emperipolesis is mediated by active actin cytoskeleton rearrangement in both megakaryocyte and neutrophil
To assess the cytoskeletal processes underlying this intriguing cell-in-cell interaction, we employed targeted inhibitors. The microtubule polymerization inhibitor nocodazole showed a negligible effect, but emperipolesis was dramatically curtailed by inhibitors of actin polymerization, cytochalasin D and latrunculin A ( Figure 3A). This effect was observed when either MKs or marrow cells were exposed to these inhibitors, confirming obligate active cytoskeletal engagement by both participants ( Figure 3B  Video 1. Emperipolesis within murine bone marrow. Three-dimensional reconstitution of murine marrow, showing MKs (green) neutrophils (red), bone marrow sinusoids (white), and DNA (blue). Green, red or blue fluorescence are removed occasionally to visualize neutrophils inside MK or MK tethers. The three animations correspond to the three images shown in Figure 1A. leukocytes (Carman and Springer, 2004;Ley et al., 2007) (Figure 1-figure supplement 1E, Figure 3figure supplement 1B, and Video 2). In agreement with inhibitor findings, actin but not microtubules localized to the interface between MKs and extracellular neutrophils and was observed to encase neutrophil-containing CD41+ vacuoles ( Figure 3D and Figure 3-figure supplement 1C); by contrast, actin was not observed surrounding neutrophils that were no longer delimited by CD41-expressing membrane ( Figure 3C). These observations demonstrate that emperipolesis is an active process mediated by actin cytoskeletal rearrangement of both MKs and neutrophils.
b2 integrins bind ICAM-1, among other targets (Ley et al., 2007). Confocal microscopy showed that ICAM-1 is expressed by a population of human and murine MKs (Figure 3-figure supplement 2A-D). In agreement with previous observations in rat (Tanaka et al., 1997), emperipolesis by ICAM-1-deficient MKs was significantly impaired ( Figure 3F). In further support of this mechanism, we evaluated the role of ezrin, which mediates the attachment of the intracellular tail of ICAM-1 to the actin cytoskeleton (Heiska et al., 1998;Ley et al., 2007). Inhibition of ezrin phosphorylation impaired emperipolesis ( Figure 3G, controls of ezrin inhibition in Figure 3-figure supplement 2E). By confocal microscopy, ezrin could be detected only at sites of MK contact with tethered neutrophils, where it co-localized strongly with ICAM-1, consistent with its role as a bridge to the cytoskeleton ( Figure 3H and I). By contrast, ezrin could not be visualized in MKs not tethered to leukocytes, or with only internalized leukocytes ( Figure 3H). Together, these data show that emperipolesis is mediated in part by an interaction between neutrophil b2 integrins and MK ICAM-1/ezrin during neutrophil entry. However, absence or blockade of these factors resulted in only partial impairment of emperipolesis, indicating a role for alternate mechanisms not yet defined.

