Vitamin K antagonism impairs the bone marrow microenvironment and hematopoiesis

The complex entity of the bone marrow (BM) microenvironment (BMM) has been conceptualized as an important regulator of hematopoietic stem and progenitor cells (HSPCs).¹  Whereas the role of mesenchymal stromal cells (MSCs), which differentiate into chondrocytes, osteoblasts, and others,²  in supporting HSPCs is well established,^3-6  the role of other cells, such as mature osteoblasts, which were originally thought to be essential components of the HSC niche,^7,8  and macrophages, is controversial. Some studies reported that macrophages promote an increase in HSC number,⁹  whereas others did not.^10-12  However, macrophages contribute to retention^11,12  and engraftment¹³  of HSPC in this niche, and administration of granulocyte colony-stimulating factor leads to depletion of macrophages localized close to the endosteum.¹¹ 

Osteoblastic differentiation¹⁴  and function can be inhibited by the anticoagulant warfarin, an inhibitor of the vitamin K reducing enzyme vitamin K epoxide reductase. In fact, warfarin treatment can lead to reduction of bone density^15,16  and increased fracture risk.¹⁷  Conversely, vitamin K administration promotes bone synthesis.¹⁸ 

Warfarin has been used in 1% of the world’s population for prophylaxis or treatment of thromboembolic events¹⁹  for 6 decades. Vitamin K epoxide reductase converts vitamin K to the hydroquinone form of vitamin K, allowing vitamin K to act as cofactor for the vitamin K-dependent carboxylase, leading to γ carboxylation of glutamic residues. Enzymatic conversion of glutamic acid (Glu) residues to γ-carboxyglutamic acid (Gla) is required for the activity of vitamin K-dependent proteins.²⁰  In the coagulation system, warfarin reduces the activity of factors II, VII, IX, and X, as well as proteins C and S. In bone, osteocalcin, matrix Gla protein, periostin, and Gas6 are vitamin K-dependent proteins.^21,22  Periostin,²³  secreted by osteoblasts and MSC, was recently implicated in hematopoiesis.²³  Therefore, we hypothesized that warfarin treatment may affect hematopoiesis via impairment of the BMM or via impairment of the proteases of the coagulation system, known to be involved in HSC mobilization.²⁴ 

Here we show that vitamin K antagonists (VKAs) compromise HSC via impairment of the HSC-supportive function of macrophages of the BMM in mice and associate with reduced leukocyte counts in humans and engraftment of human HSC in immunosuppressed mice. Without providing a causal link, we demonstrate higher odds of warfarin use in patients with myelodysplastic syndrome (MDS), a disorder of HSC leading to pancytopenia. The warfarin-induced detrimental effect on HSC is shown to be a result of impairment of the interaction between integrin β3, expressed on HSC, and decarboxylated periostin in the BMM, as well as altered AKT signaling. Our data offer pathophysiological mechanisms for bone loss and altered hematopoiesis after long-term treatment with warfarin, and raise the clinical concern of continued warfarin use, especially in elderly patients, whose hematopoiesis may already be compromised.²⁵ 

First, 0.5 mg/kg/d or 0.05 mg/kg/d of warfarin, resuspended in phosphate-buffered saline, was administered to mice via subcutaneously implanted osmotic minipumps (ALZET minipumps, Cupertino, CA). Sham-operated (skin incision and suture) mice served as controls.

χ-squared test and logistic regression adjusted for age and sex were used to compare the odds of VKA use in individuals with vs without a diagnosis of MDS. A Student’s t test was used to compare the mean of 2 groups. Differences in survival were assessed by Kaplan-Meier nonparametric estimates (log-rank test). When multiple hypotheses were tested, 1-way analysis of variance (ANOVA) and a Tukey test were used as post hoc test. All data were presented as mean ± standard deviation. We used L-Calc software (Stemcell Technologies, Vancouver, Canada) to calculate HSC frequency by Poisson statistics. P values ≤ .05 were accepted as significant.

