Angiopoietin-2 is associated with sickle cell complications, including stroke risk, and decreases with hydroxyurea therapy

Sickle cell anemia (SCA) is a hemoglobinopathy caused by the homozygous recessive β^S mutation in the β-globin gene resulting in hemoglobin S (HbS) expression. Most births affected by SCA occur in sub-Saharan Africa, where mortality remains high.^1-3 In SCA, the polymerization and precipitation of deoxygenated HbS results in dense sickled red blood cells, increased red blood cell rigidity, damage to the cell membrane, and hemolysis. SCA leads to a chronic inflammatory state associated with vascular dysfunction. This interplay of red blood cell sickling, hemolysis, and chronic inflammation at the level of the endothelium results in progressive vaso-occlusive end-organ damage.^4,5 Significant sequelae include cerebrovascular complications involving large cerebral arterial vessels in patients with SCA. SCA is a leading cause of stroke in children, with a life-time risk of stroke between 25% and 30% without disease-modifying therapy.⁶ 

Endothelial activation can be assessed by measuring circulating levels of soluble cellular adhesion molecules (CAMs).⁷ CAMs regulate leukocyte trafficking and undergo proteolytic cleavage to generate soluble CAMs, including soluble intercellular CAM 1 (sICAM-1), soluble vascular endothelial CAM (sVCAM-1), and soluble endothelial selectin (sE-selectin).^4,8,9 Endothelial activation also results in the rapid mobilization and release of Weibel-Palade bodies containing P-selectin and angiopoietin-2 (Angpt-2).¹⁰ Angpt-2 is a vascular growth factor that is released in the context of endothelial activation and can lead to increased vascular responsiveness to inflammatory cytokines through the upregulation of CAMs, and can lead to vascular leak and endothelial cell apoptosis when produced in excess.¹¹ 

Hydroxyurea is a disease-modifying therapy that functions by increasing fetal Hb (HbF), reducing red blood cell polymerization and hemolysis, and reducing white blood cell (WBC) and platelet counts, at least partially, through myelosuppression.¹² Hydroxyurea therapy is associated with fewer sickle-related complications including strokes, and reduces all-cause mortality in African children.^13,14 We hypothesized that hydroxyurea may also improve clinical outcomes through improved microvascular function. We tested this hypothesis in a randomized, double-blind, placebo-controlled clinical trial with Ugandan children with SCA receiving fixed-dose hydroxyurea or placebo for 1 year (NOHARM).¹³ An a priori defined secondary end point was to evaluate changes in markers of endothelial activation and inflammation, and transcranial Doppler (TCD) velocities associated with hydroxyurea therapy. Using serial samples collected over the first year of the study, we evaluated whether markers of endothelial activation and inflammation changed with hydroxyurea treatment, and used mediation analysis to assess the extent to which hydroxyurea improved TCD velocities as a surrogate measure of stroke risk through changes in endothelial activation as an intermediate, or mediating, pathway.¹⁵ 

NOHARM was a randomized, double-blinded phase 3 placebo-controlled clinical trial of oral hydroxyurea or placebo at a fixed dose of 20 ± 2.5 mg/kg of body weight per day, as previously described (ClinicalTrials.gov identifier: NCT01976416).^13,16 Inclusion criteria were documented SCA (HbSS by Hb electrophoresis); age of between 1.00 and 3.99 years at enrollment; weight of at least 5.0 kg at enrollment; residence within the area of Kampala, Uganda, where malaria is endemic; attending the Mulago Hospital Sickle Cell Clinic; and a willingness to comply with all study-related treatments, evaluations, and follow-up. Exclusion criteria included active use of hydroxyurea; known history of HIV, malignancy, or active clinical tuberculosis; severe malnutrition, defined as a weight-for-length z score of less than −3 using the World Health Organization growth standards; preexisting hematologic toxicity; a transfusion within 30 days; or alanine aminotransferase or creatinine level of >2× the upper limit of normal for age.

All children enrolled in the study received routine care for SCA according to the Mulago Hospital Sickle Cell Clinic guidelines including daily folic acid, penicillin prophylaxis daily for children aged <5 years, sulfadoxine-pyrimethamine prophylaxis for malaria prevention, and antihelminth medication according to the Ministry of Health guidelines. All children received an insecticide-treated bed net to prevent malaria.

On enrollment, children received a standard physical examination and had a detailed history recorded to evaluate sickle-related complications within the previous year. A blood sample was collected for confirmation of sickle cell status by Hb electrophoresis, a complete blood count and reticulocyte count, and biochemistries to assess liver and kidney function. EDTA–anticoagulated whole blood was centrifuged at 1200g for 20 minutes, and then the plasma layer was carefully collected using a sterile transfer pipette, and aliquots were stored in barcoded microtubes at −80°C until testing. Plasma samples were stored for biomarker evaluation at enrollment, and at 2-, 4- and 12-months follow-up. Between 10 and 12 months of follow-up, children had cerebral arterial blood flow velocities assessed by a single, certified examiner, with results confirmed off-site by central review, as previously described.^17,18 

The study was approved in Uganda by the Makerere School of Medicine research ethics committee, the Mulago Hospital research ethics committee, the Uganda National Drug Authority, and the Uganda National Council of Science and Technology, and in the United States by the institutional review boards at Indiana University and Cincinnati Children's Hospital Medical Center. Caregivers of the children provided written informed consent for participation in the trial with the use of forms in English or the local language.

A secondary outcome of the clinical trial included a change in markers of endothelial activation with hydroxyurea treatment.¹⁶ Enzyme-linked immunosorbent assays were conducted in batch on plasma samples using DuoSet enzyme-linked immunosorbent assays by R&D Systems (Minneapolis, MN) using methods previously described.¹⁹ We measured levels of Angpt-2 (1:5 dilution; assay range, 0.24-60 ng/mL; coefficient of variation (CV), 4.97%), P-selectin (1:50 dilution; assay range, 3.13-800 ng/mL; CV, 5.17%), C-reactive protein (1:10 000 to 1:20 000 dilution; assay range, 0.16-80 mg/dL; CV, 9.87%), soluble VCAM-1 (sVCAM-1, 1:5000 dilution; assay range, 76.8-20 000 ng/mL; CV, 6.38%), sICAM-1 (1:1000 dilution; assay range, 29.8-4000 ng/mL; CV, 7.47%), and sE-selectin (1:50 dilution; assay range, 2.35-600 ng/mL; CV, 3.87%). All sample testing was conducted in duplicate by technicians blinded to participant details.

Analysis was conducted using STATA version 17 (StataCorp LP, College Station, TX) and GraphPad Prism version 7.03 (GraphPad Software Inc, San Diego, CA). Clinical and laboratory data were described using the mean and standard deviation, or number (percentage). To evaluate the relationship between log₁₀-transformed biomarker levels and a history of sickle-related complications, logistic regression was used. Differences in longitudinal biomarker levels based on study treatment were evaluated using linear mixed effects models in which observations within participants were correlated using a patient-specific intercept, and time points were treated as categorical variables and adjusted for participant sex and age at enrollment. A banded diagonal covariance matrix was used to model the within-subject variance–covariance errors, and the mixed model was fit by restricted maximum likelihood, and the Kenward-Roger approximations was used to estimate the denominator degrees of freedom. Spearman correlation was used to evaluate the relationship between biomarker levels and complete blood count measures. The Benjamini-Hochberg false discovery rate was applied at a threshold of 0.05 to adjust for multiple comparisons, as indicated.

We leveraged the serial collection of data to explore multiple physiologically possible pathways through which endothelial activation may affect TCD velocities through the inclusion of intermediate, or mediating, variables. Mediation models aim to elucidate the underlying mechanisms or pathways through which the independent variable affects the outcome. The direct effect describes the relationship between the independent variable and study outcome without considering the mediating variables, whereas the indirect effect captures the influence of the independent variable on the study outcome via its effect on the mediating variables, which then influence the study outcome. Together, the direct and indirect effects encompass the total effect of the independent variable on the outcome of interest. Models were constructed using measures of HbF at 2 months, total Hb and WBC count at 3 months, angiopoietin-2 at 4 months, and TCD velocity at 10 to 12 months. We started by examining the temporal relationships between (1) hydroxyurea and HbF at 2 months; (2) Hb or WBC count at 3 months, and angiopoietin-2 at 4 months; and (3) angiopoietin-2 at 4 months, and TCD velocities at 10 to 12 months using linear regression models. Following the method of VanderWeele and Vansteelandt for estimating mediation effects for multiple mediators,²⁰ we estimated the indirect effects of hydroxyurea on TCD velocities through Hb and through WBC count. We used the Stata Structural Equation Modeling Builder for mediation analysis, and all analyses were adjusted for age and sex.²¹ 

