A Cdc42/RhoA regulatory circuit downstream of glycoprotein Ib guides transendothelial platelet biogenesis

Blood platelets are produced by large bone marrow (BM) precursor cells, megakaryocytes (MKs), which extend cytoplasmic protrusions (proplatelets) into BM sinusoids. The molecular cues that control MK polarization towards sinusoids and limit transendothelial crossing to proplatelets remain unknown. Here, we show that the small GTPases Cdc42 and RhoA act as a regulatory circuit downstream of the MK-specific mechanoreceptor GPIb to coordinate polarized transendothelial platelet biogenesis. Functional deficiency of either GPIb or Cdc42 impairs transendothelial proplatelet formation. In the absence of RhoA, increased Cdc42 activity and MK hyperpolarization triggers GPIb-dependent transmigration of entire MKs into BM sinusoids. These findings position Cdc42 (go-signal) and RhoA (stop-signal) at the centre of a molecular checkpoint downstream of GPIb that controls transendothelial platelet biogenesis. Our results may open new avenues for the treatment of platelet production disorders and help to explain the thrombocytopenia in patients with Bernard–Soulier syndrome, a bleeding disorder caused by defects in GPIb-IX-V.

B lood platelets are anucleated cell fragments derived from bone marrow (BM) megakaryocytes (MKs) that are essential for blood clotting. MKs develop from haematopoietic stem cells by a complex maturation process, which includes DNA replication without cell division (endomitosis) and the formation of a demarcation membrane system (DMS), which functions as a membrane reservoir for newly formed platelets 1,2 . Mature BM MKs are giant, polyploid cells that localize close to BM sinusoids in order to extend and release long cytoplasmic protrusions called proplatelets into the sinusoidal lumen, from which platelets are shed under the influence of blood shear forces [3][4][5][6] . The highly controlled process of transendothelial platelet biogenesis stands in contrast to transendothelial migration of whole haematopoietic (progenitor) cells during mobilization and homing 7 . A current concept incorporates the idea that the chemokine stromal cell-derived factor-1 triggers migration of early MKs from the endosteal niche towards BM sinusoids 8,9 and that the lipid mediator sphingosine-1-phosphate elicits directed proplatelet extension into the circulation 10 . However, the molecular mechanisms regulating the trafficking of MKs and haematopoietic cells within the BM and across the endothelial barrier are poorly defined.
Rho GTPases are small proteins (20)(21)(22)(23)(24)(25) belonging to the superfamily of Ras-related proteins which are found in all eukaryotic cells 11 . They are best known for regulating cytoskeletal dynamics in virtually all cell types 12 . The best-characterized Rho GTPases are Cdc42, RhoA and Rac1, whose activation is associated with the formation of filopodia, stress fibres and lamellipodia, respectively. We previously reported that transgenic mice lacking either Cdc42 or RhoA in MKs and platelets exhibit pronounced macrothrombocytopenia, indicating a distinct role of these molecules in platelet production 13,14 . In addition, Cdc42 deficiency led to decreased filopodia formation of platelets on von Willebrand factor (vWF), suggesting a unique role of Cdc42 downstream of the glycoprotein (GP) Ib subunit of the vWF receptor complex GPIb-IX-V (ref. 13). Bernard-Soulier syndrome (BSS) is a rare platelet disorder characterized by macrothrombocytopenia, which is caused by damaging variants in either of the three genes encoding the GPIba/b or GPIX subunits of the GPIb-IX-V complex, leading to its absence from, or dysfunction at the MK and platelet surface [15][16][17][18][19] . The mechanisms by which the receptor controls megakaryopoiesis and platelet production are largely unknown.
Here, we show that lack of either functional GPIb or Cdc42 reduced MK polarization in vitro and impaired MK localization at sinusoids and transendothelial biogenesis in vivo. In contrast, absence of RhoA in MKs resulted in increased Cdc42 activity and GPIb-dependent transendothelial migration of entire MKs into BM sinusoids. These results reveal that Cdc42 and RhoA act as a regulatory circuit downstream of GPIba to coordinate MK polarization and transendothelial platelet biogenesis in vivo.
In view of the multiple intrinsic defects present in MKs deficient in GPIba 22 or GPIbb 23 we turned to a mouse line expressing a mutant version of GPIba, where the ectodomain of GPIba is replaced by that of the human interleukin-4 receptor a (IL-4Ra) (Gp1ba À / À ;tg , further referred to as Gp1ba-Tg). In these mice, the BSS-associated macrothrombocytopenia is ameliorated, but not fully reversed ( Supplementary Fig. 1a,b (refs 21,24)), implying a specific role for the ectodomain of GPIba in platelet biogenesis. Strikingly, the BM MK distribution in Gp1ba-Tg mice and Gp1ba À / À mice was similar, with a significant decrease in the proportion of MKs with sinusoidal contact (wt: 67.5±6.4%; Gp1ba-Tg: 49.4%±7.0%; Po0.0001) and a concomitant increase of BMHC-localized MKs (wt: 30.3±6.6%; Gp1ba-Tg: 49.0±6.0%; Po0.0001; two-way ANOVA with Bonferroni correction for multiple comparisons) (Fig. 2a,c). The role of the GPIba ectodomain for MK localization was further confirmed by treatment of wt mice with the monovalent Fab fragment of an antibody directed against the major ligand (vWF)-binding domain of GPIba (p0p/B-Fab) that is known not to affect platelet survival in the circulation 25 . While GPIba blockade had no effect on total MK numbers in the BM ( Supplementary Fig. 1c), it clearly reduced the fraction of MKs with direct sinusoidal contact (39.5 ± 4.4%; Po0.0001; two-way ANOVA with Bonferroni correction for multiple comparisons, Fig. 2a,c). This was associated with a reduction in peripheral platelet counts and an increase in platelet size by about one third (Fig. 2d,e) as compared to wt, similar to that seen in Gp1ba-Tg mice ( Supplementary Fig. 1a,b). Formation of the DMS, as analysed by transmission electron microscopy (TEM), was unaltered in MKs of Gp1ba-Tg mice or wt mice after GPIba blockade (Fig. 2b). Together, these findings revealed a critical role for the ectodomain of GPIba in controlling MK localization at vascular sinusoids, independent of its role in DMS development and partitioning.
We next sought to get insights into the signalling that triggers GPIba-dependent MK guidance within the BM. Using mice lacking Cdc42 in MKs and platelets (Cdc42 fl/fl Pf4-cre , further referred to as Cdc42 À / À (ref. 13)), we have previously shown that Cdc42 controls cytoskeletal dynamics downstream of GPIba in platelets and that its absence in MKs causes marked macrothrombocytopenia by unknown mechanisms 13 . Intriguingly, significantly fewer Cdc42 À / À MKs were in direct contact with BM sinusoids in line with an increased MK population in the BMHC (Fig. 2f,g) compared to wt littermate controls (52.0±3.6 P ¼ 0.0002; two-way ANOVA with Bonferroni correction for multiple comparisons). Importantly, Cdc42 À / À MKs displayed only partially reduced membrane invaginations, indicating that the thrombocytopenia in these mice was not solely the result of a lack of intracellular membranes (Fig. 2h). Another key component of the GPIba signalling machinery is phosphoinositide 3-kinase (PI3K) that has been shown to play an important role in thrombosis and GPIbainduced integrin activation 26,27 . Of note, treatment of wt mice with the PI3K inhibitor wortmannin reduced the number of MKs in direct contact with sinusoids compared to the control (55.0 ± 5.9%, P ¼ 0.003; two-way ANOVA with Bonferroni correction for multiple comparisons) ( Supplementary Fig. 2d,e), whereas it had no effect on DMS development or total MK numbers in the BM ( Supplementary Fig. 2a,b,f).
In neuronal cells, PI3K-mediated establishment of cell polarity was shown to involve the interplay of GTPases with atypical protein kinase C (aPKC) isoforms, which form part of the PAR ('partitioning defective') complex 28 . We observed reduced sinusoidal localization of MKs in mice lacking the aPKC isoform ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15838 PKCi in MKs/platelets ( Supplementary Fig. 3a,b), indicating a role of the PAR complex and atypical PKCs for PI3K-Cdc42 controlled MK polarization and migration/localization. Regulation of cell polarization by the PAR complex is characterized by a profound reorganization of the microtubule cytoskeleton 29 . Consistently, we found that microtubule disruption by nocodazole treatment prevented transmigration of RhoA-deficient MKs ( Supplementary Fig. 3c,d, see below) similar to Cdc42-deficiency.