Treatment of mice with warfarin led to a modest, but significant increase of the international normalized ratio, a derived measure of the prothrombin time, at a dose of 0.5 mg/kg (P < .0001; supplemental Figure 1A-B, available on the Blood Web site) compared with sham-operated control mice. Numbers of leukocytes (P = .009; Figure 1A), monocytes (P = .008; Figure 1B), and lymphocytes (P = .002; supplemental Figure 1C) in peripheral blood (PB) were reduced at the higher and a lower dose of warfarin. The lower dose is equivalent to the human dose (approximately 0.05-0.07 mg/kg for an international normalized ratio between 2 and 3). The percentage of CD11b⁺ myeloid cells (P = .008; Figure 1C, supplemental Table 2) was also decreased. Oral vs subcutaneous administration of warfarin similarly reduced leukocytes and monocytes (supplemental Figure 1D). However, Lin⁻ c-Kit⁺ Sca-1⁺ (LKS), LKS CD150⁺ CD48⁻ at 2 (supplemental Figure 1E-G) and 4 (supplemental Figure 1H) weeks, myeloid progenitor cells (supplemental Figure 1I-J) or macrophages (supplemental Figure 1K) in the BM were not reduced. The effects of warfarin were observed 3 days after treatment begin (supplemental Figure 2A-B), but did not persist in PB cells beyond 2 months (supplemental Figure 2C-E). The colony-forming ability of HSPCs from warfarin-treated compared with control mice was significantly reduced (P = .02; Figure 1D). In response to a 1-time dose of 5-fluorouracil (5-FU)-induced stress, leukocyte counts were lower in warfarin-treated compared with control mice (P = .004; Figure 1E), and survival was shortened in mice treated with warfarin (P = .0004; Figure 1F). HSC impairment after 5-FU stress persisted after 2 months of warfarin treatment (P = .007; supplemental Figure 2F). Transplantation-based assays revealed impaired function of whole BM cells from a warfarin-treated compared with a control BMM in competitive (P = .04; Figure 1G; supplemental Figure 2G-J), serial (P = .004; Figure 1H), and limiting dilution assays. In fact, the limiting dilution assay revealed a 7.6-fold reduction of functional HSC in warfarin-treated compared with control mice (P = .0006; Figure 1I). Colony-formation assays of Lin⁻ cells, previously exposed to vehicle or warfarin in vitro to rule out a direct toxic effect of warfarin on hematopoietic cells, did not yield significantly different numbers of colonies (Figure 1J), and culturing of Lin⁻ hematopoietic cells in warfarin-containing media had no direct effect on phenotypical Lin⁻ or LKS cells, the percentage of annexin V⁺ DAPI⁺ dead cells (supplemental Figures 1E and 3A), their cycling (supplemental Figure 3B), or myeloid cells (supplemental Figure 3C). Homing of Lin⁻ (supplemental Figure 3D) and LKS (supplemental Figure 3E) cells to a warfarin-treated BMM was uncompromised. Further, administration of vitamin K1 to previously warfarin-treated mice rescued the decreased percentage of myeloid cells (P = .01; supplemental Figure 4A), the number of leukocytes after challenge with 5-FU (P = .02; Figure 1K; supplemental Figure 4B), and HSPC function after competitive transplantation (P = .05; supplemental Figure 4C-G), while having no significant effects on non-warfarin-treated control mice (supplemental Figure 4H). This suggested that the detrimental effect of warfarin on hematopoiesis was a result of deficient γ carboxylation of vitamin K-dependent factors. These results indicate that warfarin reduces myeloid cells in PB and impairs HSPC ability to adequately respond to hematopoietic stress. This effect on hematopoiesis occurs indirectly, possibly via detrimental effects on the BMM.

