Identification of genetic polymorphisms associated with risk for pulmonary hypertension in sickle cell disease

One of the most serious complications that can occur in adult patients with sickle cell disease (SCD) is pulmonary hypertension (pHTN). Recent reports have suggested that between 5% and 30% of adult patients with SCD develop pHTN during the course of their disease.^1-5  One recent retrospective review of an adult population with SCD found that 50 (12%) of 414 patients had echocardiographic findings consistent with pHTN.³  In that study, patients with pHTN also had significantly lower pulse oximetry O₂ saturation (< 75% vs > 90%) and, most importantly, severely shortened survival. Castro et al⁶  found that the median survival for patients with pHTN was only 25.6 months after diagnosis by cardiac catheterization. This was significantly shorter than that for patients without pHTN, who had more than 70% survival at the end of the 119 months after cardiac catheterization. In a review of 306 autopsies of patients with SCD accrued between 1929 and 1996, pHTN contributed to approximately 3% of all causes of death.⁷  However, this number has likely increased in the past several decades due to a decrease in deaths due to infections.⁸  In fact, in our study population, 21% of the adult patients (18 years and older) with pHTN died within 3 years, compared with only 4% of the patients without pHTN.⁹  In a longitudinal study of patients with SCD with pHTN, risk for mortality was independent of the degree of the severity of the pHTN as well.¹⁰ 

In the present report, we examined 297 single nucleotide polymorphisms (SNPs) in 49 candidate genes to identify genetic polymorphisms associated with risk for pHTN in our patients with SCD. Most of the genes we examined are believed to be involved in vaso-occlusion through the processes of adhesion, signaling, transport, or coagulation. In addition, we examined genes involved in nitric oxide (NO) biology (NOS2 and NOS3), as NO has been hypothesized to be involved in the pathophysiology of pHTN in SCD.¹¹  Furthermore, we also examined genes that have been previously implicated in risk for primary pHTN, such as bone morphogenic protein receptor 2 (BMPR2),¹²  and risk for pHTN associated with hereditary hemorrhagic telangiectasia, activin-receptor-like kinase 1 (ACVRL1).¹³ 

Patients were ascertained as part of a multicenter study to identify genetic factors associated with adult-onset SCD complications. Blood was obtained from all patients, and DNA was extracted using standard protocols at the DNA Bank of the Duke Center for Human Genetics. We ascertained 518 unrelated, adult (aged 18 years and older) patients with sickle cell anemia (Hb SS), Hb SC, Hb Sβ⁺-thalassemia, and Hb Sβ°-thalassemia. All patients provided written informed consent, and the study was approved by the Institutional Review Boards of Duke University and University of North Carolina. Comprehensive clinical and laboratory data were collected on all patients. The entire data set of 518 individuals for whom genetic data were available was used to assess Hardy-Weinberg equilibrium for each polymorphism, as well as to estimate linkage disequilibrium (LD) among polymorphisms within the same gene.

Subjects who underwent screening for pHTN were regular outpatients from the Duke University Medical Center (DUMC) and the University of North Carolina Chapel Hill (UNC-CH) Comprehensive Sickle Cell Centers. At DUMC, 71 patients were evaluated by echocardiography (echo) due to symptoms possibly attributable to pulmonary or cardiac compromise. At UNC-CH, 40 patients were recruited randomly to be evaluated by echocardiography. Although the 2 institutions differed in their approach for identifying patients to be evaluated for pHTN, there was no statistical difference in the percentage of patients (approximately 40%) observed to meet criteria for pHTN at each institution. pHTN was not more prevalent at DUMC, as might have been expected. Thus, the different screening approaches did not lead to an apparent selection bias that would preclude including patients from both institutions in a single data set for analysis. This observation could reflect the current lack of suitable clinical parameters for detecting patients at risk for pHTN.

At both institutions, pHTN was defined by the presence of a tricuspid regurgitation jet velocity (TRjet) of 2.5 m/second or more. Echo studies reported as either normal or abnormal but not consistent with pHTN were coded as negative for pHTN. A total of 42 patients (38% of those screened) were found to have pHTN by echo, and another 2 by right heart catheterization, which was performed after echo exams failed to demonstrate pHTN despite high clinical suspicion.

Statistical analysis was performed on a total of 111 patients with Hb SS who had been evaluated for pHTN according to the procedures described. As the decision to screen for pHTN was made independently of the conduct of the larger genetic study, there was no target enrollment for the current study. In addition, the sample size varied somewhat for the analysis of each SNP, due to slight differences in missing genotype data. All subjects were adults (aged 18 years and older) with Hb SS who had undergone evaluation for pHTN as described. For analyses of “data set 1” (see “Results”), cases included all patients who had evidence of pHTN by echo or right heart catheterization, while “controls” included all patients whose echo showed a TRjet that was less than 2.5 m/second or undetectable and who did not have pHTN by right heart catheterization (if done). For “data set 2,” “cases” were the same as in data set 1, while controls were limited to patients whose echos showed no pHTN or significant left heart abnormalities.

