Bone marrow pericyte dysfunction in individuals with type 2 diabetes

Patients undergoing hip replacement surgery were recruited to the study, under informed written consent, at the Avon Orthopaedic Centre, Southmead Hospital, Bristol, UK. The study protocol complied with principles of the Declaration of Helsinki, was covered by institutional ethical approval (REC14/SW/1083 and REC14/WA/1005) and was registered as an observational clinical study in the NIHR Clinical Research Network Portfolio, UK Clinical Trials Gateway.

The main clinical characteristics of the participants are included in Table 1. Type 2 diabetes was diagnosed according to the American Diabetes Association guidelines and was defined as follows: (1) patient/referring doctor reports a previous diagnosis of diabetes; (2) HbA_1c >48 mmol/mol (9.3%) and (3) off insulin for at least 12 months after diagnosis. Exclusion criteria comprised acute disease/infection, immune diseases, haematological disorders or malignancy, unstable angina, recent (within 6 months) myocardial infarction or stroke, critical limb ischaemia, liver failure, renal failure, pregnancy and lack of consent to participate. Randomisation was not required in the experimental plan. At all the occasions, a single patient’s sample was delivered from the Orthopaedic Centre to our laboratory and freshly processed according to the described protocols. Experimenters were aware the sample was from a diabetic or non-diabetic individual. Measurements were taken from distinct samples and each sample was generally measured in duplicate or triplicate, unless specified differently.

The femoral head was removed as a standard step of hip replacement surgery. Bone marrow was scooped into a sterile pot, transferred into Falcon tubes containing PBS-EDTA (ThermoFisher, Gloucester, UK, catalogue number 28348) and placed in a fridge for collection within 1 h. Only material otherwise discarded was collected for the study.

Bone marrow samples were stratified on Ficoll Histopaque 1077 (Sigma-Aldrich, St. Louis, MO, USA, catalogue number 10771) and centrifuged at 300 g for 30 min at 25°C. Mononuclear cells sedimented at the interphase were then collected, washed twice with PBS and assessed for viability by trypan blue staining (ThermoFisher, catalogue number 15250061). An average of 1 × 10⁸ bone marrow mononuclear cells (BM-MNCs) was labelled with CD34-conjugated microbeads (Miltenyi, Woking, UK) and immunomagnetically sorted. CD34-depleted cells were labelled with CD45-conjugated microbeads (Miltenyi) and further sorted. The CD34–CD45 double-negative population was labelled with CD146-conjugated microbeads (Miltenyi) and enriched through immunomagnetic sorting. The purity of the selected cell population was assessed by flow cytometry (see below). Samples with a purity below 90% were excluded from the study.

The CD34⁻CD45⁻CD146⁺ cell fraction was then seeded onto 24-well plates at a density of 1 × 10³ to 5 × 10³ cells per cm² and expanded in an α-MEM basal media (ThermoFisher Scientific, catalogue number 32561-029) supplemented with 20% FBS (ThermoFisher Scientific, catalogue number 16000044). Four to six cell lines per group were studied between passage three and seven in the subsequent experiments.

Bone marrow mononuclear cells were labelled with primary antibodies (ESM Table 1) in staining buffer (PBS supplemented with 1% bovine serum albumin, Sigma, catalogue number A2058) for 30 min at 4°C, washed with cold PBS and resuspended in staining buffer. They were then acquired using a FACScantoII (BD Biosciences, Wokingham, UK). Quantification was performed using the FlowJo v10 software (FlowJo, Ashland, OR, USA). Flow cytometry antibodies used are reported in ESM Table 1.

Protein extracts (20 μg) were separated by SDS-PAGE, transferred to PVDF membranes (Amersham-Pharmacia) and then probed with the antibodies listed in ESM Table 2.

Portions of bone marrow were fixed in formalin 37% for 16 h, decalcified in 20% EDTA – 2% HCl solution for 4 h and then embedded in paraffin. The samples were sectioned on a rotary microtome at 2 μm, dried, deparaffinised and rehydrated. Antigen retrieval was performed by boiling the samples in a citrate buffer (10 mmol/l, Sigma, catalogue number P4809) at pH 6. After blocking non-specific binding with non-immune goat serum (ThemoFisher, catalogue number 10000C), sections were washed and then incubated with the following primary antibodies indicated in ESM Table 3: polyclonal mouse anti-human melanoma cell adhesion molecule (MCAM, BD Biosciences), polyclonal rabbit anti-von Willebrand factor (VWF, Abcam, Cambridge, UK), rabbit monoclonal anti-CD146 (Abcam), mouse monoclonal anti-protein gene product 9.5 (PGP9.5, Abcam), or monoclonal mouse anti-αSMA (Dako, Ely, UK) in PBS. All incubations were performed overnight at 4°C. The proper secondary antibodies (ESM Table 3), goat anti-rabbit or anti-mouse IgG (Alexa Fluor labelled), diluted 1:200 in PBS, were incubated for 60 min at 37°C. Nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI, 1 μg/μl, ThermoFisher, Catalogue number D1306). To remove the excess of autofluorescence, the bone marrow sections were immersed in a solution of 0.1% wt/vol. Sudan black (Sigma-Aldrich, catalogue number 86015) in 70% ethanol for 10 min at room temperature.

Cells were seeded in eight-chamber slides, grown until confluence and then fixed with 4% wt/vol. paraformaldehyde for 15 min at room temperature. This step was followed by an incubation overnight at 4°C with the primary antibodies reported in ESM Table 3: rabbit anti-human CD146 (Abcam), rabbit anti-human CXCL12 (Cell Technologies, Cambridge, UK), rabbit anti-human vasclular endothelial-cadherin (VE-cadherin, Abcam), mouse anti-human nestin (Abcam), rabbit anti-human PDGFRβ (Santa Cruz, Heidelberg, Germany), rabbit-anti-human vascular endothelial growth factor receptor 2 (VEGFR2, Abcam), rabbit LEPROT (Abcam), and mouse VWF (Abcam). All incubations were performed overnight at 4°C in PBS. Then, the cells were washed with PBS and incubated with appropriate secondary antibody diluted 1:200 in PBS for 2 h at room temperature. Finally, cells were washed with PBS and counterstained with DAPI 1 μg/μl, for 1 min at room temperature. Images of random fields were obtained at ×200 magnification. Antibodies were validated previously by using negative and/or positive cell cultures as controls, according to the literature. Negative controls have been performed for each immunohistochemistry and immunocytochemistry experiment omitting the primary antibody.

Bone marrow CD146⁺ cells were infected with an adenoviral vector carrying the coding sequence for constitutively active myristoylated AKT (Ad-myrAKT) or the empty vector Ad66 (Ad-Null) as a control (both from Addgene, Watertown, MA, USA). Cells were grown in complete medium in six-well plates until reaching 60–70% confluence and then infected for 24 h with adenoviral vectors at 100 multiplicity of infection. The medium was then changed and cells were used for experiments after 48 h. Effective expression of the transgene was confirmed by an activity assay, as previously reported [15].

All the assays were performed with cells maintained in 5 mmol/l glucose.

Fibroblast growth factor 2 (FGF2), vascular endothelial growth factor (VEGF) A and B, CXCL12, ANGPT1, ANGPT2, IGFBP2 and SEMA6 were quantified using individual cytokine-specific ELISA kits (R&D Systems, Abingdon, UK).

A Kolmogorov–Smirnov test was used to assess whether the data had a parametric distribution. Values are presented as mean ± SEM. Two-tailed independent samples t test was used to compare two groups assuming equal variances. ANOVA was used to compare multiple groups experiments, followed by Bonferroni post hoc t test to compare each group individually. All analyses were carried out using GraphPad Prism 7.0 (San Diego, CA, USA). A p value of <0.05 was considered statistically significant.

By immunofluorescence microscopy, we verified that pericytes expressing CD146, CXCL12, nestin and αSMA localise around capillaries and in the proximity of PGP9.5-positive neuronal fibres in the bone marrow of diabetic and non-diabetic individuals (Fig. 1a–c).

Diabetes has profound remodelling effects on the bone marrow microvasculature [11, 34]. However, the impact on pericytes has not been previously investigated. We stained bone marrow samples using CD146 and VWF antibodies to distinguish endothelial cells (positive for both the antigens) from pericytes (expressing CD146 only) (Fig. 2c). As previously described [11, 12, 34], we observed a decrease in VWF-positive capillaries in the bone marrow of the diabetic compared with non-diabetic individuals (0.9 ± 0.15 vs1.27 ± 0.16 capillaries/μm², p < 0.05) (Fig. 2d). The size of arterioles was reduced in the diabetic group (Fig. 2e). However, we did not observe any difference in the localisation or density of CD146⁺ VWF⁻ pericytes between the two groups.

