If not specifically mentioned, all the reagents and vasoactive substances were from Sigma-Aldrich (Merck Life Science S.L.U, Darmstadt, Germany).

In total, 36 HHT patients, 5 with HHT-1 and 31 with HHT-2, participated in the study. Relatives not affected by the disease were recruited as controls for the study. All of them signed an informed consent form. The study was approved by the Local Ethical Committee. Epistaxis Severity Score were calculated for all participants in the study as previously described (20).

Venous blood samples were drawn into commercial buffered 3.2% sodium citrate for coagulation studies. Coagulation basic studies consisted of prothrombin time (PT) and activated partial thromboplastin time (APTT) and were measured according to international recommendations (21). Factor coagulant activity (Factor:C) was measured on a BCS XP System (Siemens, Healthcare Diagnostics, Erlangen, Germany) by modification of the one-stage clotting method with the use of factor deficient or depleted plasmas. Coagulation time by BCS XP was calibrated with standard human plasma. The normal ranges were considered to be 60–120%. Von Willebrand (vW) Factor activity was analyzed using latex particles surface-coated with recombinant GPIb (Siemens, Healthcare Diagnostics, Erlangen, Germany) (22, 23).

All animal protocols were performed with approval of the Committee for the Care and Use of Animals of the University of Salamanca (Spain), project #305 and complied with the current guides of the European Union and the US Department of Health and Human Services for the Care and Use of Laboratory Animals. Generation of Eng^+/− and Alk1^+/− mice has been described previously (24, 25). The animals were a generous gift from Michelle Letarte (Hospital for Sick Children, Toronto, Canada) and Peter ten Dijke (Leiden University, Netherlands), respectively. Both transgenic lines were backcrossed with C57BL/6J mice for more than ten generations. Of note, the genetic background of these HHT animal models do not show the characteristic bleeding phenotype due to fragile telangiectases as in other strains (26). Routine genotyping of DNA isolated from mouse tail biopsies was performed by PCR using the primers previously reported (25, 27). The generation and characteristics of the transgenic mice ubiquitously overexpressing human endoglin (ENG⁺) was performed by microinjection of a pCAGGS vector containing the complementary DNA sequence (cDNA) of endoglin in the fertilized egg of CBAxC57BL/6J mice, as previously described (28). WT littermates for each model were used as controls.

For animal anesthesia, 2% isoflurane in oxygen was used. During recovery from the anesthesia, heat was provided and, when necessary, a dose of buprenorphine (0.05 mg/kg) was subcutaneously administered. Animals were sacrificed by CO₂ inhalation or cervical dislocation, depending on the experiment requirements.

Isoflurane-anesthetized mice were placed in prone position. Their tails were transected at ~3 mm from the tip (29, 30) (constant tail diameter 1.4 mm) and immediately immersed in PBS at 37°C. Each animal was monitored for 30 min even if bleeding ceased, in order to detect any re-bleeding. However, when mice bled continuously for more than 10 min, the experiment was interrupted so that the life of the animal was not threatened. Three parameters were analyzed: (i) first bleeding time, the period between the beginning of the hemorrhage and the first time it stopped; (ii) re-bleeding time, defined as the time of the rest of periods where the hemorrhage starts again; and (iii) total bleeding time, the sum of the time in which the bleeding is active.

To analyze the capacity for reflex vasoconstriction of the vessels of mice models, we conducted studies of vasomotor functionality in thoracic aorta arteries obtained from mice of the different genotypes, as previously described (31). Briefly, arteries were removed and placed in chilled Krebs solution (118 mM NaCl, 4.7 mM KCl, 2.5 mM CaCl₂, 1.2 mM KH₂PO₄, 1.2 mM MgSO₄, 25 mM NaHCO₃, and 11 mM glucose). The pH of the solution after saturation with carbogen was 7.4. The arteries were carefully cleaned, cut into rings (2 mm) and aortic rings were mounted, with Krebs solution, onto a 4-channel myograph (Multiwire Myograph System, DMT, Denmark) After an equilibration period all rings were normalized and set at a resting tension of 5 mN and allowed to equilibrate for 30 min. After that, the vessels were tested for responsiveness to a hyperpotassium solution (KPSS; KCl 120 mM) and cumulative concentration–response curves to norepinephrine, serotonin and thromboxane analog U-46619 were generated in separate rings until the maximal response was consistent. Vasoconstrictor responses were expressed as a percentage of KCl-induced contractions. The cumulative concentration–response curves, were fitted to a logistic equation and analyzed with GraphPad Prism 9.0 software. We carried out statistical analysis of concentration–response curves using the extra sum-of-squares F test principle.

Mice were anesthetized as previously described and a catheter was implanted in the jugular vein. Collagen (150 ng/g body weight; Chrono-Log, Havertown, PA, USA) and epinephrine (15 ng/g body weight) were administered intravenously and blood samples were collected from jugular vein into EDTA tubes before and after 3 min of the injection of these compounds, that lead to platelet activation, aggregation and adhesion and, consequently, a decrease in platelet count. Platelet count in these samples was performed in an automated complete blood count analyzer (AdviaTM 120 hematology; Bayer, Leverkusen, Germany). At the end of experiment, anesthetized animals were euthanized by cervical dislocation.

Murine blood samples were obtained by retroorbital puncture using citrated capillary. Whole blood samples were centrifuged at 1,300 g for 5 min and resuspended in 1 mL of Tyrode's buffer (134 mM NaCl; 20 mM HEPES; 12 mM NaHCO₃; 5 mM glucose; 2.9 mM KCl; 1 mM MgCl₂; 0.34 mM Na₂HPO₄; pH 7.4). Due to the limited volume of blood samples obtained, flow cytometry was used to test platelets activation and aggregation as previously reported (32). Aliquots of diluted blood (30 μL) were incubated for 15 min at room temperature with the appropriate anti-mouse monoclonal antibody: APC-conjugated rat anti-integrin alpha IIb (GPIIb, CD41) (6.7 μg/mL; clone MWReg30; eBioscience, Thermo Fisher Scientific, Waltham, MA, USA) and PE-labeled rat anti-mouse integrin αIIbβ3 (GPIIbIIIa, CD41/61) (25 μg/mL; clone JON/A; Emfret Analytics, Würzburg, Germany) were used to detect platelets activation. PE-labeled CD9 and FITC-labeled CD9 (5 μg/mL; clone MZ3; BioLegend, San Diego, CA, USA) were used to visualize aggregation and the double positive population was considered the aggregated platelets. Thrombin (1 U/mL) was added when necessary. To remove excess antibody, aliquots were centrifuged (2,250 g for 5 min) and resuspended in 150 μL of Tyrode's buffer. Analysis was immediately performed in a BD Accury™ C6 flow cytometer (BD Biosciences, Franklin Lakes, NJ, USA).

Mouse Lung Endothelial Cells (MLEC) were isolated from the different mice described above (33). MLECs were grown in Advanced-DMEM F-12 medium (Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 20% FBS, 50 U/mL penicillin–streptomycin, 2 mM glutamine, 30 μg/mL endothelial cell growth supplement (ECGS) (Generon, Slough, UK) and 100 μg/mL bovine heparin in Petri dishes coated with 0.1% gelatin and supplemented with 0.01% collagen (Corning, Corning, NY, USA) and 0.01% fibronectin. Cells were maintained in 90% RH, 5% CO₂ atmosphere at 37°C. MLECs between passages 3 and 6 were used for the experiments.