Emperipolesis mediates membrane transfer from neutrophil to megakaryocyte
As observed by others (Centurione et al., 2004;Thiele et al., 1984), neutrophils engaged in emperipolesis frequently localized to the DMS, the intracytoplasmic membrane network implicated in platelet production ( Figure 4A). Close examination of this interaction demonstrated membrane contiguity between the neutrophil and the DMS, suggesting that neutrophils might be able to serve as membrane donors to MKs and potentially to platelets ( Figure 4A). To test this possibility, we employed membrane labeling. Marrow cells were stained with the lipophilic dye CellVue maroon and then co-cultured with unstained MKs. Confocal microscopy showed To assess reciprocal membrane transfer from MKs to neutrophils, we stained MKs with lipid stains as above and co-cultured these with unstained marrow cells. MK-derived lipids strongly co-localized with neutrophil membrane during emperipolesis ( While lipid exchange from neutrophils to MKs was strongly inhibited by latrunculin A, transfer from MKs to neutrophils was not ( Figure 4C), suggesting that this reciprocal transfer was not mediated primarily by emperipolesis. MKs produce microparticles in great abundance (Cunin et al., 2017;Flaumenhaft et al., 2009), and PKH67-stained MKs were observed to release many PKH67 +microparticles in a latrunculin A-independent manner that could transfer membrane fluorescence to neutrophils in the absence of intact MKs (Figure 4-figure supplement 1J). By contrast, no fluorescence was detected on MKs cultured with supernatant from PKH67-stained marrow cells, rendering unlikely a role of marrow cell-derived microparticles, exosomes or apoptotic bodies in membrane transfer Video 3. Megakaryocyte tethers and neutrophil entry into a CD41+ vacuole. MKs stained with anti-CD41 (green) were co-cultured with marrow cells from mT/ mG mice (red) in the presence of Draq5 (DNA, blue). Video shows a neutrophil on the MK surface, attached by MK tethers, followed by a rapid entry through a CD41+ membrane. A few minutes after its entry, the neutrophil exits at the bottom of the field of view We then sought to determine whether membrane transfer mediates exchange of surface proteins. We performed surface biotinylation of MKs and marrow cells, and then co-cultured these cells with unstained marrow cells or MKs, respectively. Using streptavidin, we could not detect biotin on neutrophils incubated with biotinylated MKs, suggesting the absence of bulk surface protein transfer from MKs to neutrophils (not shown). However, surface biotin could be detected on some MKs after incubation with biotinylated marrow cells ( Figure 4D), confirming that membrane exchange from neutrophils to MK transfers proteins. The nature of these proteins remains to be determined since MKs remained negative for hallmark neutrophil proteins such as CD18 and Ly6G (not shown).

Neutrophil membranes transferred in emperipolesis emerge on circulating platelets
Platelets are generated by MKs via the DMS network, an impressively extended network of membrane whose biogenesis remains incompletely understood (Eckly et al., 2014;Schulze et al., 2006). EM had demonstrated membrane continuity between cytoplasmic neutrophils and the DMS. We therefore tested whether emperipolesis could transfer neutrophil membrane to platelets. MKs require shear stress for physiological platelet biogenesis, rendering the in vivo context most suitable for these studies. MKs stained with the cytoplasmic dye Green-CMFDA were incubated with marrow cells stained with the lipid marker CellVue Maroon and then engrafted intravenously into congenic recipient mice, in which production of CMFDA+ platelets was monitored by serial phlebotomy (Cunin et al., 2017;Fuentes et al., 2010;Zhang et al., 2016) and Figure 5-figure supplement 1A). Remarkably, most platelets produced by donor MKs (i.e. CMFDA+) were also positive for Cell-Vue Maroon, indicating a high frequency of incorporation of donor leukocyte membrane ( Figure 5A). The intensity of CellVue Maroon staining remained constant over time, suggesting that donor membrane was employed continuously over an extended period ( Figure 5B). Similar findings were obtained with lipid stainer PKH67 (Figure 5-figure supplement 1B). To exclude experimental artifact related to lipid stains, we employed donor marrow from mT/mG mice bearing membrane fluorescence mediated by fluorochrome associated with the inner membrane leaflet. Confocal imaging confirmed that membrane fluorescence from mT/mG marrow cells efficiently transferred into MKs in vitro ( Figure 5-figure supplement 1C). Membrane fluorescence was also detected on platelets produced in vivo by WT MKs incubated with mT/mG marrow donors, albeit with weaker signal since membrane fluorescence is less intense that with lipid stains (Figure 5-figure supplement 1D). We similarly investigated transfer of intracellular or surface protein. Marrow cells were stained with the intracellular protein stain CellTrace Violet and then co-cultured with CMFDA +MKs. Interestingly, platelets emerging in vivo contained CellTrace violet, consistent with cytoplasmic protein transfer ( Figure 5C and D). Together, these results demonstrate that lipids and intracellular proteins are transferred from marrow cells not only to MKs but also to their daughter platelets. Of note, we could not detect biotin on emerging platelet when MKs were previously co-cultured with surfacebiotinylated marrow cells ( Figure 5-figure supplement 1E). Moreover, platelets were negative for neutrophil surface proteins Ly6G, CD11b, CD18 and CD88 (not shown). We cannot exclude the possibility that other surface proteins not directly assessed may still transfer in quantities too modest to be detected by bulk biotin-streptavidin staining.
Video 6. Transfer of membrane during long-lasting emperipolesis. MKs stained with PKH67 (green) were co-cultured with marrow cells stained with PKH26 (red) in the presence of Draq5 (DNA, blue). Video shows a neutrophil residing within an MK. Green or red fluorescence is removed at some time points to visualize bi-directional membrane transfer. DOI: https://doi.org/10.7554/eLife.44031.010   We sought to exclude the possibility that this membrane transfer reflected an artifact of ex vivo MK generation and co-culture. To this end, we generated mice chimeric for WT and mT/mG marrow, allowing us to seek platelets resulting from mT/mGfiWT membrane transfer in a fully native environment. Indeed, platelets with the expected intermediate fluorescent phenotype were observed ( Figure 5E), albeit only in relatively small numbers, potentially because of the weak fluorescence in the mT/mG system and because WTfiWT, mT/mGfimT/mG, and WTfimT/mG transfer events remain undetectable. Examination of BM MKs identified examples of fluorescent neutrophils contributing membrane to non-fluorescent MKs from an intracellular location ( Figure 5-figure supplement 1F). We conclude that intracellular neutrophils transfer membrane to MKs and thereby to platelets via emperipolesis in vivo. Of note, we observed an important fraction of the 'red-intermediate' platelets in the mT/mG/WT chimeric mice expressing phosphatidylserine ( Figure 5F and G; control for Annexin V staining in Figure 5-figure supplement 1G), while we observe normal levels of CD62P and a marginal band of b1-tubulin by microscopy ( Figure 5-figure supplement 1H-I), excluding an abnormal activation phenotype (Moskalensky et al., 2018;Sadoul, 2015). Surface phosphatidylserine creates a scaffold for clotting factors and is a hallmark of pro-coagulant platelets (Heemskerk et al., 2013;Nagata et al., 2016). This result suggests the possibility that emperipolesis-derived platelets could be functionally distinct, potentially including enhanced thrombogenic capacity.

Emperipolesis enhances platelet production
As a bidirectional interaction between MKs and leukocytes, emperipolesis is likely to have multiple cellular effects. Among these, we elected to explore its impact on thrombocytogenesis. Recognizing the association of emperipolesis in humans with hematopoietic disease (Cashell and Buss, 1992;Centurione et al., 2004;Larsen, 1970;Mangi and Mufti, 1992;Stahl et al., 1991;Thiele et al., 1984), we exposed mice to several models of stress-induced platelet over-production by MKs, intraperitoneal LPS injection and IgG-mediated thrombocytopenia. The proportion of MKs containing at least one neutrophil was assessed in two-dimensional marrow sections. In each case, emperipolesis increased from a baseline of~2-5% in control mice to~6-10% under stress ( Figure 6A-C). These figures represent a minimal estimate of the 'snapshot' prevalence of emperipolesis, since they sample only 5 mm sections of MKs with a typical diameter of 20-100 mm, but nevertheless confirm that emperipolesis is common and strongly induced under physiological stress. Interestingly, an enhanced drive for platelet production was not sufficient to augment emperipolesis, because accelerated platelet production following administration of thrombopoietin, or platelet depletion by anti-CD41 was unaccompanied by an increase in emperipolesis ( Figure 6-figure supplement 1A and B). One possible explanation is that neutrophil activation may also be required, consistent with the role of neutrophil b2 integrins defined above.
To quantitate the impact of emperipolesis on thrombocytopoiesis, we employed IncuCyte highcontent live-cell microscopy (Thon et al., 2012), comparing pro-platelet generation by MKs cultured alone or together with marrow cells. These studies employed fetal liver MKs because of their superior ability to generate pro-platelets in vitro. To assess the role of cell-cell contact and bone marrow cell-derived soluble factors, including microparticles, we cultured MKs with marrow cell supernatant or with paraformaldehyde-fixed marrow cells. Co-culture with living marrow cells markedly enhanced pro-platelet production ( Figure 6D). By contrast, MKs cultured with fixed cells or cell supernatants produced fewer pro-platelets than those cultured alone, weighing against a role for contact and soluble factors and implicating emperipolesis directly ( Figure 6D).
Finally, we tested the effect of emperipolesis on platelet production in vivo via adoptive transfer. MKs were labeled either with Green-CMFDA or with CellTracker Deep Red, and one population or the other (varied across experiments) was cultured together with marrow cells. MKs were mixed 1:1 and engrafted IV into recipient animals for serial parallel quantitation of green and red platelets (Figure 6-figure supplement 1C). As predicted by the IncuCyte findings, MKs cultured with marrow cells were more efficient at producing platelets ( Figure 6E and F), consistent with promotion of thrombocytogenesis by emperipolesis.

