High BCR–ABL/GUS^IS Levels at Diagnosis of Chronic Phase CML Are Associated with Unfavorable Responses to Standard-Dose Imatinib

The BCR–ABL oncoprotein is the culprit of chronic myeloid leukemia (CML) as it transforms the hematopoietic stem cell by altering its proliferation rate, survival signaling, cytoskeleton dynamics, and immunological interactions (1–5). Imatinib mesylate (IM) has dramatically improved the outcome of patients with CML in chronic phase (CP), generating unprecedented rates of complete hematologic (CHR) and cytogenetic (CCyR) responses and sustained reductions in BCR–ABL transcripts (6–9). Despite these results, approximately 50% of patients with CML fail to achieve an optimal response as defined by the current European Leukemia Net (ELN) recommendations. Intolerance, suboptimal responses, and the emergence of drug failure all contribute to identify a group of patients that do not benefit from IM (10, 11). Interestingly, assessing disease risk at diagnosis with the Sokal and Hasford scores has maintained its clinical significance as patients classified as high risk are less likely to attain the desired cytogenetic and molecular responses (6, 12, 13).

The search for accurate baseline parameters associated with unsatisfactory outcomes has led a group of CML investigators to devise the EUTOS score that assigns patients to low/high-risk categories according to spleen size and basophil count. High-risk subjects display inferior rates of CCyR, progression-free (PFS), and overall survival (OS; refs. 14, 15). However, the original EUTOS score was developed to predict the probability of obtaining a CCyR within 18 months. Hence, in a later effort, the same investigators have proposed a EUTOS Long-Term Survival (ELTS) score that employs age, spleen size, blast number, and platelet counts to subdivide patients with CML in low-, intermediate-, or high-risk groups (16). Subjects assigned to the latter two groups display a significantly higher risk of dying from CML progression (17).

However, although each of these scores has been validated in multiple patient series, none of them is specifically associated with response to IM. In this study, we searched for an easily detectable molecular parameter that would identify, at diagnosis, CML patients unlikely to benefit from IM. This issue has become of pivotal importance with the approval of second-generation (2G) tyrosine kinase inhibitors (TKI) for the first-line treatment of the disease (8, 18, 19). Indeed, considering the excellent results achieved with IM, the availability of generic forms of the drug at highly reduced costs, and the shorter follow-up of the studies using 2G TKIs in first line, a molecular indicator associated with inadequate IM responses could distinguish patients who will benefit from the drug from those who would require alternative treatments (20–26). We report that high BCR–ABL/GUS^IS transcripts measured at baseline are suggestive of inferior probabilities of achieving optimal responses to standard-dose (400 mg/daily) IM.

Between January 1, 2006, and December 31, 2015, 272 unselected adult patients with CML in CP were accrued to the observational SCREEN (Sicily and Calabria CML REgional ENterprise) multicenter study and analyzed for clinical, cytogenetic, and molecular responses. The study is ongoing but data collection was limited to the first 272 individuals that presented ≥12 months of follow-up. The research ethics committee of each recruiting institution reviewed and approved the study protocol and all patients gave written informed consent. IM 400 mg/daily was started within 12 weeks from diagnosis and discontinued in the presence of grade 3/4 toxicities. Treatment was resumed after toxicity reduction to grade 1 or after complete resolution. IM responses were evaluated according to the 2013 ELN criteria (27).

CP-CML and CHR were defined by conventional criteria (28). Bone marrow (BM) cytogenetics were assessed at diagnosis and then every 6 months until achievement of a CCyR. Subsequently, annual BM examinations were performed although, after 2010, subjects in CCyR were mostly monitored by real-time quantitative polymerase chain reaction (RQ-PCR). Cytogenetic responses were evaluated on ≥20 marrow cell metaphases. CCyR was defined as failure to detect Philadelphia chromosome (Ph)-positive metaphases in two consecutive examinations. Confirmed detection of Ph-positive metaphases after acquiring CCyR was considered cytogenetic relapse. Chromosomal abnormalities were scored as previously described (29). When cytogenetic analyses were unavailable due to insufficient material, technical failure or patient's refusal to undergo a BM aspirate, BCR–ABL/ABL^IS ratios ≤1% were considered equivalent to a CCyR as previously reported (30).

BCR–ABL transcripts were measured from peripheral blood (PB) samples drawn at diagnosis and every 3 months thereafter using RQ-PCR (31). All RQ-PCR determinations were centralized in the Center of Experimental Oncology and Hematology. Samples collected at diagnosis were subjected to RQ-PCR using the TaqMan platform and both beta-glucuronidase (24) and ABL as reference genes as previously reported (32). For samples collected at diagnosis, we used GUS, as it is the more appropriate reference gene for specimens expressing high levels of BCR–ABL (32). For BCR–ABL transcripts measured at time points other than diagnosis, ABL was the only reference gene used (31, 33). BCR–ABL/GUS and BCR–ABL/ABL ratios were reported on the international scale (IS) using a conversion factor (CF) calculated—on a yearly basis—from primary CML samples shared with the laboratory at the University Hospital Mannheim. When BCR–ABL/ABL was >10% and a specific value was reported, the suffix IS was removed as according to previously published data (34) the IS should not be calculated for BCR–ABL/ABL ratios >10% based on nonlinearity of the ratio. RQ-PCR determinations were considered of appropriate quality only in the presence of no less than 24,000 GUS copies or 10,000 ABL copies as previously indicated (35). Of the 272 PB specimens collected, 32 had to be discarded because of prior HU exposure (n = 27) or poor nucleic acid quality (n = 5). Of the 240 remaining specimens, only two had BCR–ABL/ABL^IS ratios <10%. Unavailable baseline samples were distributed as follows: 17 of 178 specimens from patients achieving an OR, seven of 61 samