Letters to the Editor 961 bone marrow blasts, 2 with splenomegaly, 7 with diploid REFERENCES cytogenetics and 1 with JAK2-V617F mutation. Whether these 1 Swerdlow SH, Campo E, Harris NL, Jaffe ES, Pileri SA, Stein H et al. (eds). WHO patients share a specific genetic phenotype is conceivable, and Classification of Tumours of Haematopoietic and Lymphoid Tissues. 4th edn. molecular characterization of this subgroup is ongoing. International Agency for Research on Cancer (IARC): Lyon, 2008. Our analysis validates the diagnosis of MDS/MPN-U as a unique 2 Orazi A, Germing U. The myelodysplastic/myeloproliferative neoplasms: myelo- pathologic entity, with distinctive features such as an increased proliferative diseases with dysplastic features. Leukemia 2008; 22: 1308–1319. (15%) incidence of isolated trisomy 8. Despite the relative 3 Vardiman JW, Thiele J, Arber DA, Brunning RD, Borowitz MJ, Porwit A et al. The frequency of trisomy 8 in myeloid malignancies, remarkably little 2008 revision of the World Health Organization (WHO) classification of myeloid 13 neoplasms and acute leukemia: rationale and important changes. Blood 2009; is known about the pathogenic basis of this abnormality. Similar 114: 937–951. ongoing molecular analysis will investigate whether trisomy 8 4 Wang SA, Hasserjian RP, Loew JM, Sechman EV, Jones D, Hao S et al. Refractory associates with particular somatic mutations in this cohort. anemia with ringed sideroblasts associated with marked thrombocytosis harbors Despite statistical significance, the MDS-IPSS model is not ideal, JAK2 mutation and shows overlapping myeloproliferative and myelodysplastic as the majority (68%) of MDS/MPN-U patients had lower-risk scores features. Leukemia 2006; 20: 1641–1644. according to the MDS-IPSS, and yet had poorer survival than their 5 Atallah E, Nussenzveig R, Yin CC, Bueso-Ramos C, Tam C, Manshouri T et al. lower-risk MDS counterparts. Furthermore, the improved survival Prognostic interaction between thrombocytosis and JAK2 V617F mutation in the seen in the Int-1 category compared with the low-risk category is WHO subcategories of myelodysplastic/myeloproliferative disease-unclassifiable contrary to expectation. One prognostic model of clinicopathologic and refractory anemia with ringed sideroblasts and marked thrombocytosis. Leukemia 2008; 22: 1295–1298. variables has recently been developed from a cohort (n ¼ 92) of 6 Cannella L, Breccia M, Latagliata R, Frustaci A, Alimena G. Clinical and patients with either MDS-U or MDS/MPN-U; interestingly, this prognostic features of patients with myelodysplastic/myeloproliferative syndrome 14 cohort did not find platelet count to be of prognostic importance. categorized as unclassified (MDS/MPD-U) by WHO classification. Leuk Res 2008; 32: The MDA global model also provided a significant tool for the 514–516. MDS/MPN-U cohort (P ¼ 0.004). It is noteworthy that the 7 Greenberg P, Cox C, LeBeau MM, Fenaux P, Morel P, Sanz G et al. International MDA model was originally validated within 176 patients with scoring system for evaluating prognosis in myelodysplastic syndromes. Blood CMML and leukocytosis, suggesting that this may be an appropriate 1997; 89: 2079–2088. prognostic model to use in MDS/MPN patient populations. 8 Cervantes F, Dupriez B, Pereira A, Passamonti F, Reilly JT, Morra E et al. No treatment regimen significantly improved response. Given New prognostic scoring system for primary myelofibrosis based on a study of the International Working Group for Myelofibrosis Research and Treatment. Blood their relative novelty, only two patients received JAK2-inhibitor 2009; 113: 2895–2901. therapy. In view of the JAK2-V617 mutations present among the MDS/ 9 Greenberg PL, Tuechler H, Schanz J, Sanz G, Garcia-Manero G, Sole F et al. Revised MPN-U patients signifying overactivity of JAK/STAT signaling, JAK2- international prognostic scoring system for myelodysplastic syndromes. Blood inhibitor therapy may ultimately prove effective, and indeed a clinical 2012; 120: 2454–2465. trial incorporating the combination of ruxolitinib and azacitidine for 10 Kantarjian H, O’Brien S, Ravandi F, Cortes J, Shan J, Bennett JM et al. Proposal for patients with MDS/MPN-U is ongoing at our institution. a new risk model in myelodysplastic syndrome that accounts for events not considered in the original International Prognostic Scoring System. Cancer 2008; 113: 1351–1361. 11 Kantarjian H, Giles F, List A, Lyons R, Sekeres MA, Pierce S et al. The incidence and CONFLICT OF INTEREST impact of thrombocytopenia in myelodysplastic syndromes. Cancer 2007; 109: 1705–1714. The authors declare no conflict of interest. 12 Germing U, Hildebrandt B, Pfeilstocker M, Nosslinger T, Valent P, Fonatsch C et al. Refinement of the international prognostic scoring system (IPSS) by including 1 1 1 1 2 CD DiNardo , N Daver , N Jain , N Pemmaraju , C Bueso-Ramos , LDH as an additional prognostic variable to improve risk assessment in 2 1 1 1 1 CC Yin , S Pierce , E Jabbour , JE Cortes , HM Kantarjian , patients with primary myelodysplastic syndromes (MDS). Leukemia 2005; 19: G Garcia-Manero1 and S Verstovsek1 2223–2231. 1Department of Leukemia, The University of Texas MD Anderson 13 Mertens F, Johansson B, Heim S, Kristoffersson U, Mitelman F. Karyotypic patterns Cancer Center, Houston, TX, USA and in chronic myeloproliferative disorders: report on 74 cases and review of the 2Department of Hematopathology, The University of Texas MD literature. Leukemia 1991; 5: 214–220. Anderson Cancer Center, Houston, TX, USA 14 Liu Y, Tabarroki A, Visconte V, Hasrouni E, Bupathi M, Hamilton BK et al. A Prognostic Scoring System for Unclassifiable MDS and MDS/MPN. ASH Annual E-mail: [email protected] Meeting Abstracts 2012 2012; 120: 1701.