Discussion
Megakaryocytes anchor hemostasis via elaboration of platelets. Platelet production can occur in a cell-intrinsic manner, as for example by MKs cultured in isolation ex vivo. However, physiological platelet generation proceeds in a complex multicellular environment. The present studies establish a pathway through which this cellular context modulates thrombocytogenesis. During emperipolesis, neutrophils and other hematopoietic lineages penetrate into the MK cytoplasm, a process mediated actively by both host and donor. This process is distinct from phagocytosis since the neutrophil actively penetrates into the MK and survives to exit intact. Cytoplasmic neutrophils transfer membrane and cytosolic contents to MKs and to platelets, thereby enhancing platelet production. Donor neutrophils receive membrane in turn before they exit intact (Figure 7). Thus, emperipolesis represents a previously unrecognized pathway through which neutrophils and other hematopoietic cells engage with MKs to modulate the composition and production of circulating platelets.
We identified b2 integrins and MK ICAM-1/ezrin as contributors to emperipolesis. These proteins also mediate another form of transcellular passage, the migration of neutrophils through the cell bodies of endothelial cells (Ley et al., 2007). Unlike emperipolesis, endothelial transcellular migration is not known to involve penetration into the host cytoplasm. It remains unknown whether other mechanisms are shared between emperipolesis and transendothelial migration, such as fusion of caveolin vesicles to create an intracellular channel for passage (Ley et al., 2007;Millán et al., 2006). Tavassoli and colleagues had previously postulated that emperipolesis could represent a pathway of neutrophil egress from the bone marrow (Dziecioł et al., 1995;Tavassoli, 1986). Transit of some neutrophils through MKs over the course of just a few minutes lends plausibility to this hypothesis. Our data do not exclude the possibility that some neutrophils pass through MKs without a cytoplasmic 'detour,' thereby resembling endothelial transcellular migration even more closely.
Mechanisms of MK-emperipolesis have remained entirely obscure for almost 50 years (Larsen, 1970). Electron microscopy observations previously raised the possibility that neutrophils are not internalized by MKs but rather enter directly through the DMS, which is continuous with the cell surface (Eckly et al., 2014), to reside within DMS dilated cavities (Breton-Gorius and Reyes, 1976;Thiele et al., 1984). Consistent with these observations, myeloid cells are often found at the cell surface entrance of the DMS (Thiele et al., 1984), and increased emperipolesis has been reported in models with dilated and enlarged DMS or after pharmacological modification of the DMS (Overholtzer and Brugge, 2008). However, our confocal and electron microscopy images demonstrate that neutrophils enter MKs directly, through a vacuole, ultimately taking up residence inside the MK cytoplasm. Emperipolesis is nevertheless strikingly heterogeneous. For example, emperipolesis can be observed in MKs of all sizes and can last just minutes to over an hour. MKs may enclose a single neutrophil or encompass as many as 50 neutrophils (Figure 1-figure supplement 1C). These observations strongly suggest that there may be different types of emperipolesis, involving different molecular pathways and serving distinct functions that remain to be defined.
Because the mechanistic pathways identified in our study are not specific to emperipolesis (e.g. b2 integrin binding, actin polymerization), we have so far been unable to interrupt this phenomenon with selectivity in vivo. To address its function, we employed a combination of approaches, focused here on platelet production. This focus is justified by the enhanced frequency of emperipolesis in diseases associated with high platelet count, including essential thrombocythemia, polycythemia vera (Cashell and Buss, 1992;Vytrva et al., 2014), or with high platelet demand (gray platelet syndrome, blood loss or hemorrhagic shock [Di Buduo et al., 2016;Dziecioł et al., 1995;Larocca et al., 2015;Monteferrario et al., 2014;Sahebekhitiari and Tavassoli, 1976;Tavassoli, 1986]). Further, enhanced emperipolesis in chronic myeloproliferative disorders positively correlates with the peripheral platelet count (Thiele et al., 1984). We establish that emperipolesis accelerates platelet production both in vitro and in vivo. The quantitative importance of this contribution, and whether it reflects enhanced access to lipid membrane or some other mechanism, remains to be established.