We next performed flow cytometry analysis of the abundance of pericytes within bone marrow mononuclear cells. Fig. 3a–e shows the gating strategy and Fig. 3f–i illustrates the individual values from different bone marrow samples for the rate of CD146⁺ (Fig. 3f), CD34⁻CD146⁺ (Fig. 3g), CD45⁻CD146⁺ (Fig. 3h) and CD34⁻CD45⁻CD146⁺ cells (Fig. 3i), the latter combination being previously used to identify human bone marrow pericytes [33]. There was no difference in the number of CD34⁻CD45⁻CD146⁺ pericytes from non-diabetic vs diabetic individuals: CD34⁻CD45⁻CD146⁺ pericytes accounted for 0.50 ± 0.19% (non-diabetic) and 0.33 ± 0.15% (diabetic) of total BM-MNCs (Fig. 3i).

We next assessed the antigenic profile of CD146⁺ immunosorted bone marrow pericytes after expansion in culture. Flow cytometry analyses confirmed the maintenance of CD146 abundance in cells from non-diabetic (92.8 ± 1.8%) and diabetic individuals (94.5 ± 1.9%) (Fig. 4a,b). The expanded cells were also strongly positive for mesenchymal markers, such as CD105 (non-diabetic, 74.0 ± 8.1%; diabetic, 76.3 ± 2.2%), CD73 (non-diabetic, 79.9 ± 5.3%; diabetic, 78.8 ± 2.8%) and CD44 (non-diabetic, 74.6 ± 5.0%; diabetic, 70.0 ± 5.5%). Both groups lacked expression of haematopoietic and endothelial markers. Fluorescent immunocytochemistry showed that bone marrow pericytes expressed NG2 as well as nestin, LEPROT, PDGFRβ and VEGFR2 but were negative for VWF and VE-cadherin (Fig. 4c). The CXCL12 signal was almost absent in CD146⁺ pericytes from diabetic individuals. The reduction in CXCL12 expression was confirmed by qPCR and ELISA (see below).

We next investigated the impact of diabetes on pericytes’ function. Both viability and proliferation were significantly reduced in pericytes from diabetic individuals (p < 0.05 for both comparisons, Fig. 5a,b), which also showed a fourfold increase in the abundance of late apoptotic (Annexin V⁺/propidium iodide⁺) events (p < 0.05). The fraction of early apoptotic cells did not differ between groups (Fig. 5c–e).

Pericytes from diabetic individuals showed an impaired migration capacity towards a PDGFB gradient (Fig. 5f). Pericytes from non-diabetic individuals were able to support the formation of networks by HUVECs either directly in a co-culture system or indirectly through factors contained in their conditioned medium (p < 0.05 vs HUVECs alone) (Fig. 5g showing representative images and Fig. 5h showing mean and individual values). These effects were nullified by diabetes. Moreover, the conditioned medium of diabetic pericytes exerted an inhibitory effect on HUVEC network formation (p < 0.05 vs HUVECs alone and p < 0.05 vs conditioned medium of control pericytes) (Fig. 5g showing representative images and Fig. 5h showing mean and individual values). All functional assays were performed under ‘normal glucose’ conditions, therefore the observed alterations are compatible with persistent metabolic memory of the diabetic condition.

We next investigated the mRNA levels of angiocrine factors in pericytes and corresponding protein expression in the conditioned medium. Both ANGPT1 and ANGPT2 mRNA expression were upregulated in bone marrow pericytes from diabetic individuals (p < 0.05 both comparisons), whereas TIE2 (also known as TEK), the gene encoding the ANGPT receptor (tyrosine kinase with immunoglobulin and epidermal growth factor homology domain-2 [TIE2]), remained unchanged (Fig. 6a–c). VEGFA was not affected by diabetes, whereas VEGFB was significantly downregulated (p < 0.001 vs the control group, Fig. 6d,e). This was associated with a decrease in the gene encoding insulin-like growth factor-binding protein 2 (IGFBP2) (p < 0.05 for diabetic vs non-diabetic individuals, Fig. 6f), a transcriptional activator of VEGF [35]. In contrast, the ephrin B2 gene (EFNB2), which reportedly controls VEGFR2 internalisation and signalling [36], was expressed at a similar level in the two groups (Fig. 6g). NOTCH signalling in endothelial cells and pericytes acts downstream of VEGF to shape the vascular network during angiogenesis [37]. We found that the gene encoding Notch ligand delta-like 1 (DLK1) was downregulated by diabetes (p < 0.05 vs non-diabetic individuals), whereas the genes encoding delta-like 4 (DLK4) and jagged 1 (JAG1) remained unchanged (Fig. 6h–j). A trio of genes encoding angiogenic factors—FGF2, CXCL12 and LEP—were downregulated by diabetes (FGF2 an CXCL12, p < 0.01 vs control group; LEP, p < 0.05 vs control group; Fig. 6k–m). Neuropilins play a central role in blood vessel physiology as receptors for semaphorin (SEMA) and VEGF isoforms [38, 39]. Here, we found diabetes reduced NRP1 transcripts in bone marrow pericytes and increased SEMA6A (p < 0.05 vs non-diabetic control individuals for both comparisons, Fig. 6n,o), a known angiogenesis inhibitor [40]. Diabetes did not alter genes encoding the two angiogenesis repressors sprouty (SPRY) [41] and thrombospondin (THBS1) (Fig. 6p,q) [42].

