Proplatelet formation is selectively inhibited by collagen type I through Syk-independent GPVI signaling

Platelets are released from their immediate precursor cells, megakaryocytes, at the discontinuous endothelium of bone marrow sinusoids. Mature megakaryocytes are juxtaposed to the endothelial lining and its underlying basement membrane, which delineates the bone marrow cavity from the circulation. As shown by Lichtman and colleagues in 1978, processes extending from megakaryocytes, known as proplatelets, are continuously formed and protrude across the vascular barrier into the blood vessel lumen (Lichtman et al., 1978). Proplatelet formation (PPF) is a highly complex transformation process that has been studied in vitro (Italiano et al., 1999) and in vivo (Junt et al., 2007), revealing thrombopoietin (TPO, also known as THPO) as a pivotal cytokine that drives hematopoietic stem cells into the megakaryocytic lineage. The final step in platelet biogenesis, referred to as thrombopoiesis, however, is TPO independent as shown by functional studies (Choi et al., 1995) and a recent genetic study where c-Mpl, the TPO receptor, was deleted in the megakaryocytic lineage (Ng et al., 2014). In addition, TPO-independent mechanisms of platelet production have been described (Avecilla et al., 2004; Zheng et al., 2008). It is well known that structural proteins present within the bone marrow cavity and at the vascular niche have an impact on thrombopoiesis (Larson and Watson, 2006). Several in vitro systems have been established to study PPF on plated extracellular matrix (ECM) substrates that include the use of cord-blood-derived CD34+ cells (Balduini et al., 2008; Malara et al., 2011; Sabri et al., 2004) and murine megakaryocytes from bone marrow or fetal liver (Malara et al., 2014). These studies suggest that, although platelet biogenesis is a cell-autonomous process, collagen type I inhibits PPF by preventing premature platelet release in the bone marrow cavity, whereas collagen type IV and laminin support PPF at the sinusoids (Balduini et al., 2008). However, the mechanisms underlying these processes, including which collagen receptors mediate the activating and inhibitory signals and which laminin isoforms are involved, remain unclear.

The major collagen receptors expressed on platelets are the activatory collagen receptor glycoprotein (GP)VI (also known as GP6) and α2β1 integrin (Nieswandt and Watson, 2003). Although these receptors are expressed by megakaryocytes from relatively early stages in their maturation (Lagrue-Lak-Hal et al., 2001), their function on PPF is ill-defined. Moreover, other collagen receptors, such as leukocyte-associated immunoglobulin-like receptor-1 (LAIR1) (Steevels et al., 2010) and discoidin domain receptor 1 (DDR1) (Abbonante et al., 2013) have been proposed as inhibitory collagen receptors on early megakaryocytes, whose expression decreases during maturation and thus are absent from circulating platelets. Accordingly, once released from megakaryocytes, collagens are strong platelet agonists, with their binding leading to adhesion, aggregation and granule secretion, suggesting that collagen receptors in megakaryocytes are capable of transducing distinct signaling cascades.

Here, we identify GPVI as the pivotal collagen receptor required for directed PPF using a combination of genetic and pharmacological approaches in situ and in vitro. Our results imply that the inhibitory signal of collagen I is mediated by GPVI and dominant over the effects of collagen IV. Collagen IV, however, can displace collagen I at the vascular niche by stronger binding, diminishing the inhibitory signal and thus promoting PPF. Bone marrow laminins and fibronectin have no positive impact on PPF, which can thus be considered a cell-autonomous process that requires active inhibition rather than stimulatory signals.

To accurately characterize the cellular and ECM megakaryocyte niche in the bone marrow, we used a combination of in vivo and in vitro analyses. The most recent bone-sectioning methods were employed together with well-characterized antibodies to ECM proteins and cellular markers, and several genetic mouse models of thrombopoiesis. First, we adopted a system of automated imaging and alignment of images to allow detection of a complete femur, from proximal to distal epiphysis (Fig. S1Ai). DAPI staining was used to monitor tissue cellularity, and for detection of cutting artifacts or loss of cells during the staining and washing procedures, which might result in false low numbers. Femurs were divided into proximal epiphysis, diaphysis and distal epiphysis (Fig. S1Aii,iii,iv). Sinusoids were marked using anti-endoglin (CD105) and megakaryocytes by anti-CD41. Both stainings allowed a clear distinction between the bone cortex and marrow, which comprises the vasculature (Fig. S1A). Total megakaryocyte numbers were calculated per bone marrow area, revealing that megakaryocytes are equally distributed with a slight increase in density at the distal epiphysis (Fig. S1B). Within the diaphysis, megakaryocytes had a density of 43/mm². We found 60% in direct contact with sinusoids, although this value might be higher due to undetected sinusoids above or below the section plane. Megakaryocyte area ranged from 150 to 600 µm² (Fig. S1C) and was independent of their location in the bone, with many small megakaryocytes residing at the vessel. We found that the coefficient of determination between distance and megakaryocyte area was very low (R²=0.01196). This implies that megakaryocyte maturation might not only take place within the cavity, but also while residing at the sinusoids. To distinguish between sinusoids and arterioles, we used anti-smooth muscle actin (SMA) to mark arterioles only. Co-staining with both antibodies revealed that both vessel types were clearly distinct (Fig. S1D) but that megakaryocytes associated only with sinusoidal endothelium. The median distance between two sinusoids was 50 µm (Fig. S1E).

