AbstractClinical and hematologic characteristics of beta(β)-thalassemia are determined by several factors resulting in a wide spectrum of severity. Phenotype modulators are: HBB mutations, HBA defects and fetal hemoglobin production modulators (HBG2:g.−158C>T polymorphism, HBS1L-MYB intergenic region and the BCL11A). We characterized 54 genetic variants at these five loci robustly associated with the amelioration of beta-thalassemia phenotype, to build a predictive score of severity using a representative cohort of 890 β-thalassemic patients. Using Cox proportional hazard analysis on a training set, we assessed the effect of these loci on the age at which patient started regular transfusions, built a Thalassemia Severity Score, and validated it on a testing set. Discriminatory power of the model was high (C-index=0.705; R2=0.343) and the validation conducted on the testing set confirmed its predictive accuracy with transfusion-free survival probability (P<0.001) and with transfusion dependency status (Area Under the Receiver Operating Characteristic Curve=0.774; P<0.001). Finally, an automatized on-line calculation of the score was made available at http://tss.unica.it. Besides the accurate assessment of genetic predictors effect, the present results could be helpful in the management of patients, both as a predictive score for screening and a standardized scale of severity to overcome the major-intermedia dichotomy and support clinical decisions.
Clinical and hematologic characteristics of beta(β)-thalassemia are determined by several factors resulting in a wide spectrum of severity from no need to dependence on regular blood transfusions; they mainly include the type of disease causing mutation and the capacity production of alpha(α)-and gamma(γ)-globin chains. Mutations in the HBB gene determine lack or low level of β-globin chains synthesis in the erythropoietic cells, causing an imbalance of the α- to β-globin chains ratio. This imbalance is especially evident in patients homozygous for HBB mutations in whom the accumulation of unbound α-globin chains, forming highly toxic aggregates that precipitate in the erythroid precursor of the bone marrow, leads to ineffective erythropoiesis resulting in severe anemia due to low red blood cell survival from ineffective erythropoiesis and hemolysis. The clinical presentation is widely variable because the amount of unbound α-globin chains can be modified by both the capacity to produce α-globin chains (HBA genes variants) and the capacity to produce γ-globin chains (HBG2 gene modulators) that can bind available α-globin chains to form effective fetal hemoglobin (HBF).21 The severity of mutations in the HBB gene and defects of the HBA genes were the first determinants of the phenotype variability of β-thalassemia to be discovered. The third determinant to be identified was the XmnI polymor phism of the HBG2 promoter (HBG2:g.−158C>T) which is widely included in the diagnostic workup for thalassemia patients.43 More recently genome-wide association studies contributed to the definition of very important trans-acting modifiers of the production of fetal hemoglobin such as the BCL11A gene and the HBS1L-MYB intergenic region.105
Previous studies investigated the contribution of known genetic modifiers to the major-intermedia phenotypic classification demonstrating that a great proportion of this status is effectively genetically determined.1211 This condition makes β-thalassemia one of the few diseases with complex phenotype that could be accurately predicted on a genetic basis, opening the path for the clinical application of genetic prediction to many other diseases. Recently, we demonstrated the feasibility of a more accurate prediction of the hematologic severity of β-thalassemia, using known genetic modifiers to predict the age of the start of regular blood transfusions in a cohort of homozygous β-39 individuals (HBB:c.118C>T).13 This approach has two main benefits: 1) the severity is measured on a simple and reproducible scale that precisely assesses the overall spectrum of hematologic severity, therefore overcoming the major-intermedia dichotomy; and 2) the factors considered are exclusively genetic allowing prediction from a unique measurement already available during pregnancy, and have a wide range of clinical applications, such as screening, genetic scoring and assistance in therapeutic decisions.13
The aim of the present study was to build and validate a predictive score of severity based on known genetic markers, using a large representative cohort of β-thalassemic patients. In six centers from three countries of the Mediterranean basin, we recruited 890 patients for whom the date at which regular transfusion started was accurately registered. The whole cohort was genotyped for six markers at five loci that were already robustly associated with both HBF levels and/or the amelioration of β-thalassemia phenotype.18146 Using Cox proportional hazard analysis for the age of regular transfusion start, we built a Thalassemia Severity Score (TSS) from genetic markers, tested its predictive ability, and, finally, made it available online at: http://tss.unica.it
Study sample and phenotye assessment
For this multi-center study, an international cohort of 890 β-thalassemia patients was recruited from the pediatric and adult departments of the Microcythemia Hospital of Cagliari, the San Luigi Gonzaga Hospital of Turin, the Hematology Center for Microcythemia and Congenital Abnormalities, at Galliera Hospital, Genoa, Italy, the French Reference Center for Thalassemia from Marseille, and the Thalassaemia and Molecular Genetics Clinic of the Mater Dei Hospital from Malta.
