Mechanical forces couple bone matrix mineralization with inhibition of angiogenesis to limit adolescent bone growth

Bone growth requires a specialised, highly angiogenic blood vessel subtype, so-called type H vessels, which pave the way for osteoblasts surrounding these vessels. At the end of adolescence, type H vessels differentiate into quiescent type L endothelium lacking the capacity to promote bone growth. Until now, the signals that switch off type H vessel identity and thus limit adolescent bone growth have remained ill defined. Here we show that mechanical forces, associated with increased body weight at the end of adolescence, trigger the mechanoreceptor PIEZO1 and thereby mediate enhanced production of the kinase FAM20C in osteoblasts. FAM20C, the major kinase of the secreted phosphoproteome, phosphorylates dentin matrix protein 1, previously identified as a key factor in bone mineralization. Thereupon, dentin matrix protein 1 is secreted from osteoblasts in a burst-like manner. Extracellular dentin matrix protein 1 inhibits vascular endothelial growth factor signalling by preventing phosphorylation of vascular endothelial growth factor receptor 2. Hence, secreted dentin matrix protein 1 transforms type H vessels into type L to limit bone growth activity and enhance bone mineralization. The discovered mechanism may suggest new options for the treatment of diseases characterised by aberrant activity of bone and vessels such as osteoarthritis, osteoporosis and osteosarcoma.

P roper skeleton development requires a fine-tuned crosstalk between different types of cells and tissues present in bone 1 . Juvenile bones harbour highly angiogenic type H vessels, which are closely associated with and coordinate the differentiation of bone growth-mediating osteoblasts. In contrast, quiescent type L vessels are present in the adult bone [2][3][4] . Type H vessels are located exclusively at the sites of active bone growth, namely the ossification front (OF) and periosteum, and they are organised in columnar structure 2 . Notch signalling, Hypoxia-inducible factor 1-alpha (HIF1α) activity, blood flow speed and slit guidance ligand 3 (SLIT3) were shown to support type H vessel formation 3,5,6 . However, the mechanism that directs the switch from type H to type L vessels resulting in bone growth cessation has remained unknown. A recent study reported that endothelial cells in bone require integrins for their proper functioning and maintenance, which highlights extracellular matrix (ECM) proteins as important factors for endothelium activity 7 . Yet, little is known about further ECM proteins controlling endothelial cells in bone. Therefore, we analysed changes in the composition of ECM proteins throughout bone development.
In this work we describe the molecular mechanism that links mechanical forces and angiogenesis in bone via mechanically triggered accumulation of the extracellular matrix protein dentin matrix protein 1 (DMP1) and its regulation of vascular subtypes.

Results
DMP1 is a potential candidate controlling endothelial cell fate in bone. To dissect age-related changes in vascular and perivascular microenvironments during and after active bone growth, we performed laser microdissection (LMD) of single capillary with associated surrounding cells in the OF of juvenile (4-week-old) and adult (12-week-old) mice. We developed a protocol to cut single capillary from undecalcified and unfixed bones of endothelial cell-specific GFP-reporter mice with a high RNA preservation rate to subsequently perform next-generation sequencing (NGS) ( Supplementary Fig. 1a, b). Principal component analysis showed that the cells from the juvenile OF were very similar between different mice, whereas with ageing, greater heterogeneity was apparent ( Supplementary Fig. 1c). To identify candidates from the list of differentially expressed genes with a potential of regulating type H vasculature, we focused on ECM proteins (Fig. 1a, b and Supplementary Fig. 1d). DMP1 appeared to be a promising candidate because of its reported angiogenesismodulating activity in a retina tumour model 8 .
DMP1 is known as an essential player in matrix mineralisation 9,10 . Our detailed analysis of DMP1 protein localisation throughout postnatal bone development showed that until 5 weeks of age, the protein was mainly localised at the base of the zone of type H vessels in the OF (Fig. 1c, d). Unexpectedly, at the age of 5.5-6 weeks, a burst in DMP1 secretion into ECM occurred primarily in the central top part of the OF also reaching the growth plate (GP) (Fig. 1c, e-g and Supplementary Fig. 2a). We performed quantifications of protein staining intensities within central and peripheral regions of interest in the OF (Fig. 1f, g and Supplementary Fig. 2b). During late postnatal bone maturation with increasing age, the DMP1-occupied area spread from the centre towards the periphery of the OF. Finally, by the age of 12 weeks, the whole metaphysis together with the GP were interspersed with DMP1 (Fig. 1c, f, g). This behaviour was seen irrespective of sex.
To identify the main cellular source of DMP1 in the OF-GP area at different ages, we established a method for RNAScope in bone. Co-hybridisation was performed with probes for Dmp1 together with Runx2 for early osteoblasts, osterix (Osx or Sp7) for mature osteoblasts and sclerostin (Sost) for osteocytes ( Supplementary Fig. 2c). Sclerostin expression was not found in the OF but only in the cortical bone. At all analysed stages in the OF-GP region, Dmp1 mRNA was only detected in the OF area but never in the chondrocytes of the GP, including time points directly before (5 weeks) and during the burst in DMP1 protein secretion (5.5 weeks) ( Fig. 1h and Supplementary Fig. 2f). Dmp1 mRNA was mainly found in Osx + cells and to a lesser extent in Runx2 + cells at all examined time points, which identifies osteoblasts as the main producer of DMP1 in the metaphysis ( Supplementary Fig. 2d, e). The number and intensity of Dmp1expressing cells peaked at 4-5 weeks (Fig. 1i, j and Supplementary  Fig. 2g). Yet, there was no difference at the level of Dmp1 gene expression between the central and peripheral parts of the OF, indicating that the observed differences in DMP1 amounts between these regions emerged rather at the protein level (cf. Fig. 1c, f, g). This observation goes in line with previous reports that the control of DMP1 availability and function mainly takes place at the posttranslational rather than transcriptional level [11][12][13] explaining the discrepancy in the timing of mRNA and protein peaks (Fig. 1c, f, g, i, j). Together, these data demonstrate that the burst in DMP1 secretion from osteoblasts in the OF centre at the age of 5.5-6 weeks is controlled at the protein level and DMP1 likely diffuses from the OF to the GP (Fig. 1k).
DMP1 transforms type H vessels into type L. In addition to secreting osteogenic factors 3 , tip cells of type H vessels secrete matrix metalloproteinase 9 (MMP9) that digests cartilage matrix and allows vessel bulges to invade GP cartilage, thereby facilitating bone growth 4 . We noticed that DMP1 accumulation in the OF centre at 5.5-6 weeks correlated with a decrease in endomucin (EMCN) intensity and reduced amounts of MMP9 at the OF-GP border. This was associated with the disappearance of bulges and columnar structure of the vessels (Fig. 2a, b and Supplementary  Fig. 3). Hence, all main features of type H vessels were downregulated. Importantly, at 5.5-6 weeks, these changes occurred only in the central part of the OF.
To study whether DMP1 directly regulates bone angiogenesis and growth, we generated mice with a constitutive and an inducible deletion of Dmp1 ( Supplementary Fig. 4). Constitutive Dmp1-deficient mice were smaller with shortened and widened long bones (Supplementary Fig. 4c-f) as had been described before 9,14 , and they featured a disorganized OF and GP ( Supplementary Fig. 5a). The blood vessels in the OF were dilated and had very high EMCN and MMP9 intensity, even in adulthood, though the columnar structure was lost already in juvenile bones ( Supplementary Fig. 5c, d). This finding suggests that, presumably via its matrix mineralisation activity, DMP1 provides a scaffold for columns of type H vessels. The phenotype of Dmp1 +/− mice was identical to Dmp1 +/+ mice, which indicates that one Dmp1 allele generates sufficient DMP1 protein ( Supplementary Figs. 4c, d and 5b). This result is consistent with human genetic studies where it was shown that mutations in the DMP1 gene cause autosomal recessive hypophosphatemia 15 . OSX + and RUNX2 + osteoblasts were present in the OF to a similar extent as in the wild-type group indicating that DMP1 is not required for osteoblast differentiation and maturation 16 ( Supplementary Fig. 5a, b). Notably, the GP of Dmp1 -/mice exhibited an expanded pre-hypertrophic and hypertrophic zone (RUNX2 + and SOX9 + ) ( Supplementary Fig. 5a). We assume that this phenotype is caused by the lack of proper vessel invasion due to the loss of vessel architecture.
To test whether DMP1 induces the switch from type H to type L vessels, we generated a Dmp1-floxed mouse line and crossed it with Col1-CreERT2 mice to ablate Dmp1 in an inducible manner selectively in osteoblasts (Dmp1 iΔOB ) upon tamoxifen injection (Supplementary Fig. 4g-i). Deletion of Dmp1 was performed at P28-32 to allow normal formation of the OF together with the columnar structure of blood vessels (Fig. 2c). Tamoxifen treatment led to~75% reduction of Dmp1 mRNA expression by week 6 (P42) ( Supplementary Fig. 6a, b) and reduced the amount of DMP1 protein in the OF centre by~70% (Fig. 2d and Supplementary Fig. 6c). The body weight was not affected ( Supplementary Fig. 6d). Yet, long bones of Dmp1 iΔOB mice featured an enlarged metaphysis with overall increased EMCN and MMP9 intensities (Fig. 2e, f and Supplementary Fig. 6e). In line with the femur phenotype of constitutive Dmp1-deficient mice (cf. Supplementary Fig. 4d, f), also Dmp1 iΔOB femurs were slightly shorter than control femurs by the age of 6 weeks, presumably reflecting the key role of DMP1 in matrix   Fig. 6f). Nevertheless, detailed analysis showed that the deletion of Dmp1 in osteoblasts arrested the age-related maturation of the OF. Specifically, type H vessels together with bulges were preserved in the central part of the OF in Dmp1 iΔOB mice (Fig. 2g, h). Moreover, the difference in EMCN and MMP9 intensity between the central and peripheral parts of the OF was absent in Dmp1 iΔOB femurs (Fig. 2i, k). The overall EMCN intensity and MMP9 amount in the OF centre were higher in Dmp1 iΔOB samples compared with the control group (Fig. 2j, l). As expected, von Kossa staining showed reduced mineralisation of the OF in Dmp1 iΔOB femurs, confirming the crucial importance of DMP1 for bone calcification ( Supplementary Fig. 6g, h, j).
In line with the phenotype of Dmp1 iΔOB mice at the OF, analysis of subchondral bone revealed that Dmp1 deletion at 4 weeks led to preserved density of EMCN + vessels and high amounts of MMP9 around the vessel tips underneath the articular cartilage by week 6 ( Fig. 2m-o). Control mice featured significantly reduced vessel activity in subchondral bone, whereas Dmp1 iΔOB samples were very similar to 4-week-old wild-type subchondral bone. Thus, deletion of Dmp1 at 4 weeks prevented the development-associated switch from active to quiescent vessel fate not only at the OF but also in subchondral bone. The mineralisation of subchondral bone was also reduced in Dmp1 iΔOB femurs ( Supplementary Fig. 6g, i, k). Hence, we cannot formally exclude that the reduced mineralisation contributes to some extent to the observed phenotype in Dmp1 iΔOB bones. Nevertheless, our results suggest that DMP1 is involved in the inhibition of angiogenesis in long bones and in the transformation of active type H into quiescent type L vessels, which results in reduced EMCN and MMP9 expression and loss of vessel bulges.
DMP1 inhibits VEGF signalling in the ossification front in vivo and in endothelial cells in vitro. Vascular endothelial growth factor (VEGF) signalling is crucial for angiogenesis and bone growth [17][18][19] . The proliferation of the chondrocytes in the proliferative layer of the GP allows the bone to elongate, whereas the hypertrophic chondrocytes in the bottom layer express high amounts of VEGFA, which drives the invasion of type H vessels into the GP, thus leading to the progression of the ossification front 17,18 . To test whether DMP1 has VEGF-inhibiting properties in bone, we studied phosphorylation of the main VEGF receptor VEGFR2 (pVEGFR2) along with DMP1 accumulation in ECM. Local DMP1 amounts negatively correlated with the amount of pVEGFR2 ( Fig. 3a and Supplementary Fig. 7). The VEGFR2 phosphorylation level in 4-to 5-week-old mice was equal across the OF. However, at 5.5-6 weeks we observed a marked reduction in the amount of pVEGFR2 in the central part of the OF in wildtype mice. This difference was largely lost in Dmp1 iΔOB mice, and the overall intensity of pVEGFR2 in the OF centre was higher ( Fig. 3b-d). Of note, the total level of VEGFR2 protein was not affected by Dmp1 deletion (Fig. 3e, g).
Previous studies showed that VEGFR2 activation induces Flt4 (VEGFR3) expression in endothelial cells 20,21 . VEGFR3 is usually highly upregulated in endothelial tip cells, where it is required for vessel sprouting 7,20,22 . Bulges in the GP of Dmp1 iΔOB mice stained strongly with VEGFR3 compared with remaining bulges of control mice (Fig. 3f, h). This result suggests that osteoblastreleased DMP1 downregulates VEGFR3 in endothelial tip cells most likely by interference with VEGFR2 signalling, thus leading to the loss of vessel bulges. Further support came from the finding that treatment of the capillary endothelial cell line bEnd.3 with DMP1 abrogated the VEGF-induced phosphorylation of VEGFR2 and the expression of Fos (c-fos), a target gene of VEGF signalling 23 (Fig. 3i-k). In summary, these data demonstrate that in the OF in vivo and in endothelial cells in vitro, DMP1 inhibits VEGF signalling, which is a known driver of angiogenesis and osteogenesis 19 . Taken together with previously published results 8 , we suggest that the anti-angiogenic effect of DMP1 likely is not restricted to type H vessels in bone but it rather appears to act as a general inhibitor of angiogenesis at those sites where DMP1 is present.
DMP1 secretion correlates positively with body weight and coincides spatiotemporally with FAM20C upregulation. Hormones, especially sex hormones, are usually considered important regulators of bone growth and metabolism [24][25][26] . Mice achieve puberty by approximately week 6, which coincides in time with the burst in DMP1 secretion. We noticed that the extent of DMP1 release at 5.5-6 weeks varied considerably between individual mice (cf. Fig. 1f). We also noted that in mice with lower body weight than average, the burst in DMP1 secretion had not yet taken place at 5.5-6 weeks. Inversely, littermates heavier than average featured already very advanced DMP1 release at this time point (Fig. 4a). Indeed, we found a positive correlation between body weight and the extent of DMP1 secretion (Fig. 4b). DMP1 is a highly phosphorylated protein and a substrate of FAM20C kinase that exclusively mediates the phosphorylation and subsequent secretion of DMP1 into ECM in vivo 11,27,28 . Notably, in 6-week-old mice, we found more FAM20C, along with higher DMP1 accumulation, in the OF centre in comparison to the periphery, which was not the case in 4-week-old mice (Fig. 4c-e). In line with FAM20C acting upstream of DMP1, inducible Dmp1 deletion did not impair FAM20C upregulation at the age of 6 weeks ( Supplementary Fig. 8a, b). These findings suggest that mechanical loading through increasing body weight or/and muscle contractions, directly or indirectly, controls DMP1 secretion along with FAM20C upregulation.
Mechanical loading enhances FAM20C kinase production and DMP1 secretion. To investigate whether FAM20C production was enhanced by mechanical loading, we cultured live 400-500-µm-thick slices of 4-week-old femurs and centrifuged them twice daily on 5 consecutive days (Fig. 5a). Loading via centrifugation caused an upregulation of intracellular FAM20C protein and enhanced DMP1 secretion (Fig. 5b, c and Supplementary Fig. 8c, d). The mechanically induced FAM20C upregulation apparently occurred hormone-independently in the cultured femur slices in vitro.
To study if body weight-and/or muscle force-associated mechanical loading controls FAM20C production also in vivo, we unloaded one hind limb in the mouse by cutting the sciatic and femoral nerves at the age of 4.5 weeks (Fig. 5d). This results in paralysis of the affected hind limb, hence releasing it from loading through muscle contraction as well as body weight with all weight going through the contralateral (normal) side, whereas both limbs were exposed to the same hormonal environment. Unloading by double neurectomy prevented FAM20C upregulation and DMP1 secretion at 5.5-6 weeks in almost all of the operated limbs, whilst preserving normal DMP1 release on the contralateral side (Fig. 5e, f and Supplementary Fig. 8e, f). Thus, mechanical loading enhances local FAM20C production and DMP1 secretion.
Mechanical forces trigger PIEZO1 to enhance FAM20C kinase production and DMP1 secretion. In our transcriptome analysis, we found four mechanoreceptors that osteoblasts could potentially express in the OF during late postnatal development: Trpv4, Kcnk2, Piezo1, Piezo2 (Supplementary Fig. 8g). Recently published studies show that in osteoblasts, mainly PIEZO1 and to a lesser extent PIEZO2 are key mechanosensors mediating anabolic effects of loading on bone [29][30][31][32][33] . Thus, we tested whether PIEZO1 or PIEZO2 control Fam20c upregulation in osteoblasts by analysis of bones of 6-7-week-old mice with Piezo1 or Piezo2 deficiency in Dmp1-expressing cells (Piezo1 ΔDmp1 or Pie-zo2 ΔDmp1 ). FAM20C upregulation and the resulting burst in DMP1 secretion were lost exclusively in Piezo1 ΔDmp1 bones but normal in Piezo2 ΔDmp1 femurs (Fig. 6a-c, d-f and Supplementary  Fig. 8h, i). Osteoblast maturation and abundance were not affected by Piezo1 deletion both at 4 and 7 weeks, as determined by the staining intensity and frequency of OSX + cells in the OF ( Supplementary Fig. 8j-n). 4-week-old Piezo1 ΔDmp1 bones featured similar levels of FAM20C and DMP1 as control bones indicating that PIEZO1 mediates the loading-induced upregulation of FAM20C around periadolescence, whereas basal FAM20C expression is PIEZO1 independent (Fig. 6g-i). In addition, we compared the phenotype of Piezo1 ΔDmp1 bones to control ones using von Kossa/van Gieson staining followed by histomorphometry at 7 weeks of age. Both assessments did not show significant differences between the groups at this age ( Supplementary Fig. 8o, p). These data suggest that the enhanced FAM20C upregulation and DMP1 secretion in control bones versus Piezo1 ΔDmp1 bones is likely to be mediated directly by PIEZO1 and not by differences in bone maturation between the two groups. To provide additional evidence and validate the involvement of PIEZO1 in the proposed mechanism, we exposed femur slices from 4-week-old Piezo1 ΔDmp1 and Cre-negative (WT) littermate mice to loading via low-speed centrifugation as described (cf. Fig. 5a). Mechanical loading did not lead to upregulation of FAM20C or enhanced DMP1 release in Piezo1 ΔDmp1 femur slices, while it did so in Cre-negative (WT) samples ( Fig. 6j and Supplementary Fig. 8q). To conclude, mechanical forces trigger PIEZO1 to enhance FAM20C expression in osteoblasts and subsequent DMP1 secretion into ECM during periadolescence.

Discussion
In this study, we identify a two-step molecular mechanism by which mechanical forces drive the maturation of bone and vessels at the end of adolescence (Fig. 7). First, body weight-associated mechanical loading triggers the mechanoreceptor PIEZO1 to enhance the production of FAM20C kinase in osteoblasts, which induces a burst in DMP1 secretion into ECM. Second, large Pooled data from three independent experiments. Mean ± SD. b Quantification of pVEGFR2 amount in the centre relative to periphery in control and Dmp1 iΔOB OF. Two-tailed Wilcoxon matched-pairs signed-rank test. c, Quantification of the total central pVEGFR2 amount in control and Dmp1 iΔOB group. Two-tailed unpaired t-test. b, c Pooled data from three independent experiments. Control n = 9, Dmp1 iΔOB n = 11. Age: 6 weeks. Mean ± SD. d pVEGFR2 immunostaining showing no decrease of pVEGFR2 in the OF centre of Dmp1 iΔOB femurs in contrast to control femurs at 6 weeks of age. Scale bar 25 µm. e Quantification of VEGFR2 intensity in EMCN + endothelial cells in the OF of control and Dmp1 iΔOB mice. Control n = 4, Dmp1 iΔOB n = 4. Pooled data from two independent experiments. Two-tailed Mann-Whitney test. Median ± range. f Quantification of VEGFR3 intensity in the OF of control and Dmp1 iΔOB mice. Control n = 6, Dmp1 iΔOB n = 9. Pooled data from three independent experiments. Two-tailed Mann-Whitney test. Median ± range. g, h VEGFR2 and EMCN (g) and VEGFR3 and EMCN (h) immunostaining of the OF centre in 6-week-old control and Dmp1 iΔOB mice. Scale bar 50 µm. i pVEGFR2 immunostaining of bEnd.3 cells, which were either treated with VEGF only (left) or pre-treated with DMP1 (40 nM) for 3 h before adding VEGF (50 ng/ml). Scale bar 25 µm. j Quantification of the pVEGFR2 intensity per cell of bEnd.3 cells, which were either treated with VEGF only or pre-treated with DMP1 before adding VEGF. Two-tailed Mann-Whitney test. Mean. k qPCR of c-fos expression, one of the target genes of VEGF signalling, in bEnd.3 cells, which were treated either with DMP1 only, or with VEGF only, or first with DMP1 and then with VEGF. Tukey's multiple comparisons test. Mean + SD. i-k Representative plots of 2 independent experiments. a-c, e, f, j, k Source data are provided in Source Data file.   5 Mechanical loading enhances FAM20C kinase production and DMP1 secretion. a Schematic depiction of the position of a femur slice (≈500 µm thick) derived from 4-week-old mice in a well of a 96-well PCR plate filled with cell culture medium. The slices were either just cultured for 5 days (control) or exposed to loading via low-speed centrifugation (30 min. at 19 g) twice per day (loaded). b Quantification of FAM20C and DMP1 intensities in the loaded (centrifuged) samples relative to control (non-centrifuged) group (red line). Pooled data from two independent experiments. Mean, n = 5-6. P values for paired analysis of samples derived from one mouse. Two-tailed paired t-test. c FAM20C and DMP1 immunostaining of control and loaded (centrifuged) femur slices. Scale bar 100 µm. amounts of extracellular DMP1 inhibit VEGF signalling in the OF and transform highly angiogenic type H vessels into quiescent type L vasculature to limit bone growth activity. In parallel, extracellular DMP1 leads to rapid matrix mineralisation and strengthening of long bones 9,34 . Thus, the role of DMP1 in bone is multifaceted: it combines active inhibition of angiogenesis with mineralising activity. The discovered link between mechanical loading and FAM20C upregulation, a major kinase of the bone phosphoproteome 27,28 , provides a mechanistic explanation for the well-known positive effect of loading on bone mineralisation 35 . Furthermore, it may offer an additional mechanistic explanation for hormone-independent osteoporosis caused by lack of loading, for example in astronauts 36,37 or after long immobilisation of a fractured bone 38 .
Growth-promoting type H vessels have received much attention in the last years because of their substantial clinical potential 6,39 . Our finding that inducible Dmp1 deletion led to increased vessel activity also in subchondral bone suggests that extracellular DMP1 may protect articular cartilage from vessel invasion, which accompanies neoinnervation of the tissue and sensation of pain in osteoarthritis patients 40,41 . Moreover, DMP1mediated inhibition of angiogenesis in bone may be useful in arresting growth of tumours including osteo-and chondrosarcomas [42][43][44] . In future studies, we will address the precise molecular mechanism of how extracellular DMP1 inhibits VEGF signalling in type H endothelium. Particularly we will investigate if the inhibition of VEGFR3 expression is mediated directly by DMP1 or if it is an indirect effect via the inhibition of VEGFR2 signalling 21 .
Under the influence of systemic hormones, bones of male mice acquire higher mass and cortical thickness than females, which is already evident at early puberty 45 (3-5 weeks). As the weight of male mice is also greater than females at this age it is likely that the load perceived by individual osteoblasts in male and female bones is equivalent during growth according to the principles of isometric scaling 46 . This may explain why the triggering of FAM20C upregulation and DMP1 secretion happens in both genders at the same age despite the difference in absolute body weight, and presumably after a key mechanical threshold has been reached. Our results suggest that mechanical force is a key player in the control of angiogenesis in bone and hence bone growth, acting locally on bones in addition to systemic sex hormones [24][25][26] . However, the signalling downstream of PIEZO1 that results in enhanced FAM20C expression remains to be elucidated. Moreover, potential effects of sex hormones on PIEZO1 receptor sensitivity should be analysed in future studies.
Based on our finding that FAM20C upregulation is induced by loading and it happens first in the OF centre, we hypothesise that the central part of the OF is exposed to the highest mechanical strain since it is a spot where the loading from all four domes of the GP gets integrated. Moreover, there is no cortical bone underneath to absorb this loading (Supplementary Fig. 8r). The translation of these findings into two-legged humans, where the loading distribution and control of bone growth might be different, remains to be examined in follow-up studies.
The data presented here demonstrate that mechanical forces transmitted through PIEZO1 play a crucial role in the regulation of skeletal development and bone maturation via FAM20Cmediated extracellular DMP1 accumulation and its control of blood vessel subtypes in bone. This knowledge may be applied in bone healing studies 47 , as well as in different kinds of bone pathologies featuring abnormal bone growth activity, including osteoarthritis, osteosarcoma and osteoporosis.
In all experiments, both genders were analysed together and no gender-specific differences were observed. All mice were bred under specific pathogen-free (SPF) conditions and treated according to the requirements of the German and UK legislation. Mice were kept in approved animal-care facilities and were housed 4-6 per cage in standard individually ventilated cages, maintained with a 12 h light/dark cycle at an ambient temperature of 21°C and 45% to 65% relative humidity with wood shavings bedding and nesting material. Mice had ad libitum access to tap water and standard rodent chow (Altromin -LASQCdiet™ Rod16, Autoclavable, 10 mm).  Fig. 7 Molecular mechanism transforming type H into type L vessels to limit adolescent bone growth activity. At the age of preadolescence, body weightassociated mechanical loading at the OF-GP border is only moderate, thus allowing the tip cells of type H vessels to invade the GP and promote bone elongation. During periadolescence, mechanical loading through increased body weight and/or muscle contractions triggers the mechanoreceptor PIEZO1 (1) to enhance the production of FAM20C kinase (2) in osteoblasts, which induces a burst in DMP1 secretion into extracellular matrix (3). Large amounts of extracellular DMP1 inhibit VEGF signalling in the OF by preventing the phosphorylation of VEGFR2 (4) andpresumably indirectly via reduced activation of VEFGR2the expression of VEGFR3 (5) on the tip cells of type H endothelium. Hence, highly-angiogenic type H vessels are transformed into quiescent type L vasculature to limit bone growth activity (6). In parallel, extracellular DMP1 leads to rapid matrix mineralisation and strengthening of long bones.
The age of mice, when they were used for the experiment was between 3 and 15 weeks, the precise age is always indicated in the text. All animal experiments were performed according to relevant laws and institutional guidelines (DRFZ, Berlin, Germany and University of Oxford, Oxford, UK) and were approved by German and UK animal ethics committees (LAGeSo license number G0175/19 and Project (Vincent) license as granted by the UK Home Office).
Laser microdissection. Tie2-GFP mice were killed via cervical dislocation at 4 weeks or 12 weeks of postnatal development. Femurs were immediately collected and frozen in SCEM medium using dry ice-isopropanol-hexane cooling bath. After freezing the blocks were kept at −80°C until cryosectioning. The temperature of the cryostat chamber and the knife was −32°C and -34°C, respectively. For cryosectioning, type D knife for hard tissue was used. All surfaces including the knife were cleaned with RNaseZap before use. Cryosectioning was performed with Kawamoto's films for LMD. To minimise RNA degradation, 12 µm sections were mounted into the metal frame directly after cutting and transported in the evaporating phase of liquid nitrogen. Laser microdissection was performed using Leica LMD7 device with fluorescent lamp for endothelial cell identification. To minimize the humidity during microdissection, constant flow of evaporating liquid nitrogen was applied. The microdissected pieces were collected into lysis buffer (ARC-TURUS® PicoPure® RNA Isolation Kit). Approximately 50 pieces were collected from each sample. Immediately after collection, the Eppendorf tube containing the sample was briefly centrifuged and incubated at 42°C for 10 min. Afterwards, it was vortex shortly and frozen on dry ice and kept at −80°C until RNA isolation.
RNA sequencing and analysis. RNA was isolated using ARCTURUS® PicoPure® RNA Isolation Kit according to manufacturer's instructions. Illumina libraries were generated using SMART-Seq v4 Ultra Low Input RNA Kit (Takara Clontech) and Nextera XT DNA Sample Preparation Kit (Illumina), with up to 10 ng of purified cDNA. The quality of synthesised cDNA was checked using Bioanalyzer (Agilent). For sequencing, Illumina NextSeq500 device was used generating 75 bp paired-end reads. Reads were mapped to the mouse mm10 genome using TopHat2 (PMID: 23618408) and Bowtie2 (PMID: 22388286) with default settings 48 . Further analysis was done with R 3.4.0 using the default settings of the deseq2 package (PMID:25516281). To identify the genes with the highest significance, fold change and decent expression level, the following filtering strategy was applied: mean expression value >500, Log 2 FC > 0.66, P < 10 −5 . The heat map for selected genes was generated by plotting log2-transformed read counts.
Immunohistochemistry. Mouse bones were processed as described previously 49 . Briefly, the bones were collected, cleaned from surrounding tissue and muscles, and immediately fixed in 4% freshly prepared ice-cold paraformaldehyde (PFA) for 4 h on ice. Afterwards, the bones were washed 3 times in ice-cold PBS followed by decalcification in 0.5 M EDTA solution (pH=7.5) for 24 h under constant agitation at 4°C. Cryoprotection was done overnight in 20% sucrose and 2% PVP solution at 4°C immediately after decalcification. Afterwards, 5 ml of embedding medium (8% gelatine, 20% sucrose and 2% PVP in PBS) was added per falcon tube and the bones were incubated in it for 45 min at 60°C followed by embedding into cryomolds. The femurs were always oriented the same way with the anterior side facing the bottom of the mould. After embedding the samples were left at room temperature (RT) for 30 min for complete medium solidification. The blocks were kept at −80°C in air-tight containers until cryosectioning. Cryosectioning was performed using NX70 ThermoFischer cryostat at −25°C and −27°C for chamber and blade, respectively. The disposable low-profile blades N35 for hard tissue (207500006, Feather) were used for cutting. 100 µm sections were generated and transferred onto Superfrost slides (J1800AMNT, Thermo Fisher Scientific). The slides were kept at −20°C until staining. For immunostaining, bone sections were air-dried, permeabilized in 0.3% Triton X-100 for 15 min, blocked in 10% donkey serum at RT for 30 min, and incubated with the primary antibodies diluted in 5% donkey serum in PBS for 2 h at RT or overnight at 4°C. After primary antibody incubation, sections were washed with PBS three times and incubated with corresponding secondary antibodies for 1 h at RT. Nuclei were counterstained with Hoechst (1:1000, 33342, Thermo Fischer Scientific). After incubation with secondary antibodies, sections were washed three times with PBS and mounted with Fluoromount-G medium (00-4958-02, Thermo Fischer Scientific). The list of used primary and secondary antibodies is provided in the Supplementary Information (Supplementary Table 1 and 2).
RNAScope. Freshly dissected mouse femurs were fixed in 4% freshly prepared PFA for 24 h at 4°C followed by washings in PBS. Cryoprotection was performed via overnight incubations in 10, 20, and 30% sucrose at 4°C. Bones were frozen in SCEM medium. The frozen blocks were kept at −80°C until cryosectioning. Kawamoto cryofilms type 3C(16UF) were used for cryosectioning. The sections were air-dried in the cryotome for 15 min and stored at −80°C in the slide box with molecular sieves inside. For RNAScope assay, sections were air-dried at RT for 5 min, washed in PBS for 5 min and baked on 60°C hot plate for 30 min followed by additional fixation in 4% PFA at 4°C for 15 min. Afterwards, the sections underwent 5 min dehydration steps in ethanol gradient: 50, 70, and 100% (x2). The sections were air-dried again for 5 min followed by 10 min incubation in hydrogen peroxide and washings in distilled water. The target retrieval buffer was prewarmed in the steamer for~10 min up to 75°C. The slides were submerged into it and left in the steamer for 10 min. The temperature in the steamer was 95°C. Afterwards, the slides were washed in distilled water and dehydrated in 100% ethanol for 10-15 s. The slides were air-dried for 5 min at RT or overnight. Protease III was applied, and the slides were incubated in the Hybez Oven for 30 min followed by washings in distilled water. The rest of the RNAScope procedure was performed according to the standard RNAScope® Multiplex Fluorescent v2 Assay protocol.
Double neurectomy. Mice received pre-emptive pre-operative subcutaneous analgesia. In all, 5 mm pieces of sciatic and femoral nerves of the right hind limb were cut out in 4-4.5-week-old C57BL/6J mice under general anaesthesia using isoflurane. After surgery, the mice were housed without tubes and on soft bedding. It was observed that all mice dragged the operated right limb without stepping on it. Mice were killed at the age of 5.5-6 weeks and the bones were collected for further analysis. The contralateral left femur was always used as control and was analysed in pair with the operated unloaded right femur.
In vitro culture of bone slices. Femurs were collected from 4-week-old C57BL/6 J mice. Half of the bone together with knee metaphysis was cut and frozen in 2.5% agarose on ice with subsequent cooling in liquid nitrogen. Under the sterile work bench, the frozen block was fixed on the sample holder for cryosectioning, which was placed on dry ice. 400−500 µm-thick slices of femur were produced using sterile disposable blade for hard tissue. Every slice was placed into a well of a 96well PRC plate filled with StemMACS™ OsteoDiff Media (Miltenyi Biotec). The position of the slice is depicted in Fig. 4a. At least two slices per femur were generated, each of them was placed into a separate plate. The plates were kept at 37°C with 5% CO 2 in a humidified atmosphere. One plate with slices was centrifuged twice daily at 19 g (the lowest possible speed of the centrifuge) at 37°C for 30 min. The control plate was not subjected to centrifugation. The medium was changed every 2.5 days. After 5 days of culture, the slices were fixed in 4% PFA on ice for 30 min followed by washings in PBS and decalcification for 30 min in 0.5 M EDTA on the cooling element under constant agitation. Cryoprotection was done in 20% sucrose and 2% PVP solution in PBS at 4°C overnight. The slices were frozen in embedding medium (8% gelatine, 20% sucrose and 2% PVP in PBS). 100 µm sections were prepared using the NX70 Cryotome (Thermo Fischer Scientific). The sections were stored at −20°C until immunostaining. The staining was performed as described in the paragraph "Immunohistochemistry". Paired analysis was done based on the slices derived from the same mouse.
Image acquisition. All immunofluorescent and RNAScope-multifluorescent stainings were acquired at high resolution with Zeiss LSM-880 confocal microscope using 20x or 63x objectives. All microscope settings including laser power were kept constant throughout the experiments within the same type of samples and same staining panels. All immunostained samples were always acquired in Z-stack mode followed by maximum intensity projection.
Image analysis. The images were processed and analysed using ImageJ and CellProfiler software, in compliance with Nature's guide for digital images. Quantification of immunofluorescent images: first, the total intensity of a fluorescent staining in a region of 300 µm width and 150 µm depth (from the growth plate towards the caudal region) was measured. Next, the area occupied by the protein was calculated by creating a mask (cf. Supplementary Fig. 2). Then the normalised intensity was calculated by dividing the total intensity by the occupied area (DMP1, EMCN) or by the total cell density (FAM20C). When the total amount of a protein per region was analysed, we used the total intensity without further normalisation to the area occupied by it (DMP1, pVEGFR2, MMP9). Background subtraction using ImageJ software was always done prior to quantification. For RNAScope-multifluorescent images, the central and peripheral regions of 600 µm × 300 µm area underneath the growth plate were analysed. A cell was counted as Runx2 + or Osx + if it contained at least one bright dot of the corresponding probe at 200x magnification. The final figures were compiled using Adobe Illustrator 2021 software (version 25.0.1).
Endothelial cell culture. The mouse capillary endothelial cell line bEnd.3 50 was kindly provided by Prof. Dr. Alf Hamann (Charité, Berlin). The cells were cultured in ATCC-formulated Dulbecco's Modified Eagle's Medium (Catalogue No.  containing 10% FCS and 100 U/mL penicillin and 100 mg/mL streptomycin. The cells were maintained at 37°C with 5% CO 2 in a humidified atmosphere. Cells were cultured with recombinant DMP1 (40 nM) for 4 or 24 h. At the end of this incubation time, VEGF (50 ng/ml) was added to the cells for 15-20 min with subsequent cell fixation for immunocytochemistry or RNA collection for qPCR.
Immunocytochemistry. bEnd.3 cells were cultured in 8-or 16-well slide chambers (Tissue Tek). By the end of the experiment, cells were fixed in 4% freshly prepared ice-cold PFA for 10 min. Afterwards, they were washed 3 times with PBS and blocked in PBS containing 10% FCS for 30 min. The primary antibodies were applied for 1 h at RT in PBS containing 5% FCS and 0.1% Tween-20. After primary antibody incubation, cells were washed 3 times with PBS-Tween-20 (0.1%) and incubated with appropriate secondary antibodies diluted in PBS containing Hoechst (1:1000) for 1 h at RT. Finally, the cells were washed again three times with PBS and mounted using Fluoromount-G medium (Thermo Fischer Scientific).
Von Kossa staining. The samples were prepared as described in paragraph "RNAScope". The staining was done according to the manufacturer's instructions (Silver plating kit acc. to von Kossa, Merck Millipore, #100362). Briefly, 10 µm undecalcified femur sections were placed in distilled water for 1 min followed by incubation in silver nitrate solution for 5 min under direct sunlight. Afterwards, the samples were washed in running tap water for 3 min and incubated in sodium thiosulfate solution for 5 min. Then the samples were washed in running tap water for 1 min and mounted in water-based mounting medium. The images were acquired with Leica DM750 equipped with Leica ICC50 W Camera Module using LAS software and analysed with ImageJ software (version 2.0.0).
Histomorphometry. Dissected skeletons were fixed in 3.7% PBS-buffered formaldehyde for 18 hours, before they were stored in 80% ethanol. Bones were dehydrated in ascending alcohol concentrations and then embedded in methylmethacrylate for undecalcified histology as described 51 . Histological sections of 4 μm thickness from the sagittal plane were stained by von Kossa/van Gieson procedures as described 52 . Histomorphometry was performed according to the ASBMR guidelines using the OsteoMeasure histomorphometry system (Osteo-Metrics, Decatur, GA, USA).
Statistics. Statistical analysis was carried out using GraphPad Prism software (version 8.0d and 8.4). Normal distribution was tested using Shapiro-Wilk and Kolmogorov-Smirnov tests. In case of normal distribution parametric two-tailed paired t-test or unpaired t-test were used. For non-Gaussian distribution, nonparametric tests were used: two-tailed Wilcoxon test for paired analysis and twotailed Mann-Whitney test for unpaired analysis. P < 0.05 was considered significant. Mice with body weight that was >20% below the average weight of their littermates were excluded from the inducible Dmp1-deletion experiments. All image quantifications were done automatically upon manual blinded selection of central and peripheral OF regions of the same size. Sample numbers are indicated in figure legends. Reproducibility was ensured by several independent experiments.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
RNA sequencing data are available at gene expression omnibus (GEO) under the accession number GSE148804. Source data are provided with this paper. All other data supporting the key findings of this study are available from the corresponding authors upon request. Received: 7 October 2021; Accepted: 2 May 2022;