The bone marrow contains a variety of ECM proteins, mostly fibrillar collagens and fibronectin, although laminins have been described mainly during embryogenesis and in association with the basement membranes of the vasculature. To investigate whether megakaryocytes associated with specific ECM proteins, we performed triple staining using one ECM marker in addition to megakaryocyte and sinusoid staining. We excluded small CD41+ cells that might be hematopoietic stem cells, and compared CD41 with later markers (CD42b) and could not detect significant differences. Fibronectin was present throughout the bone marrow (Fig. 1A). von Willebrand factor (vWF) was expressed in megakaryocytes, where it is stored in α-granules (Fig. 1B). Interestingly, vWF was completely absent from sinusoidal endothelial cells or the subendothelial matrix, whereas fibrinogen was detectable in patches throughout the cavity (Fig. 1C). To define which laminin isoform was expressed, we used an antibody directed against Englebreth–Holm–Swarm (EHS) laminin 111, which recognizes laminin α1, β1 and γ1 chains equally (and is therefore a pan-laminin antibody as it would detect several isoforms), or antibodies that specifically recognize the isoform subunits α4 or α5 (Frieser et al., 1997; Sorokin et al., 1997). In our femur sections, we found staining around the larger vessels and the sinusoids, as expected (data not shown). Co-staining for SMA indicated that both laminin isoforms were present in ample amounts around arterioles. Although laminin α5 was readily detected around CD105-positive sinusoids (Fig. 1D,E,G), we found laminin α4 to be weakly expressed at capillary vessels (Fig. 1F,H). Trabecular bone and cortex were highly positive for collagen I, but negative for collagen IV (Fig. S2A–D). In order to corroborate these observations and to show that the different collagen types used for plating and antibodies used for detecting these types are specific overall, we performed an ELISA where we measured the binding of antibodies to the distinct collagen types or mixtures. Our result depicted in a heatmap (Fig. S2E) indicate that, despite substantial structural overlap between collagen types I and IV, the antibodies used in our study recognize their indicated substrates with the highest affinity, and, hence, we could perform co-stainings throughout the femur to map the detailed distribution of these matrix proteins.

Throughout the bone marrow cavity, we found thin filaments of collagen I, III or IV (Fig. 2A–H) that were partially colocalizing, suggesting a reticulated fiber network. Collagen III filaments were detected throughout the bone marrow, in direct contact with megakaryocytes (Fig. 2H), but absent from bone cortex (Fig. 2B). Collagen I staining also occurred around larger arterioles and arteries, probably in the outer fibrous layer, and subjacent to sinusoids (Fig. 2D). Collagen IV also localized to basement membranes of sinusoids (Fig. 2E). Megakaryocytes were in contact with collagen types I and IV, showing no obvious preference (Fig. 2F,G). We quantified the degree of megakaryocyte contact with ECM proteins in the sections by using four categories: no contact (0–5%), low contact (5–40%), intermediate contact (40–60%) and high contact (60–100%) as indicated (Fig. 2I). Most bone-marrow-associated megakaryocytes (without vessel association) had direct contact with collagen type I or IV. The degree of contact did not vary substantially for collagen I between cavity megakaryocytes and those found juxtaposed to endothelial cells. This was in strong contrast to collagen IV: whereas cavity megakaryocytes had a comparable association with type I, more than two thirds of all vessel megakaryocytes showed high contact with collagen type IV. As pan-laminin staining was exclusively detected in basement membranes, only vessel-associated megakaryocytes were in direct contact with laminin, albeit not completely surrounding or wrapping them.