RhoA negatively regulates GPIb-Cdc42 signalling in MKs. Rho GTPases integrate phosphoinositide signalling and cytoskeletal dynamics. This involves a cross-regulation between Cdc42 and RhoA, which tunes GTPase function in different cell types 30,31 . It was recently demonstrated that Cdc42 is important to polarize the MK DMS towards BM sinusoids 32 . Moreover, we have previously shown that mice with MK/platelet-specific RhoA-deficiency (RhoA fl/fl Pf4-cre , further referred to as RhoA À / À ) exhibit pronounced macrothrombocytopenia ( Supplementary Fig. 4a,b). This finding was striking since it was reported that RhoA negatively regulates proplatelet formation of human CD34 þ derived MKs in vitro 33 . Supporting this result, we found that proplatelet formation of RhoA À / À fetal liver cell-derived MKs in liquid culture in vitro was comparable to the wt (Supplementary Fig. 4e (ref. 14)). Taking these observations together, we hypothesized that the apparent platelet biogenesis defect in RhoA À / À mice might be mechanistically linked to defective control of GPIba-Cdc42 signalling and MK localization within the BM. Intriguingly, in BM sections of RhoA À / À mice, we found a dramatic MK mislocalization with approximately 30% of the cells being present inside the BM sinusoids (27.9±10%, Po0.0001; two-way ANOVA with Bonferroni correction for multiple comparisons; Fig. 3a,b, Supplementary Fig. 4c,d). The remaining MKs were mostly found in direct contact with sinusoids (61.2±8.6%, P ¼ 0.430; two-way ANOVA with Bonferroni correction for multiple comparisons), while only a very minor population was present in the BMHC (10.8±2.3%, Po0.0001; two-way ANOVA with Bonferroni correction for multiple comparisons). To analyse the ultrastructure of intrasinusoidal MKs in the intact BM, we performed TEM on cross-sections of intact femora (Fig. 3c, Supplementary Fig. 5). Intrasinusoidal RhoA À / À MKs showed signs of maturation including poly-lobulated nuclei. The MKs either contained a large quantity of cytoplasm with a well-developed membrane system and a pronounced peripheral zone, or showed signs of flowdependent DMS/cytoplasm detachment suggestive of proplatelet formation. Nevertheless, most MKs remained in contact with sinusoidal endothelial cells. Of note, DMS formation appeared mostly unaffected in RhoA À / À MKs when analysed by TEM on BM cross-sections ( Supplementary Fig. 5c), but we frequently found intact cells of other lineages (typically granulocytes) within the MKs cytoplasm ( Fig. 3c, Supplementary Fig. 5d), a phenomenon known as emperipolesis 34 .
Next, we utilized intravital two-photon microscopy (2PM) for dynamic visualization of MK localization and proplatelet formation in the BM of the skull over time. As previously described 4 , in wt mice multiple MKs were in direct sinusoidal contact and eventually released long proplatelets into the vascular sinus ( Fig. 3d- Although the pulmonary microvessels represent the first capillary bed that receives circulating blood from the BM 35 , we did not observe increased numbers of MKs in the lungs of RhoA À / À mice ( Supplementary Fig. 6d,e), suggesting that the intrasinusoidal MKs predominantly fragment in the BM vasculature or elsewhere in the circulation 3 . These observations collectively indicate that RhoA negatively regulates MK guidance/ localization towards BM vascular sinusoids and thereby acts as a key regulator of proplatelet extension across the endothelial barrier by providing a stop-signal for MK transmigration.