Plasma samples were available for 205 children in the NOHARM trial at enrollment and comprised 798 samples, 400 from children treated with hydroxyurea and 398 from children randomized to receive placebo. The mean (± standard deviation) age of children at enrollment was 2.2 years (±0.87), and 46.3% of children were female (Table 1). A history of sickle-cell related complications was common, with 57.1% of children hospitalized in the year before enrollment; 54.2% with a history of blood transfusion, 78.1% reporting a history of dactylitis, 81.5% reporting vaso-occlusive crises (VOC), and 15.6% reporting a history of acute chest syndrome. None of the children had a history of clinical stroke on enrollment.

Plasma levels of endothelial activation markers were assessed in enrollment samples, and we evaluated whether a history of sickle-related complications was associated with differences in endothelial activation using regression. A history of dactylitis, VOC, acute chest syndrome, and transfusion was associated with increased Angpt-2 levels at enrollment after adjustment for age and sex (all P < .05 before correction). After correction for multiplicity, only a history of dactylitis and transfusion remained significantly associated with Angpt-2 levels at enrollment (Table 2). A history of dactylitis was also associated with increased sE-selectin on admission although the relationship was not significant after adjusting for multiple comparisons.

To evaluate whether hydroxyurea was associated with changes in endothelial activation over the first year, we conducted longitudinal analyses to evaluate changes in plasma biomarker levels from treatment initiation (Figure 1). At enrollment there were no differences in levels of endothelial activation markers or inflammation by treatment arm (P > .05 for all). However, children randomized to hydroxyurea had lower levels of sE-selectin, P-selectin, and Angpt-2 than children treated with placebo, independent of child age and sex, with differences in markers (sE-selectin or P-selectin and Angpt-2) evident after 2 to 4 months of treatment, respectively, and sustained over 1-year follow-up (Figure 1).

Because the endothelium is an important regulator of vascular reactivity and function, we evaluated the relationships between markers of endothelial activation and TCD velocities after 10 to 12 months of study treatment. A log₁₀ increase in enrollment values of both Angpt-2 (adjusted β coefficient, 31.5 cm/s; 95% confidence interval [CI], 13.4-49.7) and sE-selectin (adjusted β coefficient, 38.7 cm/s; 95% CI, 17.5-59.9) was independently associated with a higher TCD velocity (adjusted P < .05).

Of 161 children with TCD measurements between 10 and 12 months, 34 children (21.1%) had a conditional or abnormal TCD velocity (≥170 cm/s), which confers an increased risk of primary stroke.²² After adjustment for child age, sex, and hydroxyurea therapy, a log₁₀ increase in Angpt-2 or C-reactive protein on enrollment was associated with increased odds of a conditional or abnormal TCD at 10 to 12 months, with an adjusted odds ratio of 6.42 (95% CI, 1.03-39.99) and 2.00 (95% CI, 1.01-3.93), respectively (P < .05 for both before correction). However, after correcting for multiple comparisons, these adjusted odds ratios lost significance.

We evaluated the relationship between our markers of endothelial activation and inflammation and hematologic parameters at enrollment and 12-months follow-up with results stratified by treatment arm (Figure 2). sE-selectin and Angpt-2 were negatively correlated with Hb and HbF and positively correlated with WBC and neutrophil counts, whereas P-selectin was positively correlated with WBC, neutrophil, and platelet counts.

By leveraging the longitudinal collection of data over the course of the study, we were able to evaluate causal pathways associated with hydroxyurea effect on TCD velocity as a surrogate measure of stroke risk (Figure 3). We ultimately found 2 statistically significant pathways through which hydroxyurea may indirectly affect TCD velocities. The first pathway showed that hydroxyurea was associated with an increase in HbF at 2 months (β = 6.91; P < .001), which, in turn, was associated with an increased total Hb at 3 months (β = 0.09; P < .001), which, in turn, was associated with a decrease in transcranial velocities at 10 to 12 months (β = −7.33; P = .001). The second showed that hydroxyurea was associated with an increase in total Hb at 3 months (β = 0.43; P = .008) independent of HbF, which, in turn, was associated with a decrease in TCD velocities at 10 to 12 months (β = −7.33; P = .001). Apart from these 2 indirect pathways, hydroxyurea was not shown to have a direct effect on TCD velocities (β = 1.73; P = .698). Of note, multiple pathways involving angiopoiten-2 at 4 months approached statistical significance, including increased total Hb at 3 months associated with a decrease in angiopoiten-2 at 4 months (β = −0.17; P = .063) and decreased WBC count at 3 months associated with a decrease in angiopoietin-2 at 4 months (β = 0.03; P = .057).

