Hypoxia mimetics restore bone biomineralisation in hyperglycaemic environments

Diabetic patients have an increased risk of fracture and an increased occurrence of impaired fracture healing. Diabetic and hyperglycaemic conditions have been shown to impair the cellular response to hypoxia, via an inhibited hypoxia inducible factor (HIF)-1α pathway. We investigated, using an in vitro hyperglycaemia bone tissue engineering model (and a multidisciplinary bone characterisation approach), the differing effects of glucose levels, hypoxia and chemicals known to stabilise HIF-1α (CoCl2 and DMOG) on bone formation. Hypoxia (1% O2) inhibited bone nodule formation and resulted in discrete biomineralisation as opposed to the mineralised extracellular collagen fibres found in normoxia (20% O2). Unlike hypoxia, the use of hypoxia mimetics did not prevent nodule formation in normal glucose level. Hyperglycaemic conditions (25 mM and 50 mM glucose) inhibited biomineralisation. Interestingly, both hypoxia mimetics (CoCl2 and DMOG) partly restored hyperglycaemia inhibited bone nodule formation. These results highlight the difference in osteoblast responses between hypoxia mimetics and actual hypoxia and suggests a role of HIF-1α stabilisation in bone biomineralisation that extends that of promoting neovascularisation, or other system effects associated with hypoxia and bone regeneration in vivo. This study demonstrates that targeting the HIF pathway may represent a promising strategy for bone regeneration in diabetic patients.

Diabetes mellitus, a disease presenting with abnormally high levels of blood glucose (hyperglycaemia), can lead to bone metabolism disorders, increasing the risk of fractures, delayed union, the occurrence of non-union fractures and osteoporosis 1 . Clinical studies have reported that fracture healing in diabetic patients is prolonged by 87% 2 . Growing evidence indicates that the role of hyperglycaemia in bone remodelling may be associated with the hypoxia inducible factor (HIF) pathway 3,4 . This oxygen sensing pathway is important for endochondral bone development 5 , bone fracture repair 6 and bone homeostasis 7 , in addition to a number of steps in fracture repair e.g. inflammatory cell 8 and bone marrow derived mesenchymal stem cells (BMSCs) recruitment 9 , soft callous formation 5 , bone remodelling [10][11][12] and angiogenesis 13 . The role of the HIF pathway in bone biomineralisation is less clear with previous papers demonstrated decreased biomineralisation in hypoxic conditions 7 but others reporting increased osteoblast proliferation 14 .
In contrast, hyperglycaemia has been shown to decrease the stabilisation of HIF-1α and reduce the expression of hypoxia responsive element (HRE) genes associated with bone regeneration 4,15 . A number of mechanisms for this hyperglycaemic impaired HIF pathway have been suggested (as reviewed by Xiao et al. 16 ) including; the inhibition of HIF-1α stabilisation through hyperglycaemia-induced reactive oxygen species (ROS) generation and through advanced glycation end products (AGEs). ROS may increase Ras-related C3 botulinum toxin substrate (Rac1) expression, induce HIF-1α degradation by activating the prolyl hydroxylase domain (PHD) and increase ubiquitin-proteasome activity 17,18 . It has also been reported that hyperglycaemia reduces HIF-1α binding to the HRE and destabilises HIF-1α due to significant homology between the glucose responsive elements and HRE 19 .
Hypoxia plays an important role in bone fracture repair, where microvascular damage following bone fracture causes a hypoxic environment 20 and HIF-1α stabilisation 21 . This causes HIF-1α mediated inflammatory/ angiogenic factor production and initiation of the inflammatory phase of fracture repair vital for normal bone regeneration 8,22 . Hypoxia and HIF-1α stabilisation has been shown to be important in BMSC recruitment 23 , BMSC proliferation 24 , regulation of BMSC differentiation into chondrocytes and osteoblasts 22 . HIF-1α has been

Results
Hyperglycaesmia inhibited nodule formation. Mature nodules, with well-defined edges were observed in normoxia (20% O 2 ) normal glucose concentration (5.5 mM) (Fig. 1a), with mineralised extra-cellular collagen fibres as determined with TEM (Fig. 1g). However, moderate (25 mM) and high glucose (50 mM) environments inhibited nodule formation in a concentration dependant manner. Only a few smaller bone nodules were detectable in moderate glucose levels (Fig. 1b), with very few areas of extra-cellular mineral evident, whilst high glucose completely inhibited nodule formation (Fig. 1c). Collagen fibres and few mineralised structures were observed in TEM micrograph of moderate glucose (Fig. 1h), while high glucose only exhibited collagen fibres and nodules were hardly detectable in this condition (Fig. 1i).
Hypoxia (1% O 2 ) in normal glucose concentrations, completely prevented organic matrix deposition and the formation of bone structures (Fig. 1d). Biomineralisation deposition in hypoxic cultures was limited to small discrete calcified sites that were not associated with extracellular matrix (ECM-collagen fibres), this is sometimes referred to as dystrophic biomineralisation (Fig. 1d,j). SEM micrographs also confirmed presence of nodules that are higher than the culture surface in normoxia in normal glucose and dystrophic biomineralisation in hypoxia ( Supplementary Fig. 1a,b). Neither nodules nor biomineralisation were observed under hypoxia moderate ( Fig. 1e) and high (Fig. 1f) glucose conditions, although TEM revealed abundant collagen fibres in these conditions (unlike hypoxia low glucose where no collagen fibres were present) (Fig. 1k,l).
Quantification of bone nodule size was determined using interferometry ( Fig. 2a) and 2D image quantification of mineralised area (Fig. 2b) which confirmed the morphological observations with Alizarin Red stain (ARS). Nodules formed in normoxia normal glucose were larger (with 15.9 ± 0.2% of the surface above 30 µm, Fig. 2a) and covered more area than the discrete biomineralisation observed in hypoxia normal glucose (only 0.1 ± 0.1% of the surface above 30 µm, Fig. 2b, Supplementary Fig. 1c,d). Moderate and high glucose conditions showed a decrease in nodule formation compared to normoxia normal glucose (Fig. 2a). A larger mineralised area was observed in hypoxic conditions measured using image quantification, compared to interferometry (Fig. 2a,b), this is expected considering the numerous small discrete mineralised dystrophic nodules as opposed to the raised, mineralised ECM (bone nodules) observed in normoxia and highlights the caution needed in interpreting area quantification of biomineralisation alone without volume or nodule size measurements (e.g. with interferometer analysis) or ultrastructure characterisation.
Hypoxia radically reduced ALP expression (P ≤ 0.001) in all glucose environments (Fig. 2c). In normoxia high glucose condition, increased ALP expression (P ≤ 0.001) after day 1 but a decreased expression in ALP was observed in both moderate and high glucose conditions after 14 days in culture (P ≤ 0.01and P ≤ 0.05). All glucose levels showed similar ALP activity after 21 days. CoCl 2 and DMOG restored bone nodule formation in hyperglycaemic environments. To investigate the effects of HIF-1α stabilisation on bone formation in hyperglycaemic conditions, osteoblasts were treated with the hypoxia mimetic CoCl 2 and DMOG for 21 days. The effect of CoCl 2 , DFO and DMOG concentrations on osteoblasts metabolic activity and proliferation were initially performed. DFO significantly inhibited cell proliferation at all concentrations (12.5-50 µm) and was therefore excluded from future experiments. Transmission electron microscopy (TEM) micrographs of (g) normoxia normal glucose, (h) moderate glucose and (i) high glucose showed a glucose concentration dependent inhibition of bone nodule formation. COL fibres were not observed in (j) hypoxia normal glucose, but (k) moderate and high glucose environments in hypoxia showed some collagen fibres. Scale bar for (a-f) is 200 µm and for (g-i) is 2 µm. (n = 5) (N: nodule, COL: collagen fibres, OB: osteoblast).  (Fig. 3b,c). Although the size of the nodules was smaller than those in normoxia normal glucose (Fig. 3a).
The effect of hypoxia mimetics and hyperglycaemia on bone ultrastructure. Increased glucose levels completely inhibited ECM biomineralisation in normoxia as observed by TEM ultrastructure (Fig. 4a-c). CoCl 2 and DMOG did not prevent extracellular biomineralisation (Fig. 4). Differences were observed, however, between CoCl 2 and DMOG at differing concentrations. The mineralised ECM of the CoCl 2 and DMOG treated cultures in normal glucose (Fig. 4d,g,j,m) were less electron dense (not so dark), compared to the untreated culture (Fig. 4a), possibly indicating less mature biomineralisation. In moderate glucose conditions, 25 µM CoCl 2 appeared to be less densely mineralised (Fig. 4g) compared to both 12.5 µM CoCl 2 and untreated normal glucose cultured (Fig. 4a). Interestingly, although extracellular biomineralisation was observed in both 12.5 µM CoCl 2 and 25 µM CoCl 2 in moderate and high glucose conditions, the nodules restored in moderate glucose did not seem to be as mature compared to those in high glucose conditions (Fig. 4e,h compared to f and i). Abundant collagen fibres were observed in all hypoxia mimetic treated samples, but a reduction of number and organisation was observed in 500 µM DMOG treated cells in high glucose.

Figure 2.
Inhibition of bone nodule formation in hypoxia and hyperglycaemia quantified by (a) interferometry (area above 30 µm) and (b) image analysis of Alizarin Red staining (total area). Nodules cultured in normoxia normal (5.5 mM) glucose covered a substantially larger area (a and b total area) than both normoxia moderate (25 mM) and high (50 mM) glucose. (c) ALP activity per unit protein revealed that all normoxic conditions had higher ALP production than hypoxic conditions. Error bars represent the SD from the mean values. *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001 (n = 4).   and DMOG) increased bone nodule size (percentage of surface area with nodules above 30 µm) and area compared to untreated controls with similar level of glucose (Fig. 5). Whilst in normal (5.5 mM) glucose conditions, CoCl 2 had no significant effect on nodule formation, DMOG decreased the size of the nodules (area above 30 µM, P ≤ 0.001). In hyperglycaemic conditions, the level of bone nodule HIF-1α mediated increase depended upon the type and concentration of HIF-1α mimetic, and the glucose environment. CoCl 2 had a bigger influence on nodule formation, in terms of the size of the nodules (Fig. 5a,c) and area (Fig. 5b,d) than DMOG. CoCl 2 increased the area of nodules above 30 µm in all hyperglycaemic conditions by (200-790%), DMOG only showed an increase of 39% and 43% in 250 µm in HG and 500 µm DMOG in MG, respectively (Fig. 5c). While treatment with CoCl 2 significantly increased bone nodule formation in both moderate and high glucose conditions, it is important to note that it did not restore nodule formation to the same level as normal glucose controls.
The effect of hypoxia mimetics (DMOG and CoCl 2 ) on ALP expression. The HIF stabilising factor, CoCl 2 decreased ALP activity in all glucose conditions on day 1 (compared to untreated conditions) but did not affect the expression of ALP at later time points (day14, 21). On day 7, 12.5 µm CoCl 2 normal glucose and 25 µM CoCl 2 high glucose exhibited significantly enhanced ALP activity compared to the untreated controls (Supplementary Fig. 3a). The percentage change in ALP activity also showed that DMOG decreased ALP activity in all glucose conditions on day 1 (compared to untreated conditions) but did not affect the expression of ALP at later time point (day21). On day 7, 250 µM DMOG high glucose and on day 14, both 250 µm DMOG and 500 µM DMOG in moderate and high glucose exhibited significantly increased ALP activity compared to the untreated controls ( Supplementary Fig. 3b). www.nature.com/scientificreports/ Compositional analysis of the bone nodules. Raman spectroscopy was also performed to assess the biomolecular composition of the bone nodules (Fig. 6a,b). The nodules formed in normoxia normal glucose and those treated with CoCl 2 had a similar mineral to matrix ratio (PO 4 −3 ν 1 area : amide III) to the native bone, and similar position of the phosphate peak (~ 960 cm −1 ), while hypoxia normal glucose exhibited a more crystalline non-substituted hydroxyapatite as indicated by phosphate peak position (~ 962 cm −1 ) (P ≤ 0.0001) and a much increased mineral: matrix ratio (P ≤ 0.01). This marries with the ultrastructure (TEM) and morphological observations of dystrophic biomineralisation in hypoxic conditions, where there is much reduced protein contribution to the small crystalline minerals formed (i.e., in hypoxia there is not mineralised extracellular collagen fibres as observed in normoxia and CoCl 2 conditions). Unfortunately, DMOG and hyperglycaemic conditions Raman spectral analysis wasn't possible due to the smaller number and smaller size of the nodules available.

Hyperglycaemia impairs the HIF pathway and downregulates VEGF expression. Hypoxia
increased HIF-1α stabilisation (Fig. 7a) and VEGF expression (Fig. 7b) in normal glucose conditions. Hyperglycaemia decreased HIF-1α stabilisation (Fig. 7a) and VEGF expression (Fig. 7b) in response to hypoxia after 1 day (P ≤ 0.05). Interestingly, moderate but not high glucose conditions also decreased HIF-1α stabilisation in normoxia (Fig. 7a). Prolonged culture in hyperglycaemic conditions further decreased the osteoblastic response to hypoxia with reduced VEGF expression by 42% and 45% compared to normal glucose hypoxic response (Fig. 7b). CoCl 2 and DMOG stabilise HIF-1α and increase VEGF. CoCl 2 and DMOG increased HIF-1α stabilisation in all glucose conditions (Fig. 7c,d respectively). The amount of HIF-1α stabilised by the hypoxia mimetics were less than observed in hypoxia (1% O 2 ) but caused substantial increases in VEGF expression. High glucose conditions decreased the hypoxia mimetic stabilisation capability (P ≤ 0.001 in CoCl 2 , P ≤ 0.0001 in DMOG) and the VEGF response for both DMOG and CoCl 2 on day 1 compared to untreated controls. Both HIF-1α mimetics increased VEGF expression in all glucose environments compared to their normoxic controls (Fig. 7e,f), with the exception of VEGF expression in response to CoCl 2 in high glucose environments on day 1. DMOG increased VEGF expression substantially more than CoCl 2 and to a similar level observed in response to hypoxia on day 1 for 250 µM DMOG, and considerably exceeding the hypoxia response with 500 µM DMOG treatment. On day 7, whilst still significant, there was reduced VEGF expression in response to hypoxia mimetics in cultures exposed to DMOG compared to day 1.   36,37,40,41 , which is much higher than in diabetic patients, where the term hyperglycaemia is used for blood glucose concentrations above 70-110 mg/dL (3.9-6.1 mM) 42 . The justification for the use of these high levels in vitro, is to replicate cell behavioural responses (including reduced regenerative responses to hypoxia 13,14 ) caused by the long-term consequences of impaired glucose metabolism in vivo.
In this study we used atmospheric O 2 pressure as "normoxia", this is clearly far from the normal oxygen levels found in bone, as reviewed by Stuart et al. 43 . The oxygen level in a healthy human femur has been reported to be ~ 9.4% and upon fracture, it reduces to 1.54% 44,45 . In a fracture haematoma of a rabbit model, 0.8% O 2 and 3.8% O 2 were observed 2 days and 4 weeks post fracture, respectively [46][47][48] . Whilst mimicking the in vivo environment is undoubtedly important for in vitro modelling, it must also be considered that in vitro there isn't a multicellular and systemic response to hypoxia, as observed in vivo. Indeed an in vitro study by Utting et al. found that atmospheric oxygen pressures of 20% formed the most bone compared to lower oxygen pressures (with inhibition of nodule formation in 12% and 5%, strong inhibition at 2% and 1% and complete inhibition in 0.2%) 49 . A key component of tissue engineered models is to demonstrate the ability to form 3D tissue that is similar to the native tissue. This model demonstrated the ability to form mineralised ECM (TEM Fig. 1), that is biochemically similar to native bone (Raman Spectroscopy Fig. 6) and in sufficient volumes that differences in glucose levels were quantifiable with interferometer measurements (Fig. 5).

Hypoxia inhibits bone formation but causes aberrant extracellular biomineralisation.