We next performed ELISA of secreted factors in the pericyte conditioned medium (Fig. 7). Diabetes induced an increase in immunoreactive ANGPT2 (p < 0.05) but did not alter ANGPT1, VEGFA, VEGFB, IGFBP2 or SEMA6A (Fig. 7a–e,h). Moreover, FGF2 and CXCL12 were reduced by diabetes at the protein level (p < 0.001, Fig. 7f,g).

AKT activity controls the ability of bone marrow endothelial cells to secrete angiocrine factors [43, 44] and AKT suppresses ANGPT2 secretion [45]. Moreover, type 1 diabetes inhibits AKT activity in bone marrow endothelial cells [15, 16]. Here, we found that type 2 diabetes does not alter AKT phosphorylation in bone marrow pericytes (see electronic supplementary material [ESM] Fig. 1). Nonetheless, AKT might be uncoupled from downstream signalling. Therefore, we attempted to rescue the pericytes’ function by either transducing them with constitutively active myristoylated AKT or inhibiting ANGPT2 by adding a blocking antibody to the conditioned medium.

The presence of Ad-myrAKT did not affect the proliferation, viability or apoptosis of pericytes from non-diabetic individuals. In contrast, in pericytes from diabetic individuals, infection with Ad-myrAKT resulted in increased proliferation (p < 0.01 vs Ad-Null, Fig. 8a), improved pericyte viability (p < 0.01, Fig. 8b) and reduced late apoptosis (p < 0.05, Fig. 8c–e). In pericytes from the non-diabetic group, Ad-myrAKT increased the migratory activity of pericytes towards PDGFB (p < 0.05 vs Ad-Null, Fig. 8f) but did not enhance their angiogenic activity (Fig. 8g,h). In pericytes from diabetic individuals, Ad-myrAKT blunted the migratory deficit (p < 0.05 vs Ad-Null) but did not abolished the difference vs non-diabetic group (p < 0.01, Fig. 8f). Ad-myrAKT abrogated the detrimental effect of diabetes on the promotion of endothelial networking by pericytes or their conditioned medium (p < 0.01 vs Ad-Null, Fig. 8g,h). This effect was associated with a reduction in ANGPT2 (p < 0.05) and an increase in FGF2 levels (p < 0.01) (Fig. 8i–l).

Addition of an ANGPT2 blocking antibody to the pericyte conditioned media abrogated the inhibitory effect of diabetes on promotion of endothelial network formation (p < 0.01 vs IgG isotype control, Fig. 9a–d), suggesting that excess production of ANGPT2 may be accountable for the loss of angiogenic activity of bone marrow pericytes in diabetes.

This study newly identifies a detrimental effect of type 2 diabetes on human bone marrow pericytes, involving the inactivation of the ANGPT–FGF2 angiocrine signalling pathway. Given the limited group size and the unavailability of recent HbA_1c measurements in some of the participants, it was not possible to conclude that pericyte dysfunction correlates with metabolic control. Interestingly, pericytes were expanded and functionally assessed under in vitro conditions of normal glucose, thus suggesting a persistent memory of the diabetic environment. These data could have important translational implications and require further investigation. For instance, pericyte dysfunction might contribute to bone marrow microangiopathy, with an indirect impact on proper mobilisation of stem/progenitor cells into the circulation. Furthermore, a reduced angiocrine signalling from pericytes and endothelial cells might contribute to alterations of the haematopoietic system, including cell-intrinsic alterations of haematopoietic stem cells, leading to a decline in self-renewal, immune system dysregulation, predisposition to myeloid neoplasms and delayed haematopoietic recovery after myelosuppression.