Patients receiving long-term treatment with warfarin experience bone loss¹⁶  and an increased risk for fracture.¹⁷  Enumeration of Nestin-GFP⁺ MSC, known to support HSPC,⁶  revealed a significant reduction in warfarin-treated compared with control mice (P = .02; supplemental Figure 5A). Further, histomorphometric analysis of femora of warfarin-treated mice revealed a significant reduction in cancellous bone mass (Figure 2A), shown by a decreased ratio of bone volume/tissue volume (P < .001; Figure 2B), reduced trabecular number and thickness, and increased trabecular separation (Figure 2A; supplemental Table 1), compared with control mice. In addition, osteoid volume/bone volume (P = .046; Figure 2C; supplemental Figure 5B) was significantly increased on treatment with warfarin, suggesting reduced mineralization and partially explaining the decreased bone mass. Osteoblast surface (P = .0008; Figures 2D; supplemental Figure 5B), as well as osteoblast numbers (P = .0001; supplemental Table 1; supplemental Figure 5B) were significantly increased in warfarin-treated compared with vehicle-treated mice. Impairment of osteoblast function was supported by the increase in osteoid volume, suggesting that osteoblast-driven matrix mineralization was impaired. Consistently, relative expression of osteoblastic genes such as Col1 (Col1a1) and others were significantly reduced in primary calvaria cells treated with warfarin compared with vehicle (supplemental Figure 5C). However, osteoclast surface (supplemental Table 1; supplemental Figure 5B) and pit formation on bone discs by warfarin-treated osteoclasts (P = .04; supplemental Figure 5D) were increased, partially explaining the reduced bone mass. In vitro treatment of osteoclasts with warfarin did not change their number (supplemental Figure 5E), size (supplemental Figure 5F), or tartrate-resistant acid phosphatase activity (supplemental Figure 5G). Taken together, these data suggest that warfarin treatment impairs osteoblast function and increases bone resorption.

As the influence of osteoblastic cells on HSPC is controversial¹  and macrophages contribute to HSC dormancy,⁹  we assessed the support of HSPC by vehicle- or warfarin-treated macrophages, obtained via their adherence to plastic²⁶  and reaching a purity of 95% CD11b⁺ F4/80⁺ cells (supplemental Figure 6A-C). Plating of untreated Lin⁻ hematopoietic cells on macrophages revealed significantly decreased numbers of cobblestone-forming areas on previously warfarin-treated macrophages (P = .02; Figure 3A). Consistently, coculture of Lin⁻ hematopoietic cells on sorted, previously warfarin-treated F4/80⁺ CD11b⁺ macrophages revealed a decrease of LKS cells (P = .04; supplemental Figure 6D). We also observed a modest increase of annexin V⁺ DAPI⁺ Lin⁻ cells in coculture on warfarin-treated macrophages (P = .04; supplemental Figure 6E).

To test HSPC support by macrophages directly, we intrafemorally cotransplanted vehicle- or warfarin-treated macrophages with untreated Lin⁻ BM cells. This led to decreased engraftment of total donor BM (P = .029; Figure 3B) and donor-derived myeloid cells (P = .023; Figure 3C) with persistence of injected macrophages (supplemental Figure 6F-G). In confirmation of a prominent role for macrophages in the warfarin-induced effects, ablation of macrophages by clodronate¹²  in wild-type mice or diphtheria toxin in LysM-iDTR mice²⁷  did not lead to a reduction of leukocytes after 5-FU challenge (supplemental Figure 6H) or reduced leukocyte counts (supplemental Figure 6I) after warfarin treatment, respectively.

Hypothesizing that HSC-supportive cytokines may be reduced in a warfarin-treated BMM, we cultured untreated Lin⁻ BM cells in conditioned medium (CM) harvested from macrophages or MSC grown in vehicle- or warfarin-containing medium. This revealed decreased hematopoietic cells (P = .04 [macrophages] and P = .009 [MSC]; Figure 3D), as well as a decrease of the myeloid (P = .001; supplemental Figure 6J) and Lin⁻ (P = .0001; supplemental Figure 6J) cell fractions in cultures grown in CM from warfarin-treated macrophages. Colony-formation assays of these Lin⁻ cells, previously exposed to the CM of vehicle- or warfarin-treated macrophages or MSC, revealed reduced colony formation after culture in warfarin-treated CM from macrophages and on replating also from MSC (P = .001 [macrophages]; P = .01 [MSC]; supplemental Figure 6K). A focused gene expression analysis revealed significant decreases of angiopoietin (Angpt1), insulin-like growth factor 1 (Igf1), and others in warfarin-treated macrophages (supplemental Figure 6L; supplemental Table 3). In summary, these data suggest that warfarin impairs the HSPC-supportive ability of macrophages, and possibly MSC, of the BMM, leading to compromised function of HSPC.