Hydroxyurea increases HbF production and decreases inflammation, but its impact on endothelial function is not well characterized. In this study, hydroxyurea decreased plasma levels of sE-selectin, P-selectin, and Angpt-2 in children with SCA; lower levels of Angpt-2 and sE-selectin at enrollment and 12-month follow-up were associated with decreased TCD velocities, a strong predictor of stroke. Mediation analysis suggested that hydroxyurea therapy decreased TCD velocities, in part, by increasing HbF, which then increases Hb, and that there may be some Angpt-2–mediated pathways involving Hb and WBC count. Together, these findings provide new insight into how hydroxyurea may decrease SCA-related complications, including stroke, and suggest that Angpt-2 may be both a surrogate marker for hydroxyurea therapeutic response and potentially a target for SCA therapies.

The endothelium is a highly dynamic cell layer that serves as a critical regulator of health and disease.²³ Endothelial activation in SCA occurs through multiple pathways. Upregulated CAMs recruit immune cells to tissues leading to inflammation and adhesion of sickled erythrocytes to the endothelium, resulting in obstructed capillary flow, ischemia-reperfusion injury, and hemolysis.^24,25 Angpt-2 is a proangiogenic and proinflammatory protein that destabilizes the endothelium. Systemic increases in Angpt-2 are associated with sepsis,^26,27 retinopathy,^28,29 diabetes,^30-32 severe malaria,^19,33 neurocognitive injury and decline,^34-36 and ischemic stroke.³⁷ In the present study, Angpt-2 levels at enrollment were associated with a history of prior transfusion, dactylitis, acute chest syndrome, and VOC. The relationship between Angpt-2 and SCA-related complications on enrollment is consistent with previous studies reporting elevated Angpt-2 levels in children with SCA in steady state and during VOC,³⁸ in adults with SCA and a history of leg ulcers,³⁹ and in malaria.⁴⁰ Increases in Angpt-2 levels may reflect an underlying proangiogenic state triggered by ischemic insults, spanning from small-vessel complications (eg, dactylitis) to large-vessel complications (eg, stroke).

Several studies have evaluated hydroxyurea’s impact on endothelial function using in vitro or animal models of SCA with supporting evidence from plasma of people with SCA.^4,41 In the HUSTLE study (ClinicalTrials.gov identifier: NCT00305175), hydroxyurea decreased levels of sE-selectin but not sICAM-1 or sVCAM-1.⁴² This effect was associated with reduced pneumococcal burden and improved survival in an experimental SCA mouse mice.⁴² Hydroxyurea was associated with reduced levels of sE-selectin and sVCAM-1 in a US pediatric population of children with SCA, and reduced Angpt-1 and vascular endothelial growth factor in adults with SCA.^43,44 In this study, hydroxyurea decreased sE-selectin 2 months into treatment, and Angpt-2 and P-selectin 4 months into treatment, with these decreases persisting through 1 year follow-up. Consistent with the HUSTLE study, we did not observe differences in sVCAM-1 or sICAM-1. These data demonstrate that hydroxyurea selectively affects the endothelium and may provide insights into the mechanisms of protection.