To assess whether the transmigration of RhoA À / À MKs was indeed GPIba-dependent, we blocked the major ligand-binding site of the receptor in wt and RhoA À / À mice by injection of p0p/ B-Fab. Strikingly, whereas this treatment did not affect overall MK numbers in the BM ( Supplementary Fig. 7), it reduced the intrasinusoidal localization of RhoA À / À MKs by more than 70% (8.2±4.4%, Po0.001; two-way ANOVA with Bonferroni correction for multiple comparisons, Fig. 4a,d). This was associated with an increased presence of MKs in direct contact with BM sinusoids and a decreased percentage of BMHClocalized MKs compared with wt controls and vehicle-treated RhoA À / À mice, respectively (Fig. 4d). In contrast, treatment with Fab fragments of antibodies directed against the major ( Supplementary Fig. 7b,c), nor was it able to revert the transendothelial migration of RhoA À / À MKs, demonstrating a specific role for GPIb in these processes. We also generated RhoA À / À /Gp1ba-Tg mice, in which we found similar results as compared to p0p/B-Fab treatment (9.5 ± 4.5% MKs intrasinusoidal, Fig. 4b,d). Interestingly, inhibition of PI3K by wortmannin treatment likewise significantly reduced the number of intrasinusoidal MKs in RhoA À / À mice (12.3 ± 5.2%, Po0.0001; two-way ANOVA with Bonferroni correction for multiple comparisons), but did not increase the number of MKs with direct sinusoidal contact ( Supplementary Fig. 2d,e). Total MK numbers in the BM were unaltered in wortmannin-treated compared to vehicle-treated RhoA À / À mice ( Supplementary  Fig. 2f) and also their DMS appeared largely unaltered ( Supplementary Fig. 2c). Together, these data indicate that RhoA may act as a negative regulator of GPIba-Cdc42-driven MK guidance towards BM sinusoids and transendothelial migration. To further test this hypothesis, we generated MK/platelet-specific RhoA/Cdc42double-deficient mice (RhoA/Cdc42 fl/fl Pf4-cre , further referred to as RhoA/Cdc42 À / À ; Supplementary Figs 8a and 11a-c). Strikingly, RhoA/Cdc42 À / À MKs massively accumulated around vascular sinusoids (82.9 ± 5.9%, P ¼ 0.0002; two-way ANOVA with Bonferroni correction for multiple comparisons, Fig. 4c-e, Supplementary Movie 6), but failed to transmigrate into the lumen (3.9 ± 1.8% MKs intrasinusoidal, Fig. 4c,e) or to efficiently form transendothelial proplatelets in vivo (Fig. 4f,g, Supplementary Movie 6). Consequently, these animals were severely thrombocytopenic ( Supplementary Fig. 8b,c). Of note, intrasinusoidal localization of RhoA À / À MKs was not altered by the concomitant lack of Rac1 (RhoA/Rac1 À / À ) ( Supplementary  Fig. 9a-e), thus indicating a specific RhoA-Cdc42 crosstalk downstream of GPIba in this process.
RhoA controls MK polarization by limiting Cdc42 activity. We next investigated whether a possible crosstalk between Cdc42 and RhoA regulates MK polarization. Importantly, mature  Cdc42 À / À and Gp1ba-Tg BM MKs displayed defective DMS polarization after 4 days of liquid culture in vitro (Fig. 5a-c,f), whereas increased polarization of not fully mature DMS was observed in RhoA À / À MKs (Fig. 5d,g). Strikingly, in vitro treatment with p0p/B-Fab (anti-GPIb) markedly reduced DMS polarization in cultures of both wt and RhoA À / À BM MKs (Fig. 5e,g), suggesting that the regulatory role of GPIba in this process depends on intracytoplasmic GPIb-IX-V signalling and may operate in the absence of an ectopic ligand.