Citrated blood samples were obtained from healthy donors, who had not taken any medication for at least 10 days. Whole blood was centrifuged at 100 g for 20 min to obtain platelet-rich plasma (PRP). PRP was washed twice by diluting with washing buffer (103 mM NaCl, 5 mM KCl, 2 mM CaCl₂, 1 mM MgCl₂, 5 mM glucose and 36 mM citric acid; pH 6.5) supplemented with platelet activation inhibitor PGE1 (2 × 10⁻⁷ M) and the ADP scavenger apyrase (1 U/mL), followed by centrifugation for 12 min at 1,240 g. Platelets were finally resuspended at 2.5 × 10⁸ platelets/mL in Walsh's buffer (137 mM NaCl, 20 mM PIPES, 5.6 mM dextrose, 1 g/L BSA, 1 mM MgCl₂, 2.7 mM KCl, 3.3 mM NaH₂PO₄; pH 7.4). Finally, washed platelets were stained with 1 μM calcein (Thermo Fisher Scientific, Waltham, MA, USA). Prior to any experimental procedure, platelets were typically left at room temperature for 30 min. Under resting conditions, platelets did not show activation.

MLECs were seeded in slides (Millicell EZ slide; Millipore, Burlington, MA, USA) and grown to reach a monolayer. Human-stained-platelets (10⁷ platelets/250 μL) were added on the MLECs monolayer. After an incubation for 30 min in the presence of CXCL12 (200 ng/mL; R&D Systems, Minneapolis, MN, USA), chamber wells were washed twice with PBS. Platelet adhesion was observed under a fluorescent microscope (Observer D1; Zeiss, Jena, Germany) coupled to a CCD camera (QIclick FCLR-12; Qimaging, Surrey, Canada) and quantifications were performed with ImageJ software.

Microfluid devices were used to perform flow assays as previously described (34). Briefly, the channels were coated with a monolayer of MLECs. Calcein-stained platelets (2.5 × 10⁸ platelets/ml) were perfused through the channels at 2 dynes/cm², allowed to adhere in the absence of flow for 10 min in presence of CXCL12. Then, the chambers were subjected to 2 dynes/cm² flow for 2 min. Real-time platelet adhesion was monitored under fluorescent microscope (Observer D1; Zeiss, Jena, Germany) coupled to a CCD camera (QIclick FCLR-12; Qimaging, Surrey, Canada). Then, images were quantified by ImageJ software.

Carotid arteries of isoflurane-anesthetized animals were exposed by a surgical opening in the ventral face of the neck. Before causing the vascular injury, a flow probe was placed around the carotid artery and carotid blood flow was recorded for 3 min using a Transonic Model TS420 flowmeter (Transonic Systems, Ithaca, NY, USA) in order to determine basal blood flow. After causing vascular damage by applying a filter paper saturated with 7.5% FeCl₃ on top of the vessel for 1.5 min, carotid blood flow was recorded for 30 min and represented as the percentage of carotid basal blood flow. The time to carotid occlusion was determined as the time required to achieve a flow rate of <5% that remains stable for more than 5 min.

Murine EDTA-blood samples were collected before and after 30 min, 6 h or 24 h since the intravenous administration of collagen (150 ng/g body weight; Chrono-Log, Havertown, PA, USA) and epinephrine (15 ng/g body weight) to generate a thrombogenic event and fibrinolysis system activation. Anticoagulated EDTA-blood was centrifuged and stored according to ELISA kits instructions. Plasminogen, t-PA, PAI-1 and D-Dimer levels were analyzed, respectively, by Mouse Plasminogen (PLG) ELISA kit (Cusabio, Houston, TX, USA), Mouse t-PA (Tissue-type Plasminogen Activator) ELISA kit (Elabscience, Houston, TX, USA), RayBio® Mouse PAI-1 ELISA kit (RayBiotech, Norcross, GA, USA) and Mouse D-Dimer, D2D ELISA kit (Cusabio, Houston, TX, USA) strictly following datasheet instructions. In all cases, we analyzed statistically, by two-way ANOVA, whether in each mouse model there were significant changes in the levels of the factor studied after the thrombotic event.