Supplementary Information accompanies this paper on the Leukemia website (http://www.nature.com/leu)

Elucidation of the structural basis of interaction of the BCR-ABL kinase inhibitor, nilotinib (Tasigna) with the human ABC drug transporter P-glycoprotein

Leukemia (2014) 28, 961–964; doi:10.1038/leu.2014.21 transporters P-glycoprotein (P-gp) and ABCG2.1,2 This is clinically important as the transporters not only hamper the bioavailability of these TKIs but may also cause the emergence of drug resistance Nilotinib, imatinib (structures are shown in Supplementary Figure in patients. We have previously shown that imatinib and nilotinib S1) and other tyrosine kinase inhibitors (TKIs) have been shown to interact at the substrate-binding pocket of ABC transporters, be transported by the ATP-binding cassette (ABC) drug but do not interact at the adenosine triphosphate (ATP)

Accepted article preview online 14 January 2014; advance online publication, 7 February 2014

& 2014 Macmillan Publishers Limited Leukemia (2014) 935 – 979 Letters to the Editor 962 sites of these transporters.3 Identification of the key structural and Methods. Crude membranes from High-Five insect cells features of nilotinib and similar TKIs is essential for under- expressing similar levels of mutant proteins (Figure 1b) were used standing their interaction with P-gp. Toward this goal, molecular to determine the interaction of these mutant P-gps with nilotinib. docking, mutational mapping and quantitative structure–activity The effect of nilotinib was evaluated on ATPase activity and on relationship (SAR) were used to identify nilotinib’s binding site photolabeling of mutant P-gps with [125I]-IAAP binding (Figure 1c on P-gp. and Supplementary Table S1), as these approaches can be used to Nilotinib was docked in a human P-gp homology model that determine the interaction of substrates at the substrate-binding was developed on the basis of the mouse P-gp crystal structure4 pocket of P-gp.7,8 Nilotinib’s ability to stimulate the ATPase activity using the XP-Glide docking method to understand the orientation of Y307C-, M949C- and A985C-mutant P-gps was significantly and the complementarity of pharmacophore features of nilotinib reduced or abolished compared with Cys-less WT P-gp with respect to the residues in the drug-binding pocket of (Supplementary Table S1). Similarly, nilotinib’s ability to compete P-gp (Figure 1a). Comparison of binding energy data for the for [125I]-IAAP photolabeling was significantly reduced for Y307C- docked poses of nilotinib at sites 1–4 (ref. 5) suggested site-1 and almost completely lost for M949C- and A985C-mutant P-gps (QZ59-RRR site)4,6 as the most favorable site (binding energy score (Figure 1c and Supplementary Table S1). These observations of À 9.52 kcal/mol). The binding pocket is lined by residues that provided experimental support to the in silico docking studies. The form electrostatic and hydrophobic contacts with a pyridine, a residues Y307, M949 and A985 contribute to nilotinib binding, pyrimidine, a methyl-substituted phenyl ring, the carbonyl oxygen indicating that site-1 may be the primary binding site for nilotinib atom of the amide linker and the trifluoromethylphenyl ring on P-gp. In silico introduction of these mutations in the homology