To decipher the underlying mechanism, we took advantage of lentiviral FRET biosensors derived from the Raichu probe to monitor Cdc42 activity in the polarized DMS 32 . Wt MKs showed a polarized Cdc42 activity at the DMS/F-actin complex, which was reduced by 68% in Gp1ba-Tg MKs (Fig. 6a,b,d), demonstrating the importance of controlled Cdc42 activation by GPIba for correct DMS polarization. Strikingly, polarized Cdc42 activity showed a 2.4-fold increase in the absence of RhoA (Fig. 6c,d), indicating that RhoA controls DMS and subsequently MK polarization by limiting Cdc42 activity. In order to determine how the nucleotide state of Cdc42 and RhoA influences this regulation, we utilized single amino acid mutants Cdc42/F28L and RhoA/F30L known to induce a constitutive exchange of GDP for GTP, thus resulting in hyperactive GTPases 38 . To restrict the expression of these mutants to the megakaryocytic lineage, we generated BM chimeric mice by transplantation of HSCs after lentiviral gene transfer, in which proteins were expressed under transcriptional control of the MK/platelet-specific human GP6 promoter (Fig. 7a,b) 39 . Of note, control mice transplanted with GFP-expressing HSCs showed reduced sinusoidal MK localization compared with non-irradiated mice (44.3±2.94%, Fig. 7c,f), indicating that irradiation influences MK localization. Compared to GFP-positive control cells, vessel association of Cdc42/F28L expressing MKs (GFP þ /GPIX þ ) was increased (57.0±9.0%, Fig. 7c,d,f)   of GPIba compared to normal RhoA À / À mice (dark grey) (n ¼ 9, 3 and 9). Untreated wt is shown in black (n ¼ 9 biological replicates). (e) Quantification of MK localization in the BM of RhoA/Cdc42 À / À mice reveals reduced intrasinusoidal localization compared to RhoA À / À mice (light grey) and MK clustering at sinusoids (n ¼ 4 and 9). Wt (black): n ¼ 10, Cdc42 À / À mice (dark grey); n ¼ 3. (f) Intravital two-photon microscopy of wt (black) and RhoA/Cdc42 À / À MKs (white) in the skull (n ¼ 7 and 4). Scale bar, 50 mm. (g) Quantification reveals reduced proplatelet-formation (ppf). Bar graphs represent mean ± s.d. (d,e) Two-way ANOVA with Bonferroni correction for multiple comparisons; (g) Unpaired two-tailed Student's t-test; *Po0.05; ***Po0.001 compared to wt; ### Po0.001 compared to RhoA À / À . located inside the BM sinusoids, whereas none were found in this compartment in control mice (Fig. 7c,d,f). In contrast, RhoA/F30L expression resulted in a further reduction of MKs in direct contact with BM sinusoids (30.4±6.4%, Fig. 7c,e,f) confirming the opposing functions of active Cdc42 (go-signal) and RhoA (stop-signal) in controlling MK localization and transendothelial proplatelet formation (Fig. 8).

Discussion
MKs are unique among haematopoietic progenitor cells in that they normally do not cross the endothelial lining of BM sinusoids, but extend and release portions of their cytoplasm into the blood stream by so far unknown mechanisms 4,5 . We now provide compelling evidence that this key step of unidirectional transendothelial proplatelet formation is tightly controlled  within MKs by a crosstalk of Cdc42 and RhoA and requires a functional GPIba ectodomain to occur efficiently (Fig. 8). Our data reveal a central antagonistic role of Cdc42 and RhoA in MK polarization and suggest that absence of RhoA supports locally increased Cdc42 activity and thereby directs the sites of cell protrusion. Our findings also highlight a critical function of the GPIba ectodomain for efficient MK polarization, Cdc42 activation and thus sinusoidal localization and transendothelial proplatelet formation of mature MKs. Thus, these data indicate that MK mislocalization within the BM contributes to the low platelet counts in BSS patients and may mechanistically explain the defective transendothelial (pro-) platelet formation by impaired DMS polarization and Cdc42 activation. Of note, vWF-deficient humans and mice display normal platelet counts 40 , and MK localization in the BM was not altered in Vwf À / À mice ( Supplementary Fig. 10a,b), which excludes a major role of this ligand in GPIba-dependent MK polarization.