Taken together, our results show that collagen types I, III and IV cannot be exclusively assigned to a specific compartment within the bone marrow cavity or the vascular niche. Within the bone marrow parenchyma megakaryocytes colocalize with all analyzed collagen types, but show increased colocalization (or overlap in staining) only with type IV at the vascular niche.

This observation prompted us to ask to what degree megakaryocytes adhere to different ECM proteins. We isolated fetal liver cells and differentiated them for 3 days in the presence of TPO prior to enrichment. Megakaryocytes were enriched through a one-step BSA density gradient on day 3 and seeded onto 24-well plates coated with either collagen types I, III, IV, laminin or fibronectin. The area of all CD41+ megakaryocytes was measured and calculated as the fraction of total area per well. Surface coverage was low on uncoated or bovine serum albumin (BSA)-coated control wells, but adhesion was high on collagen types I and III, and intermediate on type IV and fibronectin (Fig. 3A and 3A, insets). Only weak adhesion was detected on laminin-111. Megakaryocytes also started to spread on ECM proteins (especially collagen I and III; Fig. 3B), as their area was significantly enlarged when compared to BSA or uncoated surfaces.

We next asked whether this adhesion and spreading on distinct matrices would affect megakaryocyte maturation. After 54 h, megakaryocytes were removed from the substrates by gentle pipetting (which was controlled by light microscopy) and subjected to ploidy analysis. Surprisingly, all ECM molecules led to a shift towards lower ploidy levels, independently of whether it was typical for bone marrow stroma or a vascular niche protein, with laminin-111 causing the strongest reduction of ploidy levels (Fig. 3C). However, the proportions of each ploidy level were not statistically significant between the different matrix proteins. For size analysis, megakaryocytes were cultured on the ECM substrates for 6 h (day 3 of culture), 30 h (day 4) or 54 h (day 5), removed and cytospun onto slides. Megakaryocyte area was determined after β1-tubulin staining (Fig. 3D). Fibronectin did not have any effect, and collagen type I was the only factor that reduced megakaryocyte size significantly. All other ECM substrates led to increased size (Fig. 3E), suggesting that collagen IV and laminin contribute to the final megakaryocyte maturation.

The generation of proplatelets and platelets by fetal-liver-derived megakaryocytes has been studied extensively and is considered a system that partly mimics platelet formation in vivo (Italiano et al., 1999; Junt et al., 2007). We therefore adopted a standardized system to study the influence of ECM proteins on PPF in a time-dependent manner. Megakaryocytes were plated onto purified ECM proteins and PPF was evaluated 6 h later and twice on day 4 and 5 (Fig. 4A). As described above, we found that, on BSA, PPF increased linearly until day 5 before it peaked. Later time points could not be accurately determined due to massive shedding of proplatelets and platelets. PPF was markedly reduced (Fig. 4B, top row) when megakaryocytes were plated on collagen types I and III, with collagen type I being the stronger inhibitor.

It is worth mentioning that we still find PPF on collagen I, but that it was strongly reduced and attenuated (Fig. 4B, top row), and none of the basement-membrane-associated proteins (collagen IV, laminin and fibronectin) supported PPF beyond the level seen in the BSA control (Fig. 4B, bottom row and data not shown; the increase in PPF on fibronectin on day 4 was not statistically significant). As ‘inhibitory’ collagen I is also present at the vascular niche where PPF takes place, we sought to address whether mixing with a second, distinct vascular niche protein can diminish the inhibitory effect on PPF when mixed with collagen IV or whether the inhibition is dominant. Thus, we extended our system to study serial dilutions of collagen I with collagen IV or laminin isoforms by maintaining the total protein concentration at 10 µg/cm² surface, which is highly saturating for surface coating. When collagen IV was present at less than 50%, collagen-I-mediated inhibition of PPF was dominant. However, when collagen IV became the predominant protein, we found increased PPF, implying that collagen IV might counteract the collagen I signal and thus promote PPF at the vascular niche. This finding also suggests that both collagen types interact with the same receptors and compete for binding or signaling. By contrast, increasing concentrations of laminin by up to 87% could not abrogate the inhibitory effect by collagen I (Fig. 4C). Even a small fraction of 13% collagen type I in the mixture was sufficient to inhibit PPF to the same degree as pure collagen I.