Data are expressed as the mean + standard error of the mean. All results are from at least 3 independent experiments and subjected to statistical test. Data were checked for Gaussian distribution using the D'Agostino-Pearson normality test. For normally distributed datasets, Student's t-test, ANOVA and two-way ANOVA was used, followed by Sidak's test for multiple comparisons. The statistical analysis of non-parametric datatests was made by the Mann-Whitney test (for two datasets) and Kruskal-Wallis test (more than two datasets). Fisher's exact test was used for the comparison of proportions when necessary. Results were considered statistical significant if p < 0.05 (ns: non-significant; ^*p < 0.05; ^**p < 0.01; ^***p < 0.001; ^****p < 0.0001). Graphs and statistical analysis were performed using GraphPad Prism 9 software (GraphPad software, San Diego, CA, USA).

Epistaxis Severity Score (ESS) was used to measure the severity and frequency of epistaxis in a group of HHT-1 and HHT-2 patients, as well as in relatives not affected by the disease who are the controls of the study. The distribution of controls and patients in age and gender is shown in Figure 1A. In the patient's group there is a slightly more number of men than women and a higher mean age compared to controls, but no significant statistical differences between the two groups were found.

As expected, the ESS of HHT patients is significantly higher than that of their healthy relatives (Figure 1B). In line with previous data in the literature and Rendu's statements when first describing the disease (35), there are no differences between control subjetcs and HHT-1 or HHT-2 patients in the coagulation basic studies, neither coagulation factor activity of the intrinsic or extrinsic coagulation pathways (Figures 1C,D, Supplementary Figure S1).

Having confirmed that increased bleeding in HHT patients is not due to coagulation alterations, and because dissecting other components of the hemostasis process in patients is not feasible due to ethical reasons, we decided to investigate it in murine genetic models of the disease: endoglin heterozygous mice (Eng^+/−) for HHT-1 and Alk1 heterozygous mice (Alk1^+/−) for HHT-2. As indicated in the methods section, the genetic background of these models do not present a bleeding phenotype due to fragile telangiectases as in other strains (26).

First, we aimed to analyze whether HHT mouse models exhibit increased bleeding when the lesion occurs in a normal vessel, rather than in a fragile vascular lesion as an HHT telangiectasia. For this purpose, we measured the tail bleeding time, which allows the analysis not only of the bleeding time of mice after wounding, but also the presence and duration of any spontaneous rebleeding.

Although, our data show that there is no difference in the first bleeding (Figures 2A,B), when rebleeding is analyzed, the number of mice with intense rebleeds (longer than 120 s) is significantly higher in both Eng^+/− and Alk1^+/− mice with respect to their controls (Figures 2C,D). Actually, rebleeding can become so severe that in some Eng^+/− and Alk1^+/− animals the bleeding may not stop spontaneously after 30 min from injury and, thus, it must be stopped by the investigator (Figures 2E,F). For these reasons, the rebleeding time, i.e., the sum of the times of the subsequent spontaneous bleedings, is significantly longer in both Eng^+/− or Alk1^+/− mice with respect to their controls (Figures 2G,H). In conclusion, both murine models of the disease have a prolonged bleeding time suggesting an impaired hemostasis.

Hemostasis is firstly dependent on a vascular spasm that reduces blood flow and increments the efficacy of the following steps of hemostasis, such as platelet activation and aggregation. Thus, we analyzed in artery rings obtained from thoracic aorta of HHT-1 and HHT-2 animal models whether they presented a lower response to different vasoconstrictor agents involved in vascular spasm, such as norepinephrine, serotonin and thromboxane analog U-46619, which could explain the increased bleeding in these animals. As shown in Supplementary Figure S2, significant differences were only found with respect to the control in the response of Eng^+/– mice to serotonin (Supplementary Figure S2C). However, these mice showed an increase in the contraction response, rather than a reduction. Taken together, our results support the conclusion that the areterial contractile response is not reduced in both HHT animal models.