Besides the vWF binding site, the p0p/B antibody Fab fragment also completely blocks thrombin binding to mouse GPIba ( Supplementary Fig. 10c), raising the possibility that this process might play a role in MK localization/polarization. However, MK localization in the BM of mice expressing human GPIba with a mutation (D277N) that abolishes a-thrombin binding 41 was similar to that of mice expressing wt hGPIba ( Supplementary  Fig. 10d,e), arguing against a role of thrombin in this process. Intriguingly, however, in vitro treatment of cultured MKs with p0p/B-Fab was sufficient to almost completely revert the DMS hyperpolarization observed in RhoA-deficient MKs. Thus, our findings do not indicate that GPIba requires binding of an ectopic ligand to control MK localization and polarization in the BM. Rather, our results support the hypothesis that GPIba-mediated regulation of MK polarization might be a cell intrinsic process, which can be modulated by altering GPIba signalling through changing the conformation and/or membrane localization/ clustering of the receptor in response to p0p/B-Fab binding. In line with this, enhanced endogenous binding of mutant vWF to GPIba in patients with vWF disease-type 2B results in giant platelets and thrombocytopenia and was reported to be associated with dysregulation of the LIM kinase/cofilin pathway in MKs, together with upregulated RhoA signalling 42 . Thus, altered GPIb-IX-V-mediated intracellular signalling can lead to abnormal cytoskeletal signalling and severely abnormal MK function.
Taken all observations together, our data support a model in which GPIba controls MK localization and transendothelial MK migration through intracellular signals, namely through the regulation of cell polarization via the Rho GTPases RhoA and Cdc42. While this process presumably involves MK PI3K function, we cannot rule out that treatment with the PI3K inhibitor wortmannin also exerts effects on vascular cells. Thus, deciphering the exact role of PI3K and/or PIs in MK localization requires further investigation. On the other hand, it is well known that Cdc42 and PI3K play a central role in establishing morphological and molecular cell polarity in eukaryotic cells 28,29 . In neurons, establishment of cell polarity during axon specification requires PI3K-dependent crosstalk of GTPases with aPKC isoforms which form part of the PAR complex 28 . Consistently, Cdc42-mediated cell polarity in migrating astrocytes occurs through activation of the aPKC isoform PKCz 28,43 . Thus, our observation of reduced sinusoidal localization of MKs in wortmannin-treated mice and mice deficient in the aPKC-isoform PKCi is supportive of a role of PI3K signalling for Cdc42/RhoA-controlled MK polarization and localization.
During the last years, experimental evidence has accumulated suggesting that MKs possess an intrinsic network involving the Our data from RhoA-deficient mice clearly exclude that size restrictions limit MK transmigration through BM sinusoids and emphasize RhoA as a critical regulator of proplatelet extension across the endothelial barrier by providing a 'stop' signal for MK polarization and thus transmigration (Fig. 8). In line with this, it has been described that excessive RhoA activation in BM-derived macrophages leads to decreased polarization, motility and cell spreading 50 . Of note, RhoA-dependent activation of Rho kinase (ROCK) and NMM-IIa was shown to negatively regulate proplatelet formation in MKs derived from CD34 þ cells in vitro 33 . Mutations in the gene encoding NMM-IIa, MYH9, affect platelet biogenesis in patients with MYH9-related disorders 51 . Interestingly, MKs from mice with MK-restricted NMM-IIa deficiency (Myh9 À / À ) exhibit increased proplatelet formation in vitro 52 and a significant number of MK nuclei was observed in lungs of Myh9 À / À animals compared to the wild-type 53 . However, given the multiple intrinsic defects present in Myh9 À / À MKs, further studies using different experimental approaches will be required to decipher the role of NMM-IIa in RhoA-induced MK polarization and platelet biogenesis.
In summary, the herein described Cdc42/RhoA regulatory circuit downstream of GPIb ensures a tight control of transendothelial platelet biogenesis in the BM and opens new avenues to treat both inherited and acquired forms of thrombocytopenia that are major causes of bleeding in man, and additionally to modulate in vitro production of platelets in the field of transfusion medicine.