Our co-stainings of collagen I and IV (Fig. 2F,G) indicate that bone marrow megakaryocytes are in contact with both types. Megakaryocytes express several collagen receptors, of which GPVI is considered the major receptor for platelet activation enabling firm adhesion and aggregate formation, whereas integrin α2β1 contributes to platelet adhesion, implying that distinct receptors are involved in collagen binding and transmission of inhibitory signals for PPF (Nieswandt and Watson, 2003). We therefore took advantage of transgenic mouse models in which either GPVI (Gp6^−/−) (Bender et al., 2011) or the α2 integrin subunit (Itga2^−/−) (Holtkotter et al., 2002) are not expressed and asked to what degree both receptors affect collagen-I-mediated inhibition of PPF. Megakaryocytes derived from Gp6^−/− mice showed normal PPF on collagen I whereas Itga2^−/− megakaryocytes maintained the PPF reduction. These results were replicated upon the use of the respective blocking antibodies (JAQ1 for GPVI, LEN/B for integrin α2) (Fig. 4D). When both receptors were blocked simultaneously, PPF inhibition was again abrogated, indicating that the dominant signaling response to collagen I is mediated by GPVI. Interestingly, the inhibitory effect of collagen I could be titrated by intermixing collagen I with type IV (Fig. 4E,F). The collagen GPO motif (single letter amino acid nomenclature, where O is hydroxyproline) comprises ∼10% of the amino acids in types I and III (Morton et al., 1995; Nieswandt and Watson, 2003), whereas it is less abundant in collagen IV (Cowan et al., 2000). Collagen-related peptide (CRP) concentrates these motifs by crosslinking peptides with a GKO-(GPO)₁₀-GKOG motif and is a strong platelet activator by inducing signaling through GPVI and the Syk–LAT signalosome (Pasquet et al., 1999). As JAQ1 blocks the CRP-binding site of GPVI (Schulte et al., 2001), we tested the effect of CRP and its mixture with collagen I on PPF. Interestingly, CRP had no effect on PPF and, when mixed with collagen I, it acted as collagen IV in competing with binding sites without transducing any inhibitory signal (Fig. 4G).

As laminin did not support PPF, but also did not dampen the collagen-I-mediated inhibitory signals, we mixed laminin with collagen I and plated collagen-receptor-deficient megakaryocytes (Fig. 4H) or wild-type (WT) megakaryocytes blocked with antibodies (Fig. 4I). Gp6^−/− megakaryocytes always had normal PPF, independently of matrix composition. In Itga2^−/− megakaryocytes or megakaryocytes treated with LEN/B, PPF remained reduced unless only a small fraction of collagen I was present, indicating that α2 integrin supports GPVI interactions at low collagen I concentrations. Of note, EHS laminin (111) might not reflect the situation within bone marrow. Thus, we additionally mixed recombinant laminin isoforms 211, 411, 421, 511 or 521 with collagen I in a 50:50 ratio and compared PPF at day 4. None of the laminin isoforms used was able to abrogate the inhibitory effect of collagen I (Fig. 4K). This included laminins 511 and 521 that we detected in the subendothelial basement membrane (Fig. 1E,G). These isoforms were able to induce weak platelet aggregation in light transmission aggregometry (data not shown) and platelet adhesion (Schaff et al., 2013) indicating that they are, in principle, functional.

Taken together, these findings indicate that PPF is a rather autonomous process that is selectively inhibited by interaction of collagen I with GPVI and not actively supported by other vascular ECM proteins. Collagen IV and CRP are able to counteract the inhibitory effect of collagen I in a dose-dependent manner. Apparently, when collagen I is present in small amounts, GPVI is not sufficient to mediate the full inhibitory signal and collagen I mediates its effect, in part, through α2β1 integrin. This implies that α2 integrin reacts toward collagen types I and IV with different impact on thrombopoiesis and suggests that interaction of collagen type I with GPVI is the major event in inhibition of PPF, whereas the interaction between collagen type IV and α2 integrin dampens this process.