of nilotinib (Figure 1a). Among these, the Y307 residue showed model helped to visualize the local changes in the binding pocket significant interaction through hydrogen bonding to the pyridine (Supplementary Figure S3). In the nilotinib-docked model of P-gp, ring (–N—HO–Y307, 2.4 Å), whereas A985 had hydrophobic pyridine nitrogen was present at a position 2.4 Å from the side contact with the CF3 group (3.3 Å), phenyl ring (3.2 Å) and chains of Y307; M949 was 5.1 Å from the imidazole ring, while imidazole ring (4.1 Å) of nilotinib. Furthermore, M949 also showed A985 was 4.1 Å from the imidazole ring of nilotinib (Figure 1). hydrophobic contact with the imidazole ring (5.1 Å) of nilotinib In the triple mutant, the pyridine nitrogen atom lost one critical (highlighted in red in Figure 1a). Therefore, the residues (Y307, hydrogen bonding interaction with the Y307 residue, increasing M949 and A985) that interact with three major functional groups the distance to 5.9 Å (Supplementary Figure S3). Similarly, the (pyridine, CF3 and imidazole) of nilotinib were selected for further hydrophobic interactions with the imidazole ring and the analysis. The docking studies indicated these residues might trifluoromethyl aniline moiety were lost when M949 and A985 determine the orientation and stabilization of nilotinib within the were mutated to a hydrophilic cysteine residue (Supplementary substrate-binding site of P-gp. These residues were mutated to Figure S3). These data, taken together, provide clear evidence that Cys residue in a Cys-less P-gp to verify their role in interaction site-1 is indeed the primary site of nilotinib binding on P-gp, with with nilotinib. Control Cys-less wild-type (WT) P-gp, Y307C, M949C Y307 interacting with the pyridine ring, A985 interacting with the and A985C P-gp mutants were expressed in HeLa cells trifluoromethylphenyl group and M949 interacting with the (Supplementary Figure S2; mutants exhibited similar expression imidazole ring of nilotinib. and transport of fluorescent substrates as Cys-less WT P-gp) and To further validate the importance of functional groups of High-Five insect cells, as described in Supplementary Materials nilotinib for interacting with P-gp, five structural derivatives of

Figure 1. Docking of nilotinib in the drug-binding pocket of human P-gp and analyses of mutant proteins. (a) Glide-predicted binding pocket of nilotinib in a homology model of human P-gp. Nilotinib was docked in a human P-gp homology model using Glide, as described in Supplementary Materials and Methods. The amino acids that contribute to nilotinib’s binding site are shown here. Three residues (Y307, M949 and A985) used for mutational analyses are highlighted by red boxes. The predicted distance of these residues from the closest functional group of nilotinib is marked. (b) Expression of mutant P-gps. Colloidal blue stain of crude membrane protein (10 mg/lane) from Cys-less WT, Y307C, M949C and A985C P-gps expressed in High-Five insect cells. (c) Nilotinib does not inhibit the labeling of mutant P-gps with [125I]-IAAP. A representative autoradiogram from three independent experiments with Cys-less WT, Y307C-, M949C- and A985C-mutant P-gps photo- 125 crosslinked with [ I]-IAAP in the absence or presence of 5 mM nilotinib is shown.