Overall data were pooled into a single dataset composed of 49.9% females and 50.1% males, representing 45 β-thalassemic mutations (65 β-thalassemic genotypes) from 88.24% transfusion-dependent thalassemia (TDT) and 11.76% non-transfusion-dependent thalassemia (NTDT) patients. The time elapsed between birth and the start of regular transfusions was continuously distributed and reflected the variability of the hematologic severity among uncensored thalassemia patients (TDT, median transfusion-free survival time of 11 months), whereas the time between birth and last follow up was taken into account for censored patients (NTDT, median survival time 44 years). Data relative to the transfusion program of the patients were collected through the WebTHAL computerized clinical records database (http://www.thalassemia.it) in use for the daily management of patients in Italian centers, and through direct transmission of data regarding the other two centers. Criteria to start transfusion were: hemoglobin level lower than 7 g/dL for more than two weeks in absence of infections, moderate to severe spleen enlargement, initial skeletal changes and/or poor growth. Patients were considered as regularly transfused when undergoing more than eight blood transfusions a year.
This retrospective study was registered with local ethical committees and conducted in accordance with the Declaration of Helsinki; all patients gave informed consent for DNA analysis and research study.
Marker selection and genotyping
We selected a set of genetic markers robustly associated with the disease severity of β-thalassemia to use as genetic predictors: the type of HBB gene mutation, the HBG2:g.−158C>T polymorphism, the type and number of HBA genes defects, two SNPs from the second intron of BCL11A gene (rs1427407 and rs10189857) and one from the HBS1L-MYB intergenic region (rs9399137). These markers have been independently associated to both HBF levels and the amelioration of β-thalassemia phenotype in different studies (for details see Table 1, Online Supplementary Methods and Online Supplementary Table S1).1813643
DNA was extracted with standard methods from venous peripheral blood. Mutation analysis of the β-globin gene was performed by direct DNA sequencing and discovered 45 different mutations (Table 1 and Online Supplementary Table S1). α-globin gene defects considered [3.7 kb rightward type I, II and III (NG_000006.1:g.34164_37967del3804, HbVar ID: 1078 and 1079, respectively), 4.2 kb leftward (HbVar ID: 1079), Mediterranean type I (NG_000006.1:g.24664_41064del16401), 20.5 kb (NG_000006.1:g.15164_37864del22701) large deletions, as well as HphI (HBA2:c.95+2_95+6delTGAGG) small deletion and NcoI (HBA2:c.2T>C) polymorphism, see Table 1] were determined using GAP-PCR or restriction enzyme digestion for large deletions and other defects, respectively.19 The HBG2:g.−158C>T (rs7482144) polymorphism was determined as previously described, while other SNPs were directly genotyped using TaqMan SNP genotyping assay (Applied Biosystems, Warrington, UK).20
Accurate data quality controls were performed on the dataset, including consistency check of data and missing value analysis, after which 31 samples were removed, leaving 859 samples with complete data. All statistical procedures were performed using SPSS v.20.0 software package (SPSS Inc., Chicago, Illinois, USA).
The cohort was divided into a training set (71% selected randomly) and a testing set (the remaining 29%). The training set was used to build a Cox proportional hazard model for the age of regular transfusion start using genetic markers as predictors, and the testing set was used to test the predictive ability of the model using Cox proportional hazard analysis, Kaplan-Meier survival analysis and ROC curves for TDT-NTDT status prediction. The choice of Cox proportional hazard model was driven by the fact that death for all causes is a negligible competing event with respect to start of transfusion in β-thalassemia patients.21 For score elaboration, all variables were treated as categorical: sex was codified 0 when female and 1 when male, HBB mutations were codified upon their severity as reported in literature (http://globin.bx.psu.edu/hbvar) in mild/mild, mild/severe or severe/severe (from 1 to 3), HBA gene defects were classified as 0, 1, or 2 according to the number of deleted or mutated copies of the HBA gene, while for each SNP a variable was defined with the value of 0, 1, or 2 according to the number of copies of the less frequent allele (Table 1). For survival analysis, center of origin was used as a stratification variable, and patients were considered uncensored when blood transfusion occurred or censored when blood transfusion did not occur. We report Cox and Snell R due to low proportion of early censoring as well as Harrell’s concordance index (C-index) to assess how well the model performed. Finally, after validation on the testing set to check prediction accuracy, we derived a 0–10 scale of severity by standardizing the sum of Cox model hazard ratios as detailed in Online Supplementary Methods.