Primary hemostasis depends on the ability of platelets to activate and aggregate in response to different stimuli. For in vivo analysis of these processes we studied the effect of intravenous injection of a sublethal mixture of collagen with epinephrine, which causes the activation and aggregation of circulating platelets forming thrombi throughout the vessels. Thus, after treatment, a decrease in non-aggregated platelets is observed, which correspond to those platelets that have formed aggregates. We found a significant difference between control and collagen/epinephrime treatment but no significant differences between WT and Eng^+/− or Alk1^+/− mice (Figures 3A,B). Thus, our results suggest that Eng or Alk1 deficiency does not alter the ability of platelets to activate or form platelet aggregates.

We confirmed these results by in vitro experiments performed with platelets isolated from Eng^+/− or Alk1^+/− mice. As shown in Figure 3, platelets from both mice are able to activate (Figures 3C,D) or aggregate (Figures 3E,F) in response to thrombin, with no differences in these processes between Eng- or Alk1-deficient mice and their respective controls.

Because the excessive bleeding in Eng^+/− and Alk1^+/− mice does not appear to be due to a defect in coagulation, a defective contractile response or defects in platelet function, we posit that HHT proteins, predominantly expressed in endothelial cells, could be involved in later stage endothelium-dependent processes of hemostasis, namely thrombus stabilization or fibrinolysis.

Although the quiescent endothelium is remarkably refractory to platelet adhesion, platelet-endothelium adhesion can occur after endothelial stimulation and this mechanism is important for proper thrombus stabilization. Recently, we have described that endoglin may participate in this interaction between platelets and endothelium through integrin binding (17). We therefore analyzed whether Eng, or Alk1, heterozygosity in the endothelium was causing reduced platelet adhesion. To this end, monolayers of MLECs from Eng^+/− and Alk1^+/− animals and their respective control mice, were incubated for 30 min with calcein-labeled platelets and the integrin activator CXCL12 and binding of platelets was measured. We found that platelets displayed a lower adhesion to Eng^+/− MLECs compared to controls (Eng^+/+ MLECs) (Figures 4A,B). In addition, to confirm that the effect is dependent on endoglin expression, we analyzed platelet adhesion to MLECs from ENG⁺ mice, that constitutively and ubiquitously express high levels of endoglin. Our results show that adhesion to ENG⁺ MLECs was significantly higher than to WT MLECs (Figures 4C,D). Interestingly, this result does not hold in Alk1-deficient cells, since no significant differences were found when we compared platelet adhesion to Alk1^+/− vs. Alk1^+/+ MLECs (Figures 4E,F).

Next, we assessed the role of Eng in resistance to shear under flow, using microfluidic devices. Platelets were allowed to adhere to endothelial cells, previously stimulated for 10 min with the integrin activator CXCL12, and the resistance to perfusion at 2 dynes/cm² for 2 min was measured (Supplementary Figures S3A,B, Supplementary Videos 1, 2). We found that platelets resistance to flow was strongly reduced in Eng^+/− compared to WT endothelial monolayers. Similar to the static adhesion results, the resistance to shear flow of platelets attached to the Alk1^+/− MLECs monolayer is comparable to that of WT MLECs (Supplementary Figures S3C,D and Supplementary Videos 3, 4). These results suggest that the presence of Eng in the monolayer of MLECs contributes to guarantee a platelets' firm adhesion and this may be critical to ensure proper thrombus formation and stabilization.