Several proteins produced by cells in the BMM require vitamin K for their γ carboxylation²²  and function. However, no relevant hematological phenotype has been observed in mice deficient for the γ-carboxylated factors osteocalcin,²⁸  matrix gla protein,²⁹  or Gas6.³⁰  Periostin, however, an extracellular matrix protein^31,32  secreted by MSC,²²  is the most abundant Gla-containing protein and regulates HSC function,²³  whereas periostin-deficient mice are characterized by anemia, myelomonocytosis, and lymphopenia.²³  As reported,³³  F4/80⁺ macrophages are a source of the vitamin K-dependent protein periostin (Figures 4A; supplemental Figure 7A). However, coimmunoprecipitation of total periostin in CM from vehicle-treated vs warfarin-treated macrophages revealed no significant reduction of total periostin levels (supplemental Figure 7B-C). In contrast, γ-carboxylated periostin was undetectable in CM of warfarin-treated macrophages in a coimmunoprecipitation, using an anti-gla antibody (Figures 4B; supplemental Figure 7D). Treatment of mice with carboxylated periostin (Figure 4B) and warfarin restored or rescued the number of colonies formed by BM cells (P = .01; Figure 4C), the percentage of CD11b⁺ cells in PB (P = .05; Figure 4D), and significantly reduced annexin V⁺ apoptotic LKS cells (P = .03; supplemental Figure 8A) compared with mice treated with warfarin alone. Cotreatment with periostin and warfarin also rescued the number of leukocytes in PB after 5-FU-challenge (P = .03; Figure 4E) and increased the engraftment of donor BM in competitive transplantation (P = .0001; Figure 4F) compared with the warfarin-treated control. Furthermore, competitive transplantation of total BM from CD45.1⁺ donor mice treated with vehicle or periostin (without warfarin) into CD45.2⁺ recipients led to temporarily and modestly increased engraftment of CD45.1⁺ periostin-treated BM (P = .02; Figure 4G). In contrast, treatment of periostin knockout mice with warfarin did not lead to a decrease of leukocytes, monocytes (supplemental Figure 8B), or LKS cells (supplemental Figure 8C) compared with control mice, and repetitive doses of 5-FU to warfarin-treated periostin knockout mice did not alter leukocyte counts (supplemental Figure 8D) or survival (data not shown) compared with control periostin knockout mice. This suggested that periostin and, likely, the γ-carboxylation of periostin play a role in HSC impairment and the decrease of myeloid cells after warfarin treatment.

In vivo administration of periostin to mice treated with warfarin also rescued the number of LKS cells positive for integrin β3, whose binding to its ligand periostin³⁴  mediates HSC support (P = .003; Figures 5A; supplemental Figure 9A-B).³⁵  Warfarin treatment had increased the percentage of integrin β3⁺ CD11b⁺ myeloid cells (P = .05; supplemental Figure 9C). An altered physical interaction between periostin and integrin β3 became evident, as periostin in the CM of macrophages exposed to vehicle, but not warfarin (Figure 4B), efficiently bound integrin β3 overexpressed on 293T cells (Figures 5B; supplemental Figure 9D-F). Further, addition of periostin to cocultures restored adhesion of Lin⁻ cells to macrophages in warfarin-treated conditions (P = .03; Figure 5C). Overall, these data suggest that impairment of HSPC and myeloid cells after warfarin treatment is at least partially mediated by decreased binding of decarboxylated periostin to integrin β3 on HSPC leading to decreased HSC support. Differential effects of warfarin on myeloid vs LKS cell number may be a result of differences in expression of integrin β3.