Activation of GPVI in platelets typically leads to phosphorylation of the FcRγ chain by kinases of the Src family (SFKs), finally activating Syk and the LAT signalosome prior to phospholipase C (PLC)γ2 activation. We asked whether a lack of central proteins from this pathway could phenocopy the loss of PPF inhibition seen in GPVI-deficient megakaryocytes. Surprisingly, megakaryocytes lacking LAT or Syk showed reduced PPF when plated on collagen I, comparable to control megakaryocytes (Fig. 5A,B), indicating that this branch of the cascade is not required. We next applied the SFK inhibitor PP2 during the assay and measured PPF. Interestingly, PP2 could fully reverse the inhibition. Of note, addition of JAQ1 or plating megakaryocytes on CRP always resulted in full PPF (Fig. 5C). Finally, we analyzed whether Col I leads to phosphorylation of Syk. Megakaryocytes were plated on collagen I or CRP for the indicated time prior to lysis and subjected to immunoblot analysis. We did not detect phosphorylation of Syk on either coating. As a control, we used platelets treated with CRP under stirring conditions demonstrating that Syk becomes activated (Fig. 5D). However, when platelets are treated the same way as megakaryocytes and incubated on coated collagen I or CRP, we could only detect Syk phosphorylation on CRP (Fig. 5E). Taken together, our genetic, pharmacological and biochemical data imply that GPVI and its associated SFKs are essential for the inhibitory signal, whereas Syk and the LAT signalosome are dispensable for the inhibition of PPF.

The interplay of collagen type I and IV with their receptors GPVI and α2-integrin implies that the lack of one of these collagen receptors should affect megakaryopoiesis and platelet production in vivo. Surprisingly, platelet counts are overall normal in mice lacking GPVI (Xu et al., 2003) or integrin α2 (Holtkotter et al., 2002) (data not shown). However, megakaryocyte numbers in collagen-receptor-deficient mice could be elevated as a compensatory mechanism. We therefore analyzed megakaryocytes in femur sections from Gp6^−/− and Itga2^−/− mice and compared them with C57BL/6 controls. Sections were stained with anti-CD41 and either anti-GPVI or anti-α2-integrin antibodies and nuclei were visualized with DAPI. Fig. 6A confirms absence of GPVI on megakaryocytes in Gp6^−/− mice and absence of α2 integrin on Itga2^−/− megakaryocytes. Their density was normal in Gp6^−/− bones compared to controls, whereas in Itga2^−/− bones, the number of megakaryocytes doubled (Fig. 6B). Megakaryocyte area was slightly reduced in the absence of integrin α2, but remained unaffected in Gp6^−/− mice (Fig. 6C). The fraction of megakaryocytes in contact with CD105-positive vessels was similar in all strains with a slight increase in Gp6^−/− mice (Fig. 6D). Co-staining for collagen I or IV revealed that lack of either collagen receptor had little effect on association with collagen I, independent of whether megakaryocytes were present in the bone marrow stroma or were vessel-bound (Fig. 6E). By contrast, in mice lacking GPVI, we found more megakaryocytes in high contact with collagen IV and this did not change when megakaryocytes resided at the sinusoids. This finding is in strong contrast to both WT and Itga2^−/− animals, in which almost two thirds of vessel-associated megakaryocytes were embedded within a collagen type IV matrix. We finally asked to what degree blocking of collagen receptors would affect binding to the two collagen types. Megakaryocytes were seeded onto uncoated 96-well plates or plates coated with collagen type I, III, IV, laminin 111 or BSA. ATP content of adherent cells was measured by a luminometric assay as a correlating parameter for total cell numbers. As expected, we found strong adhesion on collagen I, which was overall unaffected when GPVI was blocked by JAQ1 antibody. By contrast, binding was reduced almost to control levels after blocking of α2 integrin by LEN/B, and when both antibodies were combined, megakaryocyte adhesion on collagen was the same as on BSA (Fig. 6F). This confirms that collagen type I binds more strongly to integrin α2, whereas the inhibitory signal for PPF is mediated by the rather low affinity binding (or adhesion) of collagen I to GPVI. Thus, fibrous collagen types I and IV compete for binding to both receptors, especially when in close proximity (Fig. 2F,G), so that the inhibitory signal by collagen I could get diminished at the sinusoids. In contrast, in our experimental setting laminins did not affect PPF at all, and blocking of either collagen receptor did not alter megakaryocyte binding to laminin.