The Thalassemia Severity Score web-tool (http://tss.unica.it) is freely available to any user to calculate the TSS and plot the transfusion-free survival curve corresponding to the entered genotypes. The web tool calculates the TSS by transforming the sum of preloaded linear predictor scores as reported in the Online Supplementary Methods. The calculated TSS is then associated to one of the four statistically different survival curves observed in the studied cohort for a graphical presentation of the prediction.
The distribution of selected genetic markers in the cohort is presented in Table 2. The genotypes from HBB gene mutations were 65 in total and the most frequent were HBB:c.118C>T/HBB:c.118C>T (65.5%), HBB:c.93–21G>A/HBB:c.118C>T (5.6%) and HBB:c.92+6T>C/HBB:c.118C>T (4.1%). The most common HBA genes defect was the 3.7 kb rightward deletion (22.0% with one deletion and 6.5% with two), among the four defects observed. Regarding other loci, the minor allele frequencies were 3% for rs7482144 (HBG2:g.−158C>T), 18% for rs9399137 (HBS1L-MYB intergenic region), 21% for rs1427407 and 40% for rs10189857 (BCL11A). HBB genes defects frequencies differed between centers. In particular, in Italian centers we mainly observed samples homozygous for HBB:c.118C>T and heterozygous HBB:c.93–21G>A/HBB:c.118C>T and HBB:c.92+6T>C/HBB:c.118C>T; in non-Italian centers the HBB:c.92+6T>C/HBB:c.92+6T>C genotype prevailed in Malta, and the HBB:c.93−21G>A/HBB:c.93–21G>A genotype in Marseille. In Sardinia, the mutation observed was almost exclusively HBB:c.118C>T/HBB:c.118C>T.
Results from the Cox proportional hazard model built on a random sample of 71% of the cohort (training set, n=613) are presented in Table 3. The type of HBB mutation had the strongest effect on the hematologic severity of β-thalassemia with a 30-fold increased risk of transfusion start for severe/severe phenotype versus mild/mild (mild/mild: HR=0.032; P<0.001), followed by HBG2:g.−158C>T polymorphism (homozygous T/T: HR=0.052; P<0.001), rs9399137 (HBS1L-MYB; homozygous C/C: HR=0.158; P<0.001), HBA gene defects (two defects: HR=0.252; P<0.001), rs1427407 (BCL11A; homozygous T/T: HR=0.362; P<0.001), rs10189857 (BCL11A; homozygous G/G: HR=0.689; P=0.014) and sex (males: HR=0.825; P=0.033). The discriminatory power of the model was high (C-index=0.703; R=0.343) and most of it was attributable to β-globin (C-index=0.581, R=0.105) and to modulators of the γ-globin chain production (HBG2:g.−158C>T, BCL11A and HBS1L-MYB loci: C-index=0.607, R=0.133), whereas the remaining was attributable to defects in the production of α-globin chains and to sex (C-index=0.569, R=0.056 and C-index=0.532, R=0.010, respectively).
Validation of the model was conducted on the testing set (29% of the cohort, 246 cases) to confirm its predictive accuracy. Cox proportional hazard model highly associated the linear predictor score with transfusion-free survival probability (P<0.001). As expected, the model also demonstrated strong ability to predict TDT versus NTDT status (Area Under the ROC Curve: AUC=0.769; P<0.001).
A score, called Thalassemia Severity Score (TSS), was derived from the linear predictor score distribution to obtain a scale of increasing severity ranging from 0 to 10. As for the original model mentioned above, when applied to the testing set, as expected, the derived score also highly correlates with the probability of transfusion-free survival (P<0.001) and TDT versus NTDT status (AUC=0.769; P<0.001). Finally, four significantly different survival curves can be drawn from the distribution of TSS (P<0.001) (for both sets see Figure 1). Details of the four groups are presented in Online Supplementary Table S2. The TSS calculation has been automated and is freely available at http://tss.unica.it.
This multicenter study involving 890 β-thalassemia patients of six different centers from three countries of the Mediterranean basin demonstrated the ability to accurately score and predict the hematologic severity of the disease using genetic markers and the patient’s age at the start of regular transfusion.
Besides the accurate assessment of the effect of each genetic predictor on the disease, the present results could also be helpful in patient management. The prediction ability of the score could make a significant contribution to pre-conception screening by adapting genetic counseling according to score; however, its use after birth as a hematologic severity scale should be intended as a support to the wide range of clinical information available to medical staff.
In the present study, we created an easy-to-interpret score that uses a scale of increasing severity ranging from 0 to 10. The TSS can be used as a pre-natal predictive score to refine genetic counseling by completing β-thalassemia pre-conception and pre-natal screening whenever available, and to evaluate the severity of the disease in support of the clinically difficult decisions as to whether to start or not start a transfusion program. After birth, it could be used as a standardized scale to better define the hematologic severity of patients and overcome the major-intermedia dichotomy. Such a scale could be of great support when deciding whether to perform a stem cell transplantation: the latter are more efficient if performed early in life (especially with cord blood) but the disease phenotype can take four or even five years to stabilize, and in such a situation the possibility of anticipating the phenotype could be essential. The TSS could also be useful to sustain clinical decisions after the transfusion program has started, to further develop more complete severity scales that would incorporate clinical measurements with the TSS, or even to identify outlier patients by checking their clinical severity status against expected hematologic severity based on genetic background. Another application we envisage is the possibility to better categorize thalassemia patients enrolled in clinical trials for erythropoiesis stimulating factors. Using only genetic markers, the TSS needs a single assessment, is already available during pregnancy, has life-long validity, and could, therefore, be a good solution to maintain an informative profile as to patient’s predisposition for severity. This is important as, in order to improve quality of life, transfusions are increasingly used for patients who, in the past, would not have been transfused, and this will probably result in a drastic reduction in non-transfused cohorts in the future.
The benefits of using only genetic predictors are important, but there are also drawbacks. First, some centers might not have the necessary facilities to genotype the markers. Second, not all markers have proven their causality: if HBB and HBA genes defects are evidently responsible for lack or malfunction of the coded proteins, the direct implication and mechanisms of action of rs1427407, rs10189857, rs9399137 and even rs7482144 have not yet been fully demonstrated. These markers could be surrogates of the causative variants, and even though the final statistical model would not vary much due to the great amount of genetic association signal explained by these variants, the score should be up-dated if a different variant proved to be causative. Third, even if the populations in which these variants were successfully associated represent all the main ethnic groups, not all the populations have been tested and some particular linkage disequilibrium structures might render the model less accurate in untested populations.1814107 However, considering only countries with efficient health systems and screening programs, this retrospective study is a robust tool against mortality-driven bias. In such a context, mortality rates before ten years old are negligible (0 in our cohort), while over 90% of study patients were diagnosed, followed-up and treated before this age.2221 However, only cross validation from prospective studies, also including other thalassemia populations, will ensure full external validity of the score.
The Authors wish to thank Susan Bateman for her careful English proofreading and are grateful for Professor Alexander E. Felice and Professor Christian A. Scerri for providing patients’ DNA from Malta.
- The online version of this article has a Supplementary Appendix.
- Funding This work is dedicated to the memory of Renzo Galanello, who inspired and almost completed the study at the end of an exemplary career as clinician and researcher.
- Authorship and Disclosures Information on authorship, contributions, and financial & other disclosures was provided by the authors and is available with the online version of this article at www.haematologica.org.
- Received July 16, 2014.
- Accepted December 2, 2014.
- GeneReviews [Internet]. University of Washington, Seattle: Seattle (WA).Google Scholar
- Danjou F, Anni F, Galanello R. Beta-tha-lassemia: from genotype to phenotype. Haematologica. 2011; 96(11):1573-1575. PubMedhttps://doi.org/10.3324/haematol.2011.055962Google Scholar
- Gilman J, Huisman T. DNA sequence variation associated with elevated fetal G gamma globin production. Blood. 1985; 66(4):783-787. PubMedGoogle Scholar
- Labie D, Dunda-Belkhodja O, Rouabhi F. The -158 site 5′ to the G gamma gene and G gamma expression. Blood. 1985; 66(6):1463-1465. PubMedGoogle Scholar
- Garner C, Tatu T, Reittie JE. Genetic influences on F cells and other hematologic variables: a twin heritability study. Blood. 2000; 95(1):342-346. PubMedGoogle Scholar
- Lettre G, Sankaran VG, Bezerra MAC. DNA polymorphisms at the BCL11A, HBS1L-MYB, and beta-globin loci associate with fetal hemoglobin levels and pain crises in sickle cell disease. Proc Natl Acad Sci USA. 2008; 105(33):11869-11874. PubMedhttps://doi.org/10.1073/pnas.0804799105Google Scholar
- Sedgewick AE, Timofeev N, Sebastiani P. BCL11A is a major HbF quantitative trait locus in three different populations with beta-hemoglobinopathies. Blood Cells Mol Dis. 2008; 41(3):255-258. PubMedhttps://doi.org/10.1016/j.bcmd.2008.06.007Google Scholar
- So C-C, Song Y-Q, Tsang ST. The HBS1L-MYB intergenic region on chromosome 6q23 is a quantitative trait locus controlling fetal haemoglobin level in carriers of beta-thalassaemia. J Med Genet. 2008; 45(11):745-751. PubMedhttps://doi.org/10.1136/jmg.2008.060335Google Scholar
- Creary LE, Ulug P, Menzel S. Genetic variation on chromosome 6 influences F cell levels in healthy individuals of African descent and HbF levels in sickle cell patients. PloS One. 2009; 4(1):e4218. PubMedhttps://doi.org/10.1371/journal.pone.0004218Google Scholar
- Menzel S, Thein SL. Genetic architecture of hemoglobin F control. Curr Opin Hematol. 2009; 16(3):179-186. PubMedhttps://doi.org/10.1097/MOH.0b013e328329d07aGoogle Scholar
- Galanello R, Sanna S, Perseu L. Amelioration of Sardinian beta0 thalassemia by genetic modifiers. Blood. 2009; 114(18):3935-3937. PubMedhttps://doi.org/10.1182/blood-2009-04-217901Google Scholar
- Badens C, Joly P, Agouti I. Variants in genetic modifiers of β-thalassemia can help to predict the major or intermedia type of the disease. Haematologica. 2011; 96(11):1712-1714. PubMedhttps://doi.org/10.3324/haematol.2011.046748Google Scholar
- Danjou F, Anni F, Perseu L. Genetic modifiers of β-thalassemia and clinical severity as assessed by age at first transfusion. Haematologica. 2012; 97(7):989-993. PubMedhttps://doi.org/10.3324/haematol.2011.053504Google Scholar
- Menzel S, Garner C, Gut I. A QTL influencing F cell production maps to a gene encoding a zinc-finger protein on chromosome 2p15. Nat Genet. 2007; 39(10):1197-1199. PubMedhttps://doi.org/10.1038/ng2108Google Scholar
- Galarneau G, Palmer CD, Sankaran VG. Fine-mapping at three loci known to affect fetal hemoglobin levels explains additional genetic variation. Nat Genet. 2010; 42(12):1049-1051. PubMedhttps://doi.org/10.1038/ng.707Google Scholar
- Solovieff N, Milton JN, Hartley SW. Fetal hemoglobin in sickle cell anemia: genome-wide association studies suggest a regulatory region in the 5′ olfactory receptor gene cluster. Blood. 2010; 115(9):1815-1822. PubMedhttps://doi.org/10.1182/blood-2009-08-239517Google Scholar
- Bhatnagar P, Purvis S, Barron-Casella E. Genome-wide association study identifies genetic variants influencing F-cell levels in sickle-cell patients. J Hum Genet. 2011; 56(4):316-323. PubMedhttps://doi.org/10.1038/jhg.2011.12Google Scholar
- Farrell JJ, Sherva RM, Chen Z-Y. A 3-bp deletion in the HBS1L-MYB intergenic region on chromosome 6q23 is associated with HbF expression. Blood. 2011; 117(18):4935-4945. PubMedhttps://doi.org/10.1182/blood-2010-11-317081Google Scholar
- Galanello R, Sollaino C, Paglietti E. Alpha-thalassemia carrier identification by DNA analysis in the screening for thalassemia. Am J Hematol. 1998; 59(4):273-278. PubMedhttps://doi.org/10.1002/(SICI)1096-8652(199812)59:4<273::AID-AJH2>3.0.CO;2-3Google Scholar
- Sutton M, Bouhassira EE, Nagel RL. Polymerase chain reaction amplification applied to the determination of beta-like globin gene cluster haplotypes. Am J Hematol. 1989; 32(1):66-69. PubMedhttps://doi.org/10.1002/ajh.2830320113Google Scholar
- Ladis V, Chouliaras G, Berdoukas V. Survival in a large cohort of Greek patients with transfusion-dependent beta thalassaemia and mortality ratios compared to the general population. Eur J Haematol. 2011; 86(4):332-338. PubMedhttps://doi.org/10.1111/j.1600-0609.2011.01582.xGoogle Scholar
- Cao A, Kan YW. The prevention of thalassemia. Cold Spring Harb Perspect Med. 2013; 3(2):a011775. PubMedhttps://doi.org/10.1101/cshperspect.a011775Google Scholar