A mere reduction of HSC number in warfarin-treated mice did not account for impaired hematopoietic reconstitution upon transplantation (Figure 1G-I). However, transplantation of equal numbers of CD45.1⁺ LKS cells from control or warfarin-treated mice into CD45.2⁺ recipients led to a higher percentage of CD45.1⁺ donor cells in recipients 1 month (P = .03; supplemental Figure 10A), but not 2 months, after transplantation, if the transplanted LKS cells were derived from warfarin-treated donors. Hypothesizing that HSCs from a warfarin-treated environment enter cell cycle on transplantation into an untreated BMM and exhaust thereafter, we showed a modest increase of the percentage of Lin⁻ cells in the S/G2/M phases of the cell cycle (P = .04; supplemental Figure 10B), which may explain the recovery of warfarin-treated HSPCs on exposure to a fresh or untreated BMM. However, in mice treated with warfarin compared with vehicle, the percentage of LKS cells in the G0 phase of the cell cycle was significantly increased (P = .005; supplemental Figure 10C). Consistent with a role of pAKT in cell cycling,³⁶  in particular downstream of periostin,³⁷  we observed a significant reduction of pAKT⁺ Lin⁻ cells in the BM of warfarin-treated compared with control mice (P = .037; Figure 5D; supplemental Figure 10D) and of pAKT⁺ in LKS cells after culture in warfarin-treated CM (P = .028; Figure 5E-F). AKT was also involved in the temporary recovery of warfarin-exposed Lin⁻ cells after plating on untreated macrophages, as treatment with the AKT-inhibitor MK-2206 prevented this recovery (P = .002; Figure 5G).

Because of the role of pAKT, downstream of integrin β3,³⁸  in periostin-mediated cell survival,³⁷  we tested the effect of integrin β3 inhibition by cilengitide, an inhibitor of integrins αvβ3, αvβ5, and α5β1, on Lin⁻ cells after culture in CM from vehicle-treated vs warfarin-treated macrophages. This led to a decrease of Lin⁻ cells in the G2/M phases of the cell cycle (P = .05; supplemental Figure 10E) and a decrease of pAKT levels in LKS cells (P = .02; Figure 5H-I) cultured in CM from vehicle-treated macrophages. In vivo, 5-FU challenge of control mice treated with cilengitide led to a trend toward reduction of leukocytes similar to mice treated with warfarin (supplemental Figure 10F) and an increase of annexin⁺ DAPI⁺ LKS cells (P = .0003; Figure 5J). However, periostin knockout mice treated with cilengitide were resistant to a change in annexinV⁺ DAPI⁺ LKS cell numbers (supplemental Figure 10G) and leukocytes (data not shown). This suggested that the detrimental effect of warfarin on HSPC may be a result of induction of a more quiescent state via nonbinding of integrin β3 to decarboxylated periostin and, consequently, reduction of pAKT. Transplantation into an untreated BMM or replating on untreated macrophages led to a temporary, pAKT-mediated increase in cycling.

To test a potential effect of warfarin on human hematopoietic cells, we transplanted untreated human CD34⁺ cells into nonobese diabetic severe combined immunodeficiency interleukin-2 receptor γ knockout (NSG) mice demonstrating reduced engraftment of human CD45⁺ leukocytes in warfarin-treated compared with vehicle-treated mice (P = .015; Figure 6A). Further, intrafemoral cotransplantation of untreated human CD34⁺ cells with vehicle- or warfarin-treated macrophages (supplemental Figure 11A) demonstrated decreased engraftment of human CD45⁺ leukocytes, when the HSC were cotransplanted with warfarin-treated macrophages (P = .046; Figure 6B; supplemental Figure 11B). Although still in the normal reference range, leukocytes (P < .001; Figure 6C), monocytes (P = .006; Figure 6D), eosinophils (P = .002; Figure 6E), and basophils (P = .0008; Figure 6F) in patients receiving warfarin were significantly decreased compared with controls. Similarly, there was a trend toward reduction of leukocytes (P = .08; supplemental Figure 11C) and significant reduction of monocytes (P = .04; Figure 6G) in patients receiving fluindione, another VKA, compared with controls. Platelets were not reduced in warfarin-treated (supplemental Figure 11D) or fluindione-treated (supplemental Figure 11E) patients. Leukocytes in patients receiving the factor Xa inhibitor apixaban, belonging to the novel oral anticoagulant group, were higher than in patients receiving warfarin (P = .01; supplemental Figure 11F). Although the differences in leukocyte populations between patients receiving VKA vs controls are minor, these data and reduced human HSC engraftment in warfarin-treated NSG mice support our results in mice.