SNPs within each candidate gene were identified using the National Center for Biotechnology Information (NCBI) Single Nucleotide Polymorphism database (dbSNP) (http://www.ncbi.nlm.nih.gov/SNP)¹⁴  and by review of the medical literature. Except for the gene encoding the β2 adrenergic receptor (ADRB2), SNP genotyping was performed by TaqMan, using Assays-on-Demand or Assays-by-Design SNP genotyping products (Applied Biosystems, Foster City, CA). For all genotype assays, quality control measures were applied, including genotyping a series of blinded duplicate samples and Center d'Etude du Polymorphism Humain (CEPH) controls. The genotypes of all duplicate samples had to match 100% in order for the assay to pass quality control. Further, we required that each assay achieve 95% efficiency (ie, the genotypes of at least 95% of the samples could be called with certainty) to be considered for statistical analysis. Polymerase chain reaction (PCR) was performed on the ABI 9700 dual 384-well Geneamp PCR system (Applied Biosystems). Genotypes were analyzed using an ABI Prism 7900HT Sequence Detection System (Applied Biosystems). Each reaction contained 2.7 ng of total genomic DNA. Due to the physical proximity, SNPs in ADRB2 were identified through PCR amplification, followed by automated sequencing.

We identified 344 SNPs within or directly flanking 52 genes for analysis. Hardy-Weinberg equilibrium (HWE) was assessed using SAS Genetics (SAS Institute, Cary, NC), and exact tests implemented in the Genetic Data Analysis program (Cornell University, Ithaca, NY).¹⁵  These procedures were carried out primarily for quality control purposes. SNPs found to deviate from HWE (P < .01; n = 13) and SNPs with minor allele frequency less than 10% (n = 34) were excluded from analysis. This resulted in the inclusion of a total of 297 SNPs in 49 genes for our analysis, as seen in Table 1. Pairwise LD measures between markers within each gene were calculated using SAS Genetics. We calculated values for both D′ and r² measures of LD.

For each SNP, contingency tables and tests of association were constructed for the genotypes by pHTN in our initial analysis (data set 1; n = 111). We used an exact genotype test for association. This test is based on a 3-by-2 table of genotypes versus pHTN status. SNPs with P values less than .05 were considered nominally significant.

Given the large number of tests, there was high potential for false discovery. We addressed this in 2 ways. First, we considered nominally significant SNPs as potentially important when they occurred in genes previously associated with primary pHTN, thus reducing the likelihood of false discovery. Second, we adjusted the 297 nominal P values for multiple testing using the false discovery rate (FDR) procedure developed by Benjamini-Hochberg¹⁶  and applied a standard threshold of .10 for declaring significance. This means that on average, 10% of SNPs identified by this FDR procedure as significant will be false-positive discoveries.

We repeated the statistical analysis between the SNPs and the occurrence of pHTN for a subset of the sample (data set 2; n = 73), which excluded individuals whose echos indicated the presence of left heart abnormalities. Although the echos from these individuals were not consistent with pHTN, individuals with left heart abnormalities may also differ genetically from individuals with normal echos. Thus, we believed it was useful to repeat the analysis excluding these individuals. We applied the Benjamini-Hochberg procedure separately to the 297 nominal P values calculated on this subset.

We were also aware that association analyses can be improved by adjusting for important covariates. As shown in Table 3, we observed that pHTN was associated with age and baseline hemoglobin (hgb). hgb was defined as the average of 3 hgb measures (in g/L) taken during steady state and recorded at study entry. Thus, our second analysis used logistic regression to detect genetic associations to pHTN after adjusting for age and hgb. We obtained P values based on the likelihood ratio chi-squared statistic. This was performed on a data set including all patients, data set 1, as well as the subset of patients from which individuals with left heart abnormalities were excluded, data set 2.

In addition to adjusting for age and hgb, certain SNPs might gain or lose significance in a model that adjusted for the effect of other SNPs. For the purpose of multiple regression analysis, we next considered all SNPs that were nominally associated at the .01 level, based on individual covariate adjusted logistic regression, as well as ACVRL1, BMPR2, and BMP6. We used backward selection methods to choose a model that included multiple SNPs that were simultaneously significant at the .01 nominal level, based on the likelihood ratio statistic. For each SNP, we obtained odds ratios and confidence intervals, which were adjusted for age, hgb, and the effect of the other nominally significant SNPs. This modeling was conducted primarily for data set 1, where multiple SNPs were identified by the single SNP logistic regression. Possibly due to the small sample size of data set 2, only a single SNP was identified by backwards logistic regression, and therefore multiple SNP regression was not performed.

For each SNP there were a few, different individuals with missing genotype data. This is relatively unimportant when the SNPs are analyzed individually for association with pHTN; however, a missing genotype for a single SNP results in losing an entire record from a combined logistic analysis. For the purpose of variable selection, we imputed the heterozygote genotype for about 5% of the total genotype observations and thus were able to include many individuals who would otherwise have been excluded. We compared an alternative method of imputing the most common genotype and found that the backward selection procedures selected virtually the same variables regardless of the method of imputation. For the purpose of estimation of odds ratios and evaluation of P values, we used only the complete data where records with any missing genotypes were deleted.

The entire results for the chi-squared analysis of all 297 SNPs are available as supplementary material online at http://wwwchg.duhs.duke.edu/research/support.html (Table S1, available on the Blood website; see the Supplemental Materials link at the top of the online article).

Chi-squared analysis of pHTN by SNP genotype provided evidence for association between pHTN and genes predominantly in the TGFβ family, including ACVRL1, BMPR2, and BMP6 (Table 2). In our overall sample of patients with SCD (data set 1), 3 SNPs in ACVRL1 demonstrated nominal association with pHTN, with chi-squared P values less than .01. The ACVRL1 SNPs with strongest relationship to pHTN were not in LD with each other (r² < .01), suggesting that the signals from these SNPs could not be attributed to the other significant SNPs in the same gene. A total of 2 SNPs in BMPR2 were in high LD (r² = .94) and had nominal P values less than .01. Since both ACVRL1 and BMPR2 have been previously associated with primary pHTN, the nominal significance we observed in our data set is particularly interesting. However, we note that the FDR-adjusted P values for all SNPs in ACVRL1 and BMPR2 were .22, which is not significant at the false discovery threshold of .10.

Evidence for association with pHTN was also found in BMP6. A total of 3 SNPs, in fairly strong LD (r² > .75), had nominal P values less than .01 and FDR P values of .221. We also tested for association in the subset which excluded individuals with abnormal echos without pHTN (data set 2). Since these tests included only 73 individuals, we were not surprised that most associations became nominally insignificant in this smaller subset. However, the association with BMP6 increased (rs267192, P < .001; rs267196, P < .001). Furthermore, the associations with the 2 BMP6 SNPs in data set 2 were significant at the false discovery threshold, with FDR P values of .031 and .038, respectively.

We next compared characteristics of patients with pHTN to those patients without pHTN to detect potential covariates. In our sample, there was a strong relationship between age and pHTN (Table 3). There was also an association between hgb and pHTN, which was reduced after controlling for age, but still moderately significant (P = .03). Thus, we conducted analysis using these covariates.

The logistic regression analysis, including age and hgb as covariates, did not distinguish any newly significant SNPs for data set 1 (Table 4). The associations between pHTN and ACVRL1 and BMPR2 remained nominally significant, though again neither was FDR significant in this analysis. In data set 2, the BMP6 SNP remained highly associated with pHTN (rs267192, P < .001). This was significant after adjusting for multiple testing with FDR (P = .077).

The previous analyses, which focused on individual SNPs, provided a subset of the original candidate genes upon which further analysis could be focused. Using backward selection, we obtained a logistic regression model including age, hemoglobin, and multiple SNPs in ACVRL1 (rs3847859 and rs706814), ADRB1 (rs1801253), BMP6 (rs267192), and TGFBR3 (rs10874940), where all of these SNPs were significant at the .01 nominal level. This model was obtained using the data with imputed missing values. We fit this same model for the data without imputation in order to obtain odds ratios and P values with the observed data only. Even in this reduced data set (removing individuals with any missing data), all of the SNPs that had been nominally significant with imputation maintained significance at the .01 nominal level. SNPs in ACVRL1, BMP6, and ADRB1 had FDR P values less than .10 (Table 5). In data set 2, only a single BMP6 SNP was selected after adjusting for covariates.

Our analyses demonstrate a likely role of genetic variation in ADRB1 and genes in the TGFβ pathway for risk of pHTN. The single SNP adjusted logistic regression had implicated several genes, including ACVRL1, ADRB1, BMP6, BMPR2, LTA4H, SERPINC1, and TGFBR3, all at the nominal significance level (Table 4). When correcting for multiple testing, only BMP6 remained FDR significant in data set 2. Multiple logistic regression allowed us to control for the effects of other SNPs, as well as age and hemoglobin. In this analysis, SNPs in ACVRL1, BMPR6, and ADRB1 became FDR significant, while other SNPs did not even maintain nominal significance and thus were eliminated by backwards selection. Although we used imputation for approximately 5% of the genotype data to construct these models, when we repeated the models using only complete, observed data, the results appeared to be robust. The associations with these genes are particularly interesting because mutations in ACVRL1 have previously been associated with primary pHTN in families who have hereditary hemorrhagic telangiectasia.¹³ ACVRL1 belongs to the TGFβ pathway, as does BMP6. Further, although it did not remain in the final logistical analysis, BMPR2, also part of the TGFβ pathway, showed nominal evidence for association with pHTN in our data set. Mutations in BMPR2 have also been associated with primary pHTN in approximately 50% of familial patients, and 10% to 25% of nonfamilial patients.¹⁷  Finally, TGFBR3 also demonstrated nominal evidence for association in data set 1, and nearly met FDR significance in the multiple SNP logistic regression (Table 5).

Ultimately, of the 5 genes in the TGFβ pathway that were examined in our analysis, 4 demonstrated some level of association with pHTN. Based on our models, the TGFβ superfamily was the primary biological pathway associated with the occurrence of pHTN in our patients with SCD. Interestingly, this has also been true thus far for genes identified for primary pHTN (ACVRL1 and BMPR2). In contrast, we recently described genetic associations with priapism from the same larger multicenter data set,¹⁸  where multiple pathophysiologic pathways were implicated in the occurrence of priapism, including cell hydration, red cell adhesion, coagulation, and the TGFβ superfamily. The fact that a single biological pathway may be the most germane to the development of pHTN has important implications for developing pharmacologic treatments. That is, the TGFβ pathway is an excellent candidate pathway for targeting treatments that may prevent or ameliorate this condition.

While the TGFβ pathway may be the primary pathway contributing to risk for pHTN in SCD, it is noteworthy that this pathway has also been implicated in risk for other sickle cell complications. Genes in the TGFβ pathway have been implicated in risk for stroke,¹⁹  leg ulcers,²⁰  bacteremia,²¹  and priapism¹⁸  in SCD. TGFβ genes have also been associated with glomerular filtration rate (GFR) in SCD.²²  Since GFR is a marker for kidney function, these genes may also place individuals with SCD at risk for developing nephropathy.

Although there is increasing appreciation for the role of hemolytic anemia in NO dysregulation and ultimately the development of complications of SCD such as pHTN, priapism, leg ulcers and stroke,^23-27  we did not detect associations with genes in the NO pathway. However, because SNPs out of HWE could represent the result of a true association, we also analyzed the SNPs which deviated from HWE (n = 13) in the overall data set (data set 1). A total of 3 of these SNPs were nominally associated with pHTN in the chi-squared analysis: SNP rs7921133 in ADRB1 (P = .004), and 2 SNPs in ARG2 (rs12587111, P = .02; and rs1885042, P = .009). ARG2 encodes an arginase that has been implicated in abnormal NO biology in SCD, and higher arginase levels are found in patients with SCD who have pHTN.²⁸  Further investigation will be needed to determine if this association with ARG2 is a true association.

While we are confident of the classification of patients identified by echocardiography with TRjets greater than 2.5 m/second as having pHTN, we also recognize that basing the diagnosis of pHTN on echocardiography alone could potentially be hampered by false-negative results. It has been suggested that diagnosis of pHTN by echocardiography alone could miss up to 20% of subjects with mild pHTN (pulmonary artery systolic pressure lower than 50 mmHg).^29-31  While this may resulted in misclassifying patients with mild pHTN as negative for pHTN in our study, the most likely effect of this misclassification was to reduce our power to detect true associations.

In summary, we identified several genetic variants that are associated with the occurrence of pHTN in patients with SCD. A model including age, genetic variation in ADRB1 and genes in the TGFβ pathway predicted risk for pHTN in our study population. Further evaluation of this model in a prospectively studied cohort of patients with SCD is warranted. Our findings offer new promise for developing therapies and reducing the occurrence of this life-threatening SCD complication.

We thank the patients and families for their participation.

This work was supported by grants R01HL68959, R01HL79915, and U54HL070769 from the National Heart, Lung and Blood Institute and R21DK066605 from the National Institute of Diabetes, Digestive and Kidney Diseases, National Institutes of Health (to M.J.T.). L.D.C. is the Sickle Cell Scholar for the Duke-UNC Comprehensive Sickle Cell Center (HL070769).

National Institutes of Health

Contribution: A.E.A.-K., L.D.C., and M.J.T. designed the research and wrote the paper. L.E. and M.K. analyzed data and wrote the paper. J.J., T.L.J., J.P., K.I.A., and E.P.O. performed the research. J.M.V. supervised genotyping. M.C.L. and J.B.W. contributed reagents and expertise regarding nitric oxide pathways. A.C. wrote the paper.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Allison Ashley-Koch, Center for Human Genetics, Duke University Medical Center, Box 3400, 2007 Snyderman Genomic Sciences Bldg, 595 LaSalle St, Durham, NC 27710; e-mail: allison.ashleykoch@duke.edu.