The triple helical collagen molecule is one of the most complex matrix proteins, and is encoded by a variety of genes and exists in different compositions, mainly divided into majority types I, II, III that are present in bone, cartilage and tendons and minority collagens that often connect these types to other scaffolding proteins. Collagen type IV, in contrast, is mainly found in the basal lamina. Its fibers are overall less densely packed and form meshworks. All collagens are highly post-translationally modified with hydroxylation of proline and lysine residues being the most prominent alterations (Kadler et al., 2007). The use of standardized, pure collagen preparations for use in the laboratory has been hampered, as two different lots of a single preparation might display substantial functional differences. In addition, proteolytically digested, ‘soluble’ collagens, CRP (Knight et al., 1999) or collagen receptor agonists like convulxin are readily used for receptor activation in platelet research. Typically, the inter-species sequence differences of one collagen type are smaller than the differences between distinct collagen types (Fig. S2F). Specific isotype staining of whole femur sections was performed using a series of well-characterized antibodies. In contrast to a previous study where collagen I was exclusively found in cortical bone and collagen IV was only associated with vessels (Nilsson et al., 1998), we readily also detected both types in the bone marrow cavity (Fig. 2) in line with a recent study (Malara et al., 2014). The discrepancy between these two recent studies and the study of Nilsson might be attributed to different decalcification methods, as harsh acid treatment can destroy epitopes (Athanasou et al., 1987). Thus, our study extends these previous reports and provides a high-resolution colocalization analysis and mapping of bone marrow megakaryocytes in respect to distinct collagen filaments and vessels. In our hands, the collagen staining pattern does not allow delineation of distinct ‘niches’ with the exception that collagen IV is clearly enriched at the sinusoids. This is corroborated by the fact that wrapping of megakaryocytes by collagen I was not different between cavity and vessel, in contrast to type IV, which engulfs megakaryocytes predominantly at the vascular niche (Fig. 2I). Laminin has been considered a key protein to support PPF. Using isoform-specific antibodies, we only detected the α5 chain colocalizing with CD105 at sinusoids. EHS laminin or any other tested recombinant laminin (including the sinusoid-specific isoforms 511 and 521) had no effect on PPF beyond that seen with BSA and were unable to counteract the inhibition mediated by collagen I. It is worth mentioning that the recombinant laminins might not completely reflect the structural properties of the laminins present at the vascular niche. Fibronectin also did not support PPF beyond that seen with the BSA control and its distribution makes it unlikely that it can counteract collagen-I-mediated inhibition.

Several collagen receptors have been described that are spatiotemporally expressed in the megakaryocytic lineage. GPVI and integrin α2β1 are two well-characterized receptors on platelets (Nieswandt and Watson, 2003; Watson, 1999). They are regarded as late megakaryocyte lineage markers as compared to αIIbβ3 or GPIb, GPV and GPIX, which mark earlier stages of megakaryocyte maturation (Lagrue-Lak-Hal et al., 2001; Lepage et al., 2000). In addition, LAIR1 has been described as an early megakaryocyte inhibitory receptor for collagen, with reduced expression on mature megakaryocytes and no expression on platelets (Steevels et al., 2010). DDR1 has most recently been reported to be expressed on early megakaryocytes and to modulate their motility on collagen I (Abbonante et al., 2013) and, finally, GPV on platelets can also bind to collagen (Moog et al., 2001). Using blocking antibodies for collagen receptors on human megakaryocytes derived from cord blood CD34+ cells, Sabri and colleagues have shown that integrin α2β1 is essential for stress fiber formation by MAPK (ERK1/2) and Rho GTPase activation, whereas GPVI plays a role in maintaining ERK signaling. Interestingly, in their study, longer adhesion on the α2β1-specific collagen-mimetic peptide GFOGER led to inhibition of proplatelets (Sabri et al., 2004), which is partly in contrast to our findings (Fig. 4). However, the GFOGER peptide does not fully reflect the composition of collagen type IV (Knight et al., 1998) and difference between mouse and human cells, as well as experimental conditions like the presence of bivalent cations, might contribute to the different findings. In addition, the activation status of the α2-integrin plays an important role in thrombopoiesis (Zou et al., 2009), raising an important mode of feedback where GPVI activation increases the fraction of integrins being in the ‘on’ state. Using our system of mixing collagen I with type IV or laminin isoforms for analysis of PPF, we could unambiguously show that collagen I mediates the inhibitory signal by GPVI, as it was completely abrogated in Gp6^−/− mice or upon GPVI blockade (Fig. 4F), whereas adhesion of megakaryocytes to collagen I was rather mediated by α2β1 (Fig. 6F), implying that both receptors harbor distinct functions for collagen I binding and signaling. Platelet aggregation to fibrillar collagen is abolished in Gp6^−/− platelets, but only delayed when Itga2^−/− platelets are used (Holtkotter et al., 2002). Integrin α2β1 on vessel-associated megakaryocytes might bind collagen IV and thus prevent collagen I signaling to GPVI. We found that the inhibitory signal is at least partly transduced by α2β1. This raises the question as to whether it is the collagen type or rather the collagen receptor that defines the inhibitory signal. There are several lines of evidence that the receptors harbor different binding sites for distinct collagen epitopes (Leitinger, 2011; O'Connor et al., 2006; Schulte et al., 2001), which is corroborated by our results. This might also explain why Sabri et al. (2004) could not detect inhibition of PPF when treating megakaryocytes with convulxin. We presume that another binding site of collagen I, which is not mimicked by convulxin, can bind to GPVI and trigger the inhibitory signal. The high-affinity binding site can be blocked by JAQ1 making the receptor inaccessible for CRP binding, whereas high concentrations of fibrillar collagen I still can bind (Schulte et al., 2001; and data not shown). Intriguingly, CRP was not able to inhibit PPF and acted like BSA (Fig. 4G).

Megakaryocytes lacking Syk or LAT also displayed reduced PPF on collagen I, whereas inhibition of SFKs by PP2 treatment phenocopied megakaryocytes lacking GPVI, suggesting that the inhibitory signal bifurcates downstream of SFKs. A similar branch point has been identified for the targeted downregulation of GPVI through internalization or ectodomain shedding, respectively, providing an independent line of evidence that there are two distinct signaling pathways downstream of GPVI (Rabie et al., 2007). Moreover, the GPVI signaling cascade in megakaryocytes might differ from platelets, where ERK proteins becomes transiently phosphorylated when plated on collagen I, whereas in megakaryocytes plated on collagen, we found only weak and attenuated ERK phosphorylation (data not shown). In this study, we found that platelets plated on CRP could induce a weak phosphorylated Syk band that was not detectable in megakaryocytes (Fig. 5D). This could be explained by an inhibitor of Syk activation being present in megakaryocytes that becomes deactivated in platelets, where Syk activation is required for GPVI-dependent platelet activation. The branchpoint downstream of SFKs could indicate that the signaling complexes have a distinct composition to prevent premature GPVI signaling in megakaryocytes. These data together provide another line of evidence that Syk does not play a major role in GPVI-mediated inhibition of proplatelet formation by collagen.

Megakaryocytes plated on immobilized JAQ1 had normal PPF, indicating that the antibody itself cannot induce the inhibitory signaling (data not shown). Interestingly, mice lacking GPVI or integrin α2 have normal platelet counts and normal recovery rates after platelet depletion and do not release platelets prematurely into the marrow cavity (data not shown). Proplatelet formation has been shown to be cell autonomous (Lecine et al., 1998), which is also reflected in bone marrow explants (Eckly et al., 2012), where megakaryocytes immediately start to form proplatelets when they migrate out of the bone marrow microenvironment. The default setting for this process thus seems to be ‘on’. During their maturation within the bone marrow and when present at the endothelial barrier, premature proplatelet formation is suppressed by several ‘brakes’. Our own data clearly show that PPF on collagen type I can be reduced or attenuated, but not completely abolished (Fig. 4B). It is thus very likely that in vivo more than one inhibitory signal can act in parallel, which is not reflected by our in vitro approach. These additional ‘brakes’ would still be active in mice lacking GPVI, explaining the normal platelet counts.

In conclusion, this study reveals that the activatory platelet collagen receptor GPVI is the major receptor mediating the inhibitory effects of collagen I on megakaryocyte PPF through a Syk-independent pathway. The spatial expression pattern of collagen types I and IV at the bone marrow cavity and across sinusoids suggests a mechanism for how PPF might be orchestrated across the endothelial barrier. Our current model is shown in Fig. 7.

All animal studies were approved by the district government of Lower Franconia (Bezirksregierung Unterfranken). We used outbred strain CD1 or the inbred strain C57BL/6 as well as the transgenic strains Gp6^−/− (Bender et al., 2011), Itga2^−/− (Holtkotter et al., 2002), Lat^−/− (Zhang et al., 1999) and Syk^{fl/fl, Pf4−Cre+/−} (Lorenz et al., 2015) (all C57BL/6 background) as indicated. Male and female mice of 6 to 12 weeks were used in equal numbers for all experiments, unless stated otherwise.