We hypothesized that VKA use may be associated with an increased risk for MDS, a clonal hematopoietic stem cell disorder, which, in mice, may also arise as a result of impairment of the BMM.^39,40  In a population of 70- to 79-year-olds, VKA use was more frequent in people with vs without a diagnosis of MDS (14.66% vs 5.76%; P < .001; supplemental Table 4). The odds of VKA use was significantly higher in patients with vs without a diagnosis of MDS (adjusted odds ratio, 2.49; 95% confidence interval, 2.31-2.68; Figure 6H). In comparison, the odds of novel oral anticoagulant use were also higher in patients with vs without a diagnosis of MDS (4.74% vs 3.29%), although this association was weaker than for VKA (adjusted odds ratio, 1.35; 95% confidence interval, 1.19-1.55). As somatic mutations associated with clonal hematopoiesis and MDS risk increase with age⁴¹  and VKA use may trigger development of these mutations or increase MDS risk, we investigated VKA use in patients aged 50 to 60 years. Here, VKA use was also more frequent in patients with vs without a diagnosis of MDS (7.1% vs 0.85%; P < .001; supplemental Table 4). The odds of VKA use in this age group was significantly higher in patients with vs without a diagnosis of MDS, and higher than in patients aged 70 to 79 years (adjusted odds ratio, 8.2; 95% confidence interval, 6.64-10.13; Figure 6I). Sequencing of 5 control patients and 4 patients receiving VKA did not reveal any mutations associated with clonal hematopoiesis (supplemental Table 5). In summary, these data suggest that the odds of VKA use are increased in patients with MDS.

In this report, we show that VKAs compromise normal hematopoiesis via modulation of the BMM. In mice, the effect on PB cells is short-term, whereas the effect on HSCs persists for at least 2 months. Decarboxylated periostin and inefficient binding to integrin β3 on HSPC are mediators of this effect. Our results are consistent with previous reports ascribing a beneficial role to vitamin K for the treatment of cytopenias and MDS.^42,43  We extend these studies demonstrating a previously unreported, although descriptive and noncausal, link between VKA and human MDS.

Although a role for the periostin/ integrin αvβ3 axis for the support of hematopoiesis was elucidated,^23,44  periostin-deficient mice had an increased number of HSCs, characterized by increased cycling and faster recovery after sublethal irradiation.²³  HSCs in periostin-deficient mice exhibited poor repopulating ability on transplantation, similar to our data, but in vitro treatment of HSC with periostin decreased pAKT⁺ HSC.²³  In contrast, we observed impaired hematopoietic recovery after 5-FU stress and increased HSC quiescence, as well as a decrease of pAKT⁺ HSC after treatment with warfarin, consistent with AKT activation by periostin increasing cellular survival in colon cancer via αvβ3 integrin.³⁷  These discrepancies may reflect differences between quantitative vs functional (γ carboxylated) levels of periostin. Alternatively, genetic knockout of periostin may lead to compensatory production of proteins with similar function,⁴⁵  although this may not occur after decarboxylation of periostin by warfarin. Further, the duration of periostin deficiency is short-term in our studies vs more long-term in genetic deficiency of periostin.²³ 

Decreased bone mineral density and osteoporosis are adverse effects in patients receiving warfarin,^16,46  and fetal skeletal abnormalities in pregnant mothers receiving warfarin have been described.⁴⁷  Our data provide evidence that impaired osteoblast maturation and/or function, an increase of osteoid, decreased levels of functional periostin, and possibly increased osteoclast function may account for decreased bone mineral density. However, we do not link decreased osteoblast function to impaired hematopoiesis, but rather implicate decreased levels of γ-carboxylated, functional periostin, generated by macrophages, MSCs, and possibly other cells of the BMM.

Whether warfarin may be associated with clonal hematopoiesis,⁴¹  which increases the risk for atherosclerotic cardiovascular disease,⁴⁸  via these mechanisms is unknown and has not been established by us. Although a minor increase in thrombotic rate was observed in some patients with clonal hematopoiesis,⁴⁹  our human MDS data included patients with overt MDS only. Thrombotic risk is not characteristically elevated in MDS, but will require further investigation in clonal hematopoiesis/MDS in the future. Overall, interpretation of our human data must be done with caution, and discontinuation of warfarin, in view of lack of causal and mechanistic links, is not suggested at this time. Higher age in patients receiving warfarin and in patients with cardiovascular disease are confounding factors, and temporality between VKA use and MDS diagnosis needs to be carefully and mechanistically investigated. As patients with cardiovascular disease are frequently treated with VKA, it is possible that VKA may be associated with development and/or progression of clonal hematopoiesis. Increased HSC stress, possibly after discontinuation of warfarin therapy, may lead to reduced quiescence, possible exhaustion, accumulation of mutations, and clonal hematopoiesis. Sequencing efforts need to be extended to address this question on a large scale.

Given our results on periostin, the definitive study of the periostin/integrin β3-axis for maintenance or apoptosis of HSC is a worthwhile goal. Whether therapeutic periostin may improve engraftment²³  (Figure 4G) in HSC transplantation or improve MDS will be assessed in the future.

In conclusion, our results suggest that warfarin treatment impairs HSC function via compromising of functional periostin in the BMM. Our data implicate the periostin/integrin β3 axis as a mediator of HSC maintenance. Concomitantly, warfarin reduces osteoblast maturation and/or function, leading to impaired bone mineralization and a (possibly also osteoclast-mediated) reduction of trabecular bone. These data connect BMM impairment in warfarin-treated patients with previously underrecognized impairment of hematopoiesis, possibly establishing an association with MDS.

This work was supported by grant 2015_A238 to D.S.K. from the Else Kröner-Fresenius-Stiftung. Histomorphometric analysis was partially supported by a grant from the National Institutes of Health, National Institute of Arthritis and Musculoskeletal and Skin Diseases (P30 AR66261).

Contribution: D.V. designed and carried out experiments, analyzed the data, and critically reviewed the manuscript; R.K., R.S.P., C.K., C.Z., V.R.M. helped with cloning, co-immunoprecipitation, and other experiments and critically reviewed the manuscript; K.F. performed immunohistochemistry studies, analyzed data, and critically reviewed the manuscript; K.K. and S.H. performed the mass spectrometry experiments and reviewed the manuscript; S.Z.-C. and M.F. identified and contributed patients and analyzed the patient data; M.-J.J., J.E., and R.D.-S. analyzed the patient data from the French National Health Insurance Information System and reviewed the manuscript; F.K., K.H., and E.S. performed and analyzed the sequencing results; P.W. provided LysMCre-iDTR mice and advice on experiments; S.M. supervised the mass spectrometry experiments and reviewed the manuscript; N.F. provided statistical advice; P.D.P. performed experiments, analyzed data, and critically reviewed the manuscript; and D.S.K. designed experiments, supervised the project, analyzed data, and wrote the manuscript.

Conflict-of-interest disclosure: D.V. and D.S.K. hold patent EP18184430.9 for the clinical use of periostin.

Correspondence: Daniela S. Krause, Georg-Speyer-Haus, Institute for Tumor Biology and Experimental Therapy, Paul-Ehrlich-Str. 42-44, 60596 Frankfurt am Main; e-mail: krause@gsh.uni-frankfurt.de.