Since platelets generated through emperipolesis bear donor membrane as well as parent MK membrane, it is likely that they will be different in function. Consistent with this possibility, we found enhanced expression of surface phosphatidylserine on emperipolesis-derived platelets in our chimeric WT-mT/mG mice ( Figure 5F and G). Given the role of b2 integrins in emperipolesis, it is plausible to suspect that activated neutrophils will preferentially engage in emperipolesis, suggesting the possibility that 'angry neutrophils make angry platelets'. The identity of lipids and proteins transferred from neutrophils and other cells to MKs and platelets, and the resulting changes in cell function, will be important topics for future study.
The impact of emperipolesis is unlikely to be restricted to MKs and platelets. Our videomicroscopy data confirm that exiting neutrophils can carry MK membrane with them ( Figure 4-figure  supplement 1D), potentially translating into altered function. These effects are more difficult to study in vitro because of the capacity we identified for MKs to transfer membrane to nearby cells via MK microparticles (Figure 4-figure supplement 1H). Our prior work had identified MK microparticles as potent pro-inflammatory vectors implicated in the delivery of IL-1 during systemic inflammatory disease (Cunin et al., 2017). The present findings thus extend the understanding of MK microparticles as signaling vectors. Further, by establishing transfer of membrane not only from donor cell to MK but also reciprocally, followed by release of viable cells back into the intercellular milieu, these studies identify emperipolesis as a mechanism through which MKs may be able to 'groom' neutrophils and other immune lineages.
We are unable at present to define conclusively the proportion of circulating platelets that bear neutrophil membrane. The mT/mG chimera experiments will not accurately reflect this fraction, because only one of 4 possible donor-host pairs yields detectable platelets (mT/mG neutrophilfiWT MK) and because mT/mG fluorescence is weak, such that transfer events will likely often be invisible. We note that many platelets released by MKs co-cultured with membrane-labeled marrow donors expressed membrane label ( Figure 5). Given the speed with which cells enter and exit MKs, a snapshot prevalence of 6% is compatible with the possibility that many or even most MKs, and many neutrophils, experience emperipolesis over time, perhaps repeatedly. If this is the case, and labeling experiments accurately reflect the efficiency of membrane transfer, then emperipolesis-derived membrane could be common in circulating platelets. Alternately, if transfer were inefficient, and/or only a subset of MKs engaged in emperipolesis, then emperipolesis-modulated platelets could represent simply a small (but potentially still functionally important) subset of the circulating pool.
We recognize other limitations to these studies. Unique functional contributions of emperipolesisderived platelets remain to be established. The mechanisms through which neutrophils escape emperisomes to enter the cytoplasm, home intracellularly to the DMS, and then egress without violating MK outer membrane integrity remain to be defined. Cells deficient in b2 integrins retain the capacity to enter MKs, albeit with reduced efficiency, revealing that other ligand/receptor pairs can mediate entry. The signals driving enhanced emperipolesis in the setting of experimental stress, during hemorrhagic shock, and in aberrant marrow environments such as in hematopoietic malignancies, remain to be established. Despite these limitations, the current studies identify emperipolesis as a novel cell-in-cell interaction that mediates reciprocal transfer of membrane and other cellular components, defining thereby a new mechanism of interchange between immune and hematopoietic systems.