Two-photon intravital microscopy of the BM. Mice were anaesthetized by intraperitoneal injection of medetomidine 0.5 mg g À 1 , midazolam 5 mg g À 1 and fentanyl 0.05 mg g À 1 body weight. A 1 cm incision was made along the midline to expose the frontoparietal skull, while carefully avoiding damage to the bone tissue. The mouse was placed on a customized metal stage equipped with a stereotactic holder to immobilize its head. BM vasculature was visualized by injection of tetramethylrhodamine dextran (8 mg g À 1 body weight, 2 MDa, Thermo Scientific). Platelets and MKs were antibody stained with anti-GPIX-AlexaFluor 488 (i.v. injection of 0.6 mg g À 1 body weight). Images were acquired with a fluorescence microscope equipped with a Â 20 water objective with a numerical aperture of 0.95 and a TriM Scope II multiphoton system (LaVision BioTec), controlled by ImSpector Pro-V380 software (LaVision BioTec). Emission was detected with HQ535/50-nm and ET605/70-nm filters. A tunable broad-band Ti:Sa laser (Chameleon, Coherent) was used at 760 nm to capture Alexa Fluor 488 and rhodamine dextran fluorescence. ImageJ software (NIH) was used to generate movies.
Histology. Three-micrometer-thick sections of formalin-fixed paraffin-embedded spleens and lungs were prepared, deparaffinized and stained with haematoxylin and eosin (MHS32 and 318906, Sigma-Aldrich). MKs were stained with HRP-labelled anti-GPIba antibodies after antigen retrieval and detected with 3-amino-9-ethylcarbazole substrate. MK number, morphology and localization were analyzed with an inverted Leica DMI 4,000 B microscope.
Analysis of MK polarization. For immunofluorescence, MKs in suspension were fixed and permeabilized in one step for 30 min in PBS with 3.7% formaldehyde and 0.05% Triton X-100. Samples were saturated in 1% fatty acid free BSA in PBS. Incubation with antibodies, fluorescent secondary antibodies, AlexaFluor-labelled phalloidin or DAPI was performed for 1 h at RT. For 3D imaging, cells were kept in m-slide ibiTreat chambers (Ibidi) in PBS. Confocal images were captured with a LSM780 operated with Zen software using a Â 63, 1.4 NA Plan Apochromatic objective lens (Carl Zeiss). Profiling of fluorescence intensity was done with ImageJ (NIH). For 3D-analysis, z-stacks were taken and processed with the Imaris 6.4.2 software (Bitplane AG).
FRET-based measurement of Cdc42 activity. The Raichu-Cdc42 encoding sequence was amplified by PCR using Raichu-1054X plasmid as a template and cloned in frame into the pTRIP-IRES lentiviral vector plasmid using BamHI and NheI restriction sites (underlined) 32 . The primers were 5 0 -CGCGGATCCTTGG CAAGAATTCGGCATGG-3 (forward) and 5 0 -CTAGCTAGCGGCAGAGGGAAA AAGATCCGTCGAC-3 0 (reverse). UT711oc1 cells were transduced by incubation with lentiviral particles at a multiplicity of infection (MOI) of 1 for 2 days. Transduction efficiency was checked by fluorescence and reached about 100% of the population. Transduction of primary MKs was performed on Lin À population at an MOI of 5 for 1 day, then 50 ng ml À 1 mTPO were added for 3 additional days. FRET efficiency was represented as the colour-coded ratio image of YFP/CFP after background subtraction using MetaMorph (Universal Imaging). Acquisitions were performed with a LSM780 confocal microscope piloted by the Zen software, using a À 63, 1.4 NA Plan Apochromatic lens (Carl Zeiss).
BM transplantation. 8-12-week-old female C57Bl/6 mice were preconditioned with 10 Gy irradiation and transplanted with 5 Â 10 5 cells. Cell infusion was performed via tail vein injection in a final volume of 150 ml. All mice were kept in the specified pathogen-free animal facilities of the Paul-Ehrlich-Institute, Langen, Germany.
Statistical analysis. When comparing two experimental groups, data distribution was analysed using the Shapiro-Wilk test. Where indicated, statistical significance between two experimental groups was analysed using an unpaired two-tailed Student's t-test. Otherwise, data were analysed using two-way analysis of variance (ANOVA) with Bonferroni correction for multiple comparisons (Prism 7; GraphPad Software). P-valueso0.05 were considered as statistically significant. *Po0.05; **Po0.01; ***Po0.001 or as otherwise stated. Data are presented as mean ± s.d.
Data availability. All data generated or analysed during this study are included in this published article (and its Supplementary Information files).