Next, we analyzed thrombus formation and stabilization by measuring blood flow in the carotid artery after FeCl₃ administration on the external carotid wall. FeCl₃ diffuses to the endothelial layer and generates a damage capable of activating the hemostasis response and generating a thrombus that reduces and even blocks blood flow. Our results show that stable occlusion, determined as the reduction of carotid blood flow to <5% of baseline blood flow, was delayed in Eng^+/− compared to WT mice (Figures 5A,B). This is in agreement with the above results on platelet adhesion to the endothelium. In order to further assess the role of endoglin in these hemostasis functions, we analyzed thrombus formation in the mouse model overexpressing human endoglin (ENG⁺). We found that wound closure was significantly faster in ENG⁺ mice, which present constitutively increased levels of endoglin, than in control animals (Figures 5C,D), suggesting that endoglin is directly related to the formation of a stable thrombus and positively contributes to hemostasis. To rule out that this effect on stable thrombus formation was not due to endoglin overexpression leading to changes in platelet activity, we analyzed in vivo platelet activation and aggregation by measuring (i) non-aggregated platelets after the induction of a thrombogenic event in ENG⁺ mice and (ii) in vitro platelet aggregation in response to thrombin. Our results show that endoglin overexpression did not alter platelet activation or aggregation either in vivo (Supplementary Figure S4A) or in vitro (Supplementary Figures S4B,C). In addition, we analyzed the effect of endoglin overexpression on the bleeding time of these mice and also observed no difference between ENG⁺ and control mice (Supplementary Figure S4D).

At variance with the Eng^+/− mice, no significant differences between Alk1^+/− mice and controls were observed regarding thrombus formation and carotid closure (Figures 5E,F). This is consistent with the absence of differences in platelet adhesion to endothelial cells when comparing Alk1^+/− vs. Alk1^+/+ MLECs (Figures 4E,F and Supplementary Figures S3C,D).

These results suggest that endoglin haploinsufficiency, but not Alk1 haploinsufficiency, leads to unstable thrombus formation that may explain the occurrence of rebleeding, whereas in the Alk1-deficient condition, the difficulty in stopping bleedings seems to be due to a different mechanism.

Fibrinolysis is involved in the process of removing the clot formed upon hemostasis activation. We analyzed the main components of the fibrinolysis cascade in plasma samples from Eng^+/−, Alk1^+/− and their control mice, collected immediately before (basal levels) and 30 min, 6 or 24 h after an acute thrombotic event: the intravenous administration of collagen and epinephrine as above described. These collection times were selected by choosing the periods where the highest concentrations of each of the components are expected to be found.

The basal plasminogen concentration in plasma was similar in Eng^+/− and WT mice, and no significant changes were observed in both groups 30 min after the induction of the thrombotic event (Figure 6A). However, 6 h after the thrombotic event, tissue plasminogen activator (t-PA) (Figure 6B) and its inhibitor PAI-1 (Figure 6C) levels significantly increased in Eng^+/− mice while in the WT this increment is lower and not statistically significant. Finally, we analyzed the plasmatic levels of D-dimer, a breakdown product of the blood clot fibrinolysis. Our results show that 24 h after the thrombotic event, D-dimer levels do not increase significantly in either WT or Eng^+/−mice (Figure 6D). Therefore, Eng deficiency causes a greater release not only of t-PA, but also of its inhibitor PAI-1 after the occurrence of a thrombotic event; these changes being associated with no differences in in the plasma levels of D-dimer.

Interestingly, the fibrinolytic response of the HHT-2 animal model was substantially different from that of HHT-1 mice. Thus, after the thrombotic event induced in Alk1^+/− mice, a significant increase in plasminogen levels at 30 min was observed, whereas the control group remained unresponsive (Figure 6E). Similar to the HHT-1 model, 6 h after activating thrombogenesis, there was a statistically significant increase in plasma levels of t-PA in Alk1^+/− mice, but not in WT mice (Figure 6F). However, contrary to what was observed in Eng-deficient mice, in Alk1^+/− mice the increase in PAI-1 levels did not reach statistical significance and was clearly much lower than the significant increase that occurred in control mice (Figure 6G). Analysis of D-dimer levels 24 h after the thrombotic event showed a significant increase in plasma levels of Alk1^+/− mice with no statistical differences in control mice (Figure 6H). In summary, the generation of a thrombotic event in the murine model of HHT-2 results in a significant increase in plasminogen and t-PA. However, the increment in PAI-1 levels is much lower than in controls. Finally, an increase in D-dimer levels is detected that is not observed in controls. Together, these results support the existence of a greater fibrinolysis activity in Alk1^+/− mice.