Hydroxyurea may indirectly affect Angpt-2 levels by increasing Hb and HbF, resulting in less erythrocyte sickling and reducing WBC counts. However, hydroxyurea has also been shown to have direct effects on the endothelium through decreased neutrophil adhesion and increased nitric oxide bioavailability.⁴⁵ E-selectin expression is specific to the endothelium and is transcriptionally induced by inflammatory cytokines (tumor necrosis factor α and interleukin-1β) and P-selectin. Murine data confirm a selective decrease in sE-selectin associated with hydroxyurea treatment, independent of changes in markers of inflammation, and associated with reduced neutrophil recruitment and tissue injury in pneumococcal infection.⁴² In contrast to E-selectin, P-selectin and Angpt-2 are stored in Weibel-Palade bodies and rapidly released in response to endothelial activation in response to exogenous stimuli (ie, tumor necrosis factor α and thrombin) and environmental cues (eg, hypoxia),^11,46 and are inhibited in response to nitric oxide production.⁴⁷ 

There was a strong association between Angpt-2 levels at enrollment and TCD velocities measured at 1 year follow-up; to our knowledge, the first such report in a pediatric SCA cohort. TCD measures blood flow velocity in the middle cerebral and internal carotid arteries, with increases in blood flow velocity associated with increased stroke risk in children, which may be attributable to hyperemia related to impaired vasodilatation.^48,49 Hydroxyurea is associated with reduced TCD velocity and reduced stroke risk.^17,50,51 In the NOHARM cohort, hydroxyurea was associated with a median TCD velocity reduction of 12 cm/s after 1 year of treatment.¹⁸ Reduced TCD velocities are thought to be driven by increased HbF and total Hb expression and myelosuppression (reduced leukocyte and platelet counts).⁴⁵ Although our findings require validation, Angpt-2 levels represent a potential biomarker to risk stratify children in settings in which TCD testing is not available.

Mediation analysis suggests that hydroxyurea reduces TCD velocities, in part, through increased HbF and total Hb. Mediation analysis further suggested possible Angpt-2–mediated pathways involving both Hb and WBC count, which approached statistical significance. Mechanistically, we hypothesize that both the hydroxyurea-mediated myelosuppression and increased Hb/HbF results in improved oxygenation of red blood cells and less endothelial activation. It is important to note that hydroxyurea may also reduce TCD velocities by directly affecting the endothelium.

Study strengths include the randomized design and serial measures of endothelial activation. We identified possible causal pathways, suggesting that hydroxyurea’s impact on Angpt-2 may be mediated through increased Hb. Although this mediation analysis has limitations on implied causality, we posit a biologically plausible mechanism through which hydroxyurea-mediated increases in Hb, decrease Angpt-2 levels and stroke risk. More study is needed to validate these potential pathways and explore other possible mechanisms.

In conclusion, in this study we found that hydroxyurea treatment in children with SCA leads to persistently decreased markers of endothelial activation. These data suggest that serial measures of Angpt-2 could be used to measure clinical response to hydroxyurea therapy, may serve as a risk stratification marker for stroke, and could be an additional target for SCA therapies.

The authors thank the study participants and their families who volunteered to participate in this study, the Mulago Hospital Sickle Cell Clinic, Global Health Uganda, and the Data Coordinating Center at Cincinnati Children’s Hospital Medical Center.

The Doris Duke Charitable Foundation provided funds for the study (ICRA 2013139), and Addmedica Inc (Paris, France) donated hydroxyurea (Siklos) for the NOHARM trial. Biomarker evaluation was supported by a Doris Duke Charitable Foundation fellowship (A.C.) and a Fogarty International Center Global Health fellow grant (D43 TW009345 to F.S.).

Contribution: R.O.O., R.E.W., and C.C.J. designed the study, supervised the trial, and analyzed the results; A.L.C. designed and conducted the biomarker testing with A.C. and T.F.S.; T.L. coordinated many critical aspects of the trial to ensure its safe and successful operational execution and participated in the analysis; M.N. conducted the TCD analysis; A.L., J.M.S., K.M., and A.L.C. performed statistical analyses; T.F.S. and A.L.C. wrote the first draft of the manuscript; and all authors participated in the editing of the manuscript and approved the final version.

Conflict-of-interest disclosure: R.E.W. is a consultant for Nova Laboratories; receives research support (drug donation) from Bristol Myers Squibb and Addmedica; and serves on 2 data and safety monitoring boards. A.L.C. is an inventor on a patent related to life-threatening responses to infection that includes angiopoietin-2. None of these disclosures are relevant to the results and conclusions of the NOHARM trial. The remaining authors declare no competing financial interests.

Correspondence: Andrea L. Conroy, Indiana University School of Medicine, Ryan White Center for Pediatric Infectious Diseases and Global Health, 1044 W. Walnut St, Indianapolis, IN 46202; email: [email protected].