Materials and methods
Mice C57Bl/6, CD45.1 B6 mice, cd18 -/mice, mT/mG mice and Tg-FcgRIIA mice were purchased from The Jackson Laboratory. LyzM-GFP mice transgenic for FcgRIIA (Tg-FcgRIIA) mice were backcrossed 10 times in the C57BL/6J background. Unless stated, all experiments employed male mice aged 8-12 weeks. All procedures were approved by the local animal care committee.

Chemicals and reagents
Latrunculin A and Cytochalasin D were purchased from Cayman Chemical. Ezrin inhibitor NSC668394 was from Calbiochem. Lipids cells strainers PKH67 and PKH26 were from Sigma. Protein strainers Green CMFDA, CellTracker Deep Red and Cell Trace Violet were from Molecular probes. Surface protein biotinylation kit was purchased from Pierce.

Confocal microscopy
Cells were fixed in PFA 2% for 30 min at RT. After washing, cells were suspended in PBS supplemented with 0.1% saponin and 3% FCS (permeabilization buffer) and incubated 2 hr at RT or overnight at 4C with 10 mg/ml primary antibodies. After washing in permeabilization buffer, secondary antibodies diluted 1:200 were added for 1 hr. When indicated, Phalloidin (Molecular Probe), Draq5 (eBioscience) or Hoechst (ThermoFisher) were added for the 15 last minutes, prior washing with PBS and cytospin. Cells were mounted on slides using FluorSave mounting medium (Calbiochem). Microscopy was performed using a Nikon C1 Plus Confocal Laser Scanner confocal or a Zeiss LSM 710 or 800 Multiphoton Laser scanner confocal microscope.

Spinning disk confocal microscopy
MKs stained with PKH26 and marrow cells stained with PKH67 were co-cultured in P96 round bottom wells for at least 1 hr prior to spinning disk imaging. Cells were then resuspended in TPO medium without red phenol and supplemented with Draq5, and cultured in a micro insert 4-well dish for time lapse imaging. Movies were obtained using a YokogawaCSU-X1 or an Olympus DSU inverted spinning disk confocal microscope. Images were acquired every 4 min on 12 different Z-stacks, 1 mm per stack. Movies were analyzed using EZ element software or Volocity software.

Imaging of whole-mount bone marrow
Whole-mount-tissue preparation, immunofluorescence staining and imaging of the bone marrows were performed as described previously (Kunisaki et al., 2013). Briefly, mice were intravenously injected with AF647-labelled anti-CD31 and anti-CD144 and perfused with PBS and 4% PFA15 min after. Femurs and tibias were harvested, PFA-fixed, frozen in OCT, and shaved on a cryostat to expose the marrow. Bones were incubated in PBS containing 10% FCS and 0.5% Triton X-100, with AF594 anti-Ly6G, AF488 anti-CD41 and Hoechst for 2 days. Images were acquired using a Zeiss LSM 800 Multiphoton Laser scanner confocal microscope and reconstructed in 3D with Imaris software.

Electron microscopy
Cells were fixed using 2% PFA and 0.1% glutaraldehyde for 1 hr RT. After washings, cells were incubated with 0.1% OsO4 for 30 min prior to sectioning. 50 nm sections were observed with a JEOL 1200EX electron microscope.

In vitro pro-platelet production
MKs were co-cultured overnight without or with PKH67-stained marrow cells. Marrow cells were separated from MKs using BSA-gradient sedimentation, and cells were transferred to a P24 well plate and imaged using the IncuCyte HD system (Essen BioScience). Frames were captured every hour. Rates and extent of proplatelet production were measured in ImageJ software using investigatorcoded software (Thon et al., 2012).
In vivo platelet production 2 Â 10 5 MKs, previously stained with Green CMFDA or CellTracker Deep Red (Molecular probes) and co-cultured or not with marrow cells, were injected i.v. in 200 ml PBS. Blood was harvested by tail vein sampling at indicated time points using heparinized capillary tubes. 1 ml was blood is diluted in 500 ml PBS in the presence of an anti-CD41 antibody. Presence of green CMFDA or CellTracker Deep Red on circulating CD41+ platelets was evaluated by flow cytometry.

Emperipolesis in marrow sections
Bones were fixed in PFA 4% for 2 days prior to decalcification in Kristensen solution for 2 days prior to paraffin-embedding. Percentages of marrow MKs containing at least one neutrophil were determined on 6 mm paraffin-embedded sections stained with H and E.

Statistics
Statistical significance in emperipolesis between two conditions was determined using the Chisquare test. Number of MKs counted per sample is reported in Figure legends. To compare emperipolesis in 2 groups of mice we used the Two-tailed Mann-Whitney test. In vitro pro-platelet production (IncuCyte experiment) and in vivo platelet production over time were analyzed with the two-way analysis of variance (ANOVA). All statistical analysis were done using Prism software, *p<0.05, **p<0.01 ***p<0.001.