Leukemia (2014) 935 – 979 & 2014 Macmillan Publishers Limited Letters to the Editor 963 conformation and the mutational mapping data. Whereas the pyridine and pyrimidine moieties of nilotinib are important for interaction at the drug-binding pocket, the imidazole group is not critical for this interaction. Nilotinib and imatinib were also compared for their binding orientation in the substrate-binding pocket of P-gp (Supplementary Figure S4). As described in Supplementary Results, the observed affinity differences between nilotinib and imatinib for P-gp can be explained on the basis of the above docking analysis and its comparison with the known crystal structures of imatinib and nilotinib bound to BCR-ABL kinase.9,10 Several studies have used a docking-based approach to identify the substrate-binding pocket in P-gp, but most of those studies relied on either SAR or mutagenesis alone (reviewed in Palmeira et al.11). We used a two-pronged approach, where the docked orientation of nilotinib was not only validated by directed mutagenesis of selected residues but also was verified using the structural derivatives of nilotinib. Although the data derived from modeling and mutational studies with nilotinib and its derivatives corroborate well with the docked conformation of nilotinib, there is still a possibility that nilotinib may bind to a secondary site because of the chemical and structural flexibility of a large drug- binding pocket that can accommodate more than one ligand simultaneously.12–16 In recent years, multidrug resistance-linked transporters have gained considerable attention as potential targets to improve Figure 2. Synthesis of nilotinib derivatives and characterization of their interaction with P-gp. (a) Chemical structures of nilotinib and cancer chemotherapy and to increase bioavailability/tissue its derivatives used in this study. Nilotinib and derivatives 1, 2, 3, 4 penetration of drugs. Therefore, the interaction of these transpor- and 5 were synthesized as described in Supplementary Materials ters with targeted therapeutic drugs such as nilotinib at the and Methods. (b) A representative autoradiogram from three molecular level needs further elucidation. To our knowledge, this independent experiments with Cys-less WT P-gps photo-crosslinked is the first report that provides an understanding of the interaction 125 with [ I]-IAAP in the absence or presence of 5 mM nilotinib or of nilotinib with human P-gp through molecular modeling, derivative 3 and 5 is shown. (c) The histogram shows accumulation mutational mapping and SAR studies. We identified residues that of rhodamine 123 in the presence and absence of 5 mM of nilotinib or are crucial for binding of nilotinib to the primary site on P-gp and derivative 3 or 5 in BacMam P-gp virus-transduced HeLa by using derivatives, we defined the molecular determinants of cells (additional details are given in the legend to Supplementary Figure S2). nilotinib for binding to P-gp. We believe these findings will help to synthesize novel TKIs that do not interact with P-gp, thus minimizing the possibility of development of resistance in cancer cells. nilotinib (Figure 2a) that lacked critical functional groups such as the pyridine ring, pyrimidine ring, CF3 group and imidazole ring were synthesized (as described in Supplementary Materials and EDITOR’S NOTE Methods). These derivatives were evaluated for their interaction This report is important to identify the primary binding site of with P-gp by testing their ability to inhibit rhodamine 123 efflux nilotinib in the drug-binding pocket of P-gp. It is promising in view from HeLa cells. Compound 1 (CF3 replaced by CH3) showed of exploiting the designing of novel tyrosine kinase inhibitors that inhibition similar to that of nilotinib (data not shown), suggesting will not be recognised by ABC drug transporters. that substitution of fluorine with hydrogen at the CF3 group is not a critical determinant of nilotinib’s binding to the P-gp. Derivative 2 (no pyrimidine–pyridine ring system) was comparable to 3 CONFLICT OF INTEREST (no pyridine ring), and 4 (imidazole replaced with a benzoic acid) The authors declare no conflict of interest. was comparable to 5 (no imidazole ring) with respect to their ability to inhibit P-gp transport activity (data not shown). Therefore, 5 and 3, derivatives lacking the key imidazole or ACKNOWLEDGEMENTS pyridine/pyrimidine rings, respectively, were further tested for We thank Mr George Leiman for editorial assistance. This research was supported by interaction with P-gp. Compared with nilotinib, 3 completely lost the Intramural Research Program of the National Institutes of Health (NIH), National 125 the ability to inhibit photolabeling of P-gp with [ I]-IAAP, but 5 Cancer Institute, Center for Cancer Research and National Center for Advancing was still able to inhibit B30–35% of [125I]-IAAP photolabeling Translational Sciences at the NIH. (Figure 2b and Supplementary Table S2). Similarly, 3 did not stimulate the ATPase activity but 5 was equally effective as S Shukla1, EE Chufan1, S Singh2, AP Skoumbourdis3, K Kapoor1, nilotinib in stimulating the ATPase activity of Cys-less WT P-gp MB Boxer3, DY Duveau3, CJ Thomas3, TT Talele2 and SV Ambudkar1 (Supplementary Table S3). In addition, 5 partially inhibited the 1Laboratory of Cell Biology, Center for Cancer Research, National rhodamine 123 efflux by P-gp, whereas 3 had no effect (Figure 2c). Cancer Institute, National Institutes of Health, Bethesda, MD, USA; These results show that the interaction of nilotinib at the 2Department of Pharmaceutical Sciences, College of Pharmacy and substrate-binding pocket of P-gp is significantly affected when Health Sciences, St John’s University, Queens, NY, USA and the pyridine and/or pyrimidine ring is absent, whereas loss of the 3NIH Chemical Genomics Center, National Center for Advancing imidazole rings only slightly perturbs this interaction. Taken Translational Sciences, Rockville, MD, USA together, the results with derivatives corroborate the docking E-mail: [email protected]