In agreement with previous studies, our results demonstrated that hypoxia (1% O 2 ) in normal culture conditions (5.5 mM glucose) reduced mineralised bone nodule formation 10,49 . Utting et al. 49 reported slight inhibition of nodule formation in 12% and 5% O 2 , strong inhibition in 2% and 1% O 2 and complete inhibition in 0.2% O 2 . Our results demonstrate, through ultrastructural (TEM), biochemical (Raman spectroscopy) and size quantification (interferometry measurement), that these hypoxic nodules (or mineralised structures) were morphologically, and biochemically different to the control bone nodules or native bone. Despite covering a large percentage of the surface, the discrete mineralised structures were smaller (less than 30 µm in height), more crystalline, and had a much higher mineral:matrix ratio (P ≤ 0.01) than that observed in native bone or the control nodules (Fig. 6a). Ultrastructural analysis (TEM) found few collagen fibrils present in the hypoxic cultures (compared to the abundant fibres present in other oxygen tensions) and importantly the mineral structures were not associated with collagen fibrils. A decrease in collagen in hypoxic conditions has been mentioned in previous studies 11,50 and this lack of collagen may be because of the oxygen dependent functionality of the procollagen prolyl 4-hydroxylase, an enzyme involved in collagen formation 51 . The mineral formed in hypoxic structures is therefore not bone, but discrete hydroxyapatite mineral. It is unclear if these discrete mineral structures are formed spontaneously (possibly due to the addition of calcium and phosphate supplements commonly added to bone cultures) or is osteoblast mediated.
In addition to the reduced collagen (discussed above) other causes for hypoxia mediated impaired bone formation have been previously discussed including; the reduced proliferation and differentiation of immature osteoblasts 49 , and the reduction of soluble proteins (e.g. ALP, osteocalcin) important in osteoblast differentiation and biomineralisation 50,52,53 . In accordance with this literature, our results showed that hypoxia dramatically decreased osteoblast ALP activity and cell number in all glucose conditions 54 .
The hypoxic biomineralisation, observed here, illustrates the caution that must be used when interpreting the ARS as an indicator of in vitro bone formation. As illustrated by our results, ARS identifies calcium rich deposits, rather than a mineralised extracellular matrix found in bone. This may be especially relevant if high concentration of calcium and phosphate are present in the bone culture (e.g., through the addition of calcium and phosphate supplements, or the release of these ions by bioceramics). In contrast to bone biomineralisation, calcification is not a bone specific characteristic and can occur in other body tissues such as arteries 55 .

The difference between HIF-1α stabilisation and hypoxia on bone nodule formation. Hypoxia
(when cultured in normal 5.5 mM glucose) caused a substantial and prolonged decrease in ALP activity, together with a reduction in collagen and small dystrophic or discrete mineralised structures. However, treatment with HIF-1α stabilising factors (CoCl 2 and DMOG), revealed a different response, where extracellular mineralised collagen fibres were observed and ALP activity was only temporarily decreased (on day 1). This suggests that oxygen availability may affect HIF-1α-independent cellular mechanisms important in extracellular biomineralisation. There are a number of cellular responses to low oxygen pressure that are independent of the HIF pathway and occur prior to HIF-1α transcriptional response, these include glycolysis regulation (prior to upregulation of glycolytic enzymes and glucose transporters via HIF-1α), ROS production/regulation and the pre-transcriptional inhibition of oxygen sensitive proteins (such as the PHDs involved collagen synthesis).
The differences in biomineralisation observed between hypoxia and hypoxia mimetic stabilisation could also reflect the differing specificity and sensitivity of hypoxia mimetics (cobalt and DMOG), in targeting the HIF-1α pathway compared to hypoxia. For example, it has been reported that cobalt can upregulate HIF-1α genes whilst also down-regulating HIF-2α genes 56  www.nature.com/scientificreports/ bone biomineralisation where HIF-2α stabilisation has recently been shown to inhibit osteoblast differentiation 57 . Likewise differing oxygen pressure environments, and different durations of hypoxic exposure, have been shown to have different effects on osteoblastogenesis and bone formation 50 . Deletion of the HIF-1α pathway would help elucidate the differing effect of the chemical hypoxia mimetics and hypoxia, but may be technically challenging for in vitro bone models (requiring long-term culture), where inhibition of the pathway decreases osteoblast differentiation (osteocalcin expression) 58 and increases oxidative stress 28 . This may be why (to date) bone formation has not been demonstrated in vitro with HIF-1α inhibited primary osteoblasts.
Our results demonstrate that the use hypoxia mimetics can have favourable outcomes, compared to hypoxia, for bone tissue engineering, where high numbers of cells are needed (hypoxia mimetics did not inhibit proliferation compared to hypoxia), and bone-like nodule formation is possible.

Hyperglycaemia reduces bone nodule formation. Hyperglycaemic conditions (25 and 50 mM)
impaired the HIF-1α pathway (as observed with the reduced HIF-1α stabilisation and decreased expression of VEGF in response to hypoxia) and dramatically reduced bone nodule formation. Hyperglycaemia is known to cause HIF-1α dysfunction 59,60 and the impairment of bone formation in diabetic patients 36,61 . In vivo, this is likely to be due to the consequential decreased in angiogenesis 52 , osteoblast-osteoclast cross talk dysfunction 53 , reduced MSC recruitment 62 , changed inflammatory response 63 and other system effects. Here we show that hyperglycaemia inhibits biomineralisation in vitro, and this is in accordance with a few other reports where it is suggested that hyperglycaemia inhibits osteoblast proliferation and differentiation 40,64 .
The intracellular mechanism of how hyperglycaemia inhibits biomineralisation remains uncertain. In vitro, the glucose dependant impairment of the HIF-1α did not appear to effect osteoblast collagen formation, as observed with abundant expression in TEM images. Another study has also reported similar results, where high glucose (25 mM) did not diminish collagen production 65 . Likewise, whilst ALP activity, was drastically reduced in hypoxia it was not clearly affected by hyperglycaemia in our results. Others have reported a glucose concentration dependent effect on ALP production. Cunha et al. 66 and Gopalakrishnan et al. 54 reported that hyperglycaemia (16.5-49.5 mM) reduced ALP activity, whilst Garcia-Hernandez demonstrated that 12 mM glucose enhanced ALP activity 40 . Differences in cell type, duration of experimentation and ALP normalisation approaches may account for why the glucose concentration ALP inhibitory effects were not observed in our cell culture. The differences observed between hypoxia (non-ECM associated minerals) and high glucose levels (abundant collagen but no biomineralisation) on bone formation, suggests that glucose dependant inhibition of bone formation involves different mechanisms than hypoxia.

HIF stabilising chemicals CoCl 2 and DMOG restore bone nodule formation in hyperglycaemic environments.
Here, for the first time, we demonstrated that artificial stabilisation of HIF-1α (with CoCl 2 and DMOG) promoted bone nodule formation in hyperglycaemic conditions. The intracellular mechanism for this restoration is unclear. In vivo, hyperglycaemic impaired bone formation is often explained as due to the impaired angiogenic response in diabetic tissues (as seen here where hyperglycaemia reduced VEGF expression) 67 , or due to the increased ROS production promoting osteoclastogenesis 68 . HIF-1α mimetic restoration or enhanced bone formation are thought to be due to increased angiogenesis (as seen in Fig. 7) and restoration of osteoblast-osteoclast balance 29,69 . This clearly cannot be the explanation of the hypoxia mimetic restoration of bone-like extracellular matrix seen here (Figs. 3 and 4), where a osteoblast cell-line model was used. Rather, it suggests that that hypoxia may affect HIF-1α independent pathways important in bone mineralisation (possibly pre-transcription changes e.g. in ROS availability or glycolysis). Alternatively, it may mean that the hypoxia mimetics are also affecting other HIF independent pathways, due to their likely interaction with other oxygen sensitive Fe containing enzymes (e.g. procollagen prolyl 4-hydroxylase) 51 .
It is interesting to speculate on the mechanism of hypoxia mimetic restoration of bone mineralisation in hyperglycaemia, and a possible explanation could include the reduction of ROS through HIF-1α mediated generation of ROS scavengers, and modification of glycolysis 70 . Interestingly, Munoz-Sanchez et al. reported that hyperglycaemia and actual hypoxia impaired mitochondrial function 71 , whilst CoCl 2 treatment prevented mitochondria damage 72 . Mitochondrial damage may also inhibit mitochondrial related biomineralisation, which has been previously observed in osteoblasts 73,74 . Although this doesn't fully explain inhibition of extracellular mineralisation. Another explanation of why hypoxia mimetic agent restored hyperglycaemic inhibited bone formation, could be due to hypoxia mimetic agent activation of osteogenic differentiation factors. HIF-1α stabilisation has been shown to increase the gene expression of osteogenic differentiation factors such as Runt-related transcription factor (RUNX)-2 and osteocalcin 75 , although others have reported that hypoxia (as opposed to hypoxia mimetic agents) decreases RUNX-2 and osteocalcin expression 76 . A further explanation, related to metabolic adaptation, is the effects of the hypoxia mimetic agents on osteoblast number and differentiation. Osteoblast number is known to effect in vitro mineralisation 49 . Our analysis of proliferation (DNA quantification supplementary Fig. 4) showed that hypoxia mimetics did not cause significant change in the proliferation compared to normoxia (compared with the same level of glucose), whilst hypoxia did. Thus, differentiation might be the cause for the hypoxia mimetics mediated restoration. It is of course possible, due the broad mode of action of the hypoxia mimetics, that the hypoxia mimetic agents (Co and DMOG) are influencing HIF independent pathways, and that these are also important in restoring bone nodule formation in hyperglycaemia.
Hypoxia mimetics for treatment of diabetic related bone disorders. HIF stabilisation strategies need to control the location and duration of HIF stabilisation, as systemic HIF stabilisation or uncontrolled HIF stabilisation may cause chronic inflammation and other adverse effects [77][78][79]  www.nature.com/scientificreports/ als with the ability to form carbonated hydroxyapatite layer with exposure to biological fluids which facilitates binding of BGs to bone surface and bone regeneration 80,81 . Creating BGs with different ion release profiles e.g., silicate (Si), strontium (Sr) or cobalt (Co), for specific patient cohorts (e.g. for diabetic patients) is not only possible but probably favourable in the evolution of this field.

Conclusion
A hyperglycaemic bone model was developed that demonstrated that moderate and high glucose levels (25 and 50 mM) inhibit bone nodule formation. Interestingly, for the first time, it was demonstrated that the use of hypoxia mimetics (CoCl 2 and DMOG) helped restore bone nodule formation. This discovery demonstrates the role of the HIF pathway in bone development and the separate roles of HIF-1α stabilisation and low oxygen availability. Targeting the HIF-1α pathway in bone, offers the possibility of new interventions, including the design of biomaterials or tissue scaffolds specifically for patients with impaired bone regeneration due to a defective cellular oxygen sensing pathway.

Materials and methods
Bone nodule formation. Calvarial osteoblastic cells were isolated from 3-day-old Sprague-Dawley rats according to the sequential enzyme digestion protocol described by Orriss et al. 38 . All animal experimentation protocols were approved by the University College London (UCL) Animal Care Services Ethical Review Committee, licensed under the UK Home Office regulations and carried out in accordance with the UK Animal (Scientific Procedures) Act 1986 (Home Office, London, United Kingdom) upholding the highest standards of ethical practice and care in all aspects of this research. This study was carried out in compliance with the ARRIVE guidelines (https:// arriv eguid elines. org).
Angiogenic response. On day 1 and 7, the cell culture supernatants were collected and spun using a plate centrifuge (Hettich Universal 320R, Germany) at 1500 rpm for 5 min at 4 °C to remove cellular debris. VEGF concentration was quantified using a Quantikine ELISA kit (R & D Systems, UK) according to the manufacturers' protocol. VEGF concentration was normalised to DNA content.
HIF-1α stabilisation. Osteoblasts were seeded into T75 flasks at a density of 1.2 × 10 6 in 15 ml supplemented α-MEM and treated with the conditions mentioned above. After 72 h, nuclear extracts were prepared using the nuclear extraction kit (Abcam, UK) according to the manufacturer's instructions. The protein activity of HIF-1α was measured in nuclear extracts using HIF-1α transcription factor assay kit (Abcam, UK). HIF concentration was normalised to DNA content.
Alkaline phosphatase (ALP) activity. On day, 1, 3, 7, 14 and 21 ALP activity (Alkaline Phosphatase Assay Kit; Abcam) was determined using the manufacturer's protocol, by mixing 50μL of osteoblasts cell lysates (as explained in previous section) with 50μL of 5 mM pnitrophenyl phosphate (pNPP). After 60 min, when the colour change occurred, the assay was stopped by adding 20μL of the stop solution (NaOH), and the absorbance was measured at 405 nm using a multimode microplate reader (TECAN Infinite M200 PRO, Switzerland). ALP activity was normalised to total protein content determined using BCA protein assay kit (Merck).
Characterisation of bone nodules. Alizarin Red staining (ARS) assay. After 21 days culture, the cells were washed with PBS, fixed in 4% paraformaldehyde (PFA; Sigma-Aldrich) for 15 min, washed again and incubated with 40 mM Alizarin Red stain (ARS-pH between 4.1 and 4.5) in ddH 2 O for 20 min with gentle shaking on a shaker. ARS was removed, cells were washed five times with ddH 2 O and air dried. An EVOS XL Core light microscope (Thermo Fisher Scientific, UK) was used to take images of stained cells at 10× magnification.
Structural characterisation using transmission electron microscopy (TEM). Day 21 nodules were fixed in freshly made 2% PFA and 1.5% glutaraldehyde in a 0.1 sodium cacodylate buffer (both from Sigma-Aldrich) and kept in 4 °C fixative solution in refrigeration overnight. Nodules were then post-fixed in 1% osmium tetroxide and 1.5% potassium ferrocyanide (both from Sigma-Aldrich), suspended in 0. 3-Dimensional analysis of bone nodules using interferometry. Day 21 nodules, were fixed in 4% PFA at room temperature for 15 min, rinsed three times with ddH 2 O and air-dried at room temperature overnight. For each sample, the nodules at the centre of the Melinex disc (36 mm 2 square-shaped area; Agar Scientific) were assessed with Nexview-NX2 3D optical interferometer (Zygo, Middlefield, CT, USA) using 2.75× objective lenses, 0.5× zoom and a scan length of 145 µm. A minimum of 4 wells per treatment were analysed. The height and area covered with nodules (above 20 µm) were analysed in Mx software. The minimum was height was chosen to differentiate between spontaneous calcification (as observed in hypoxia) and mineralised ECM which is raised from the cell surface. The ARS images of were also used to measure the 2D surface covered with nodules in Image J (U.S. National Institutes of Health, Bethesda, MD, USA).
Biochemical analysis of bone nodules using Raman spectroscopy. Osteoblasts were cultured on magnesium fluoride substrate (MgF 2 ) (Crystran, UK) which is a weak Raman scatterer within the biochemical range of interest and does not interfere with the results. On day 21, the MgF 2 disks were rinsed with ddH 2 O and air dried at room temperature overnight. Air dried native bones from 3-day-old SD rat calvaria were used as controlled specimens for comparison. Spectra were collected using a 785 nm laser, 50 mW laser power, at room temperature and atmospheric pressure, on a Renishaw inVia spectrometer equipped with a Leica microscope. The laser was focused onto the sample using a 20X objective lens, ~ 3-4 µm diameter spot size. The spectra were recorded at a resolution of ~ 1-2 cm −1 in the region of 300-2000 nm, with 30 s acquisition time per spectrum and repeated 3 times per spot. A minimum of 10 bone nodules were sampled from 3 different MgF 2 for each treatment. The Raman spectra were processed and analysed with the Origin 9pro (OriginLab Corporation, Northampton, MA, USA), according to the protocol suggested by Gentleman et al. 82 . To assess the biochemical properties of the bone nodules univariate spectral analyses was performed to measure the average area of three peaks: apatite phosphate (PO 4 -3 ν 1 ) symmetric stretch (near ~ 960 cm −1 ), the type B υ 1 CO 3 peak (~ 1070 cm −1 ), the amide I band (~ 1665 cm −1 ). The mineral to matrix ratio was also determined by dividing the PO 4 -3 ν 1 band area by the matrix band area (amide I). The full width half maximum of the phosphate (PO 4 -3 ν 1 ) symmetric stretch (near ~ 960 cm −1 ), as a measure of mineral crystallinity, was also determined. and compared to native bone. Statistical analysis. All analyses were calculated using GraphPad Prism 8 software (CA, USA) unless otherwise stated. A minimum of 3 experimental repeats were considered for analyses. Significant differences between conditions were determined using the Student's-t-test with Welch's corrections for comparison between two data sets and one-way ANOVA followed by Holm-Sidak's multiple comparisons test for more than two data sets, with significance being defined as a p-value of less than 5% (P < 0.05).
Ethical approval. All animal experimentation protocols were approved by the University College London (UCL) Animal Care Services.

Data availability
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.