Fetal liver cells were differentiated as recently described (Kunert et al., 2009). Megakaryocytes were preincubated with antibodies against GPVI (JAQ1; Emfret Analytics) or anti-α2-integrin (LEN/B; Shida et al., 2014) at 10 μg/ml (10 min, 37°C) prior to seeding onto coated culture surfaces. Plates were coated with 10 μg/cm² human collagen type I, type III or fibronectin (Becton Dickinson), murine collagen type IV (Becton Dickinson or Trevigen), EHS-derived laminin (Merck Millipore) or human laminin 211, 411, 421, 511 or 521 (BioLamina) according to manufacturers' instructions, or with 7.5 µg/cm² BSA. For adhesion assays, megakaryocytes were seeded onto coated coverslips (1.4×10⁵ cells/cm²) or into 96-well plates (6.3×10⁴ cells/cm²) and allowed to adhere in the presence of 300 ng/ml CXCL12 (R&D Systems). After 4 h, cells were stained directly on the matrices, and megakaryocytes detected by use of anti-CD41 conjugated to FITC (Becton Dickinson) or by measuring intracellular ATP using the CellTiter-Glo^® kit (Promega) with a Fluoroskan Ascent™ FL luminometer (Thermo Scientific, Waltham, MA). Each experiment was performed in triplicate, unless indicated otherwise. To determine megakaryocyte maturation, megakaryocytes were plated on coated surfaces at a density of 30,000 cells/cm². After 6, 30 and 54 h, cytospins were performed, and cells were stained for β1-tubulin and DAPI. The area covered by at least 100 individual megakaryocytes was quantified for each group. Ploidy was determined as described previously (Kunert et al., 2009). The fraction of megakaryocytes that undergo proplatelet formation was determined by analyzing 80 to 120 cells per well at the indicated time points. Immunoblotting with antibodies was performed as described previously (Cherpokova et al., 2015); densitometric analysis was performed using ImageJ (http://rsb.info.nih.gov/ij).

Femurs of mice aged 6–12 weeks were sectioned as described previously (Kawamoto, 2003; Pleines et al., 2012) and megakaryocytes stained with anti-CD41 antibody. CD105 was used as an endothelial cell marker (eBioscience). Additional stainings were performed using antibodies against SMA (Merck Millipore, ASM-1, 1:50), fibrinogen (Dako F0111, 1:50), or vWF (Dako A0082, 1:50), collagen IV (Merck Millipore AB769, 1:50), collagen type I (ab21286) or III (ab7778, both Abcam, 1:50), laminin α4 (clone 377b, 1:100) (Ringelmann et al., 1999) and α5 chains (clone 504, undiluted) (Sorokin et al., 1997), GPVI (JAQ1), and integrin α2 (Sam.G4, both Emfret Analytics, 1:100). Corresponding secondary antibodies detecting IgG of rat, goat, rabbit or mouse were purchased as conjugates with Alexa Fluor 488 (A-11034), 594 (A-11007) or 647 (A-21247, A-21469, A-21244), respectively, by Life Technologies, all 1:300. All sections were also stained for DAPI to visualize all nuclei.

Antibody specificity was determined by ELISA dot plots, incubating slides with isotype-matched antibodies and cross-reactivity excluded by single staining or using sections from knockout mice. Slides were mounted in DAPI-containing medium (Southern Biotech). Megakaryocyte number and density was measured in every visual field. Three femur sections from three separate animals were examined, and 20 megakaryocytes per section were quantified with regards to area, shortest distance to sinusoids and degree of ECM contact, which was classified as: none (<5%), low (5 to 40%), intermediate (40 to 60%) and high (>60%).

All images were taken using a confocal laser-scanning microscope (Nikon A1, Chiyoda, Japan or TCS SP5, Leica Microsystems CMS, Wetzlar, Germany) with 200-, 400- or 600-fold magnification and CFI Plan Apochromat VC lenses (NA 0.75, 1.25 or 1.4) and NIS Elements AR or LAS AF software for image documentation and analysis.

Protein expression in platelets and megakaryocytes was studied by immuoblotting according to recently described protocols (Cherpokova et al., 2015; Kunert et al., 2009).

All data were subjected to Shapiro–Wilk tests for determination of their distribution curve. Significant differences of pairwise comparison of means against a specified control were determined using ANOVA, Student's t-test or a nonparametric Mann–Whitney or Kruskal–Wallis rank sum test. Differences between cell size distributions of different groups were compared by Kolmogorov–Smirnov test and frequency distributions by a χ2-test. Linear regression is indicated by the coefficient of determination (R²). Differences with P-values lower than 0.05 or 0.01 were defined as significant.

We would like to thank Lydia Sorokin for providing crucial reagents and critical comments on the manuscript and Elizabeth Haining for comments on the manuscript. We are also thankful to the Microscopy platform of the Bioimaging Center (Rudolf Virchow Center) for providing technical infrastructure.