& 2014 Macmillan Publishers Limited Leukemia (2014) 935 – 979 Letters to the Editor 964 REFERENCES 8 Sauna ZE, Ambudkar SV. About a switch: how P-glycoprotein (ABCB1) harnesses 1 Shukla S, Chen ZS, Ambudkar SV. Tyrosine kinase inhibitors as modulators the energy of ATP binding and hydrolysis to do mechanical work. Mol Cancer Ther of ABC transporter-mediated drug resistance. Drug Resist Updat 2012; 15: 2007; 6: 13–23. 70–80. 9 Schindler T, Bornmann W, Pellicena P, Miller WT, Clarkson B, Kuriyan J. Structural 2Bro´zik A, Hegedu¨ s C, Erdei Z, Hegedu+sT,O¨ zvegy-Laczka C, Szaka´cs G et al. mechanism for STI-571 inhibition of Abelson tyrosine kinase. Science 2000; 289: Tyrosine kinase inhibitors as modulators of ATP binding cassette multidrug 1938–1942. transporters: substrates, chemosensitizers or inducers of acquired multidrug 10 Weisberg E, Manley P, Mestan J, Cowan-Jacob S, Ray A, Griffin JD. AMN107 (nilotinib): resistance? Expert Opin Drug Metab Toxicol 2011; 7: 623–642. a novel and selective inhibitor of BCR-ABL. Br J Cancer 2006; 94: 1765–1769. 3 Shukla S, Sauna ZE, Ambudkar SV. Evidence for the interaction of 11 Palmeira A, Sousa E, Vasconcelos MH, Pinto M, Fernandes MX. Structure imatinib at the transport-substrate site(s) of the multidrug-resistance-linked and ligand-based design of P-glycoprotein inhibitors: a historical perspective. ABC drug transporters ABCB1 (P-glycoprotein) and ABCG2. Leukemia 2008; 22: Curr Pharm Des 2012; 18: 4197–4214. 445–447. 12 Dey S, Ramachandra M, Pastan I, Gottesman MM, Ambudkar SV. Evidence for two 4 Aller SG, Yu J, Ward A, Weng Y, Chittaboina S, Zhuo R et al. Structure of nonidentical drug-interaction sites in the human P-glycoprotein. Proc Natl Acad P-glycoprotein reveals a molecular basis for poly-specific drug binding. Science Sci USA 1997; 94: 10594–10599. 2009; 323: 1718–1722. 13 Loo TW, Bartlett MC, Clarke DM. Simultaneous binding of two different drugs in 5 Shi Z, Tiwari AK, Shukla S, Robey RW, Singh S, Kim IW et al. Sildenafil reverses the binding pocket of the human multidrug resistance P-glycoprotein. J Biol Chem ABCB1- and ABCG2-mediated chemotherapeutic drug resistance. Cancer Res 2003; 278: 39706–39710. 2011; 71: 3029–3041. 14 Lugo MR, Sharom FJ. Interaction of LDS-751 and rhodamine 123 with 6 Tiwari AK, Sodani K, Dai C-l, Abuznait AH, Singh S, Xiao Z-J et al. P-glycoprotein: evidence for simultaneous binding of both drugs. Biochemistry Nilotinib potentiates anticancer drug sensitivity in murine ABCB1-, ABCG2-, 2005; 44: 14020–14029. and ABCC10-multidrug resistance xenograft models. Cancer Lett 2013; 328: 15 Ambudkar SV, Kim IW, Sauna ZE. The power of the pump: mechanisms of action 307–317. of P-glycoprotein (ABCB1). Eur J Pharm Sci 2006; 27: 392–400. 7 Ambudkar SV, Dey S, Hrycyna CA, Ramachandra M, Pastan I, Gottesman MM. 16 Marcoux J, Wang SC, Politis A, Reading E, Ma J, Biggin PC et al. Mass spectrometry Biochemical, cellular, and pharmacological aspects of the multidrug transporter. reveals synergistic effects of nucleotides, lipids, and drugs binding to a multidrug Annu Rev Pharmacol Toxicol 1999; 39: 361–398. resistance efflux pump. Proc Natl Acad Sci USA 2013; 110: 9704–9709.

Supplementary Information accompanies this paper on the Leukemia website (http://www.nature.com/leu)

Philadelphia chromosome-negative very high-risk acute lymphoblastic leukemia in children and adolescents: results from Children’s Oncology Group Study AALL0031

Leukemia (2014) 28, 964–967; doi:10.1038/leu.2014.29 that the Ph À patients received no imatinib (see Supplementary Figure 1). Prior approval was obtained from the National Cancer Institute and the Institutional Review Boards of the COG member The Children’s Oncology Group (COG) AALL0031 study included institutions. Informed consent was obtained in accordance with very high-risk (VHR) pediatric acute lymphoblastic leukemia (ALL) the Federal guidelines. Sixty-three hypodiploid (41) and IF (22) patients who had an expected 5-year event-free survival p45%. patients were enrolled in AALL0031 after 4 weeks of a three- or The chemotherapy regimen was based on previous strategies; four-drug induction regimen for National Cancer Institute eligible patients received 4 weeks of standard induction standard and high-risk ALL, respectively. Data on adverse events chemotherapy and then were enrolled on AALL0031, which and clinically significant abnormal laboratory findings were included an intensive consolidation followed by a continuation collected using National Cancer Institute Common Terminology regimen (Supplementary Figure 1).1 COG AALL0031 enrolled Criteria version 2.0. MRD was assessed by multiparameter flow patients aged 1–21 years with VHR ALL from 14 October 2002 cytometry.2 Samples were available from 46 of 63 (73%) patients to 20 October 2006. Induction therapy was limited to a at study entry. MRD high was defined as 40.01% and low as combination of vincristine, prednisone or dexamethasone, and p0.01%. asparaginase with or without daunomycin. VHR features included The primary outcome in this report is disease-free survival (DFS). the following: (a) Philadelphia chromosome [t(9;22)(q34;q11.2)]; Overall survival (OS), DFS and event-free survival were all defined (b) hypodiploidy: defined as p44 chromosomes or DNA index as the time from the end of consolidation to the first event or last o0.81; (c) any rearrangement of the MLL gene in conjunction with contact. An event was defined as relapse at any site, secondary a slow early response X5% marrow blasts at day 15 and/or malignancy or death in remission. A historical control data set of X0.1% minimal residual disease (MRD) at the end of induction as hypodiploid patients included patients enrolled on the Pediatric detected by multiparameter flow cytometry;2,3 and (d) induction Oncology Group 8602, 9005, 9006, 9201, 9405, 9406 and 9605 failure (IF) defined as either 425% blasts (M3 marrow status) by protocols for B-ALL (January 1986–November 1999).3 The percentage histology at the end of 4 weeks of induction therapy or an M2 of patients undergoing bone marrow transplant (BMT) in these marrow (5–25% blasts) or MRD X1% by flow cytometry at the end comparator studies is unknown. IF patients were excluded from of induction followed by an M2 (or M3) marrow or MRDX1% after post-induction therapy in the historical control trial. Estimates of DFS, receiving two additional weeks of induction therapy (M2/M2 IFs). event-free survival and OS were computed using the Kaplan–Meier The therapy was identical to that presented in a previous method4 and s.e. of the estimates according to Peto and Peto.5 publication on outcomes for Ph þ ALL patients,1 except The log-rank test was used for comparison of survival curves

Accepted article preview online 17 January 2014; advance online publication, 11 February 2014

Leukemia (2014) 935 – 979 & 2014 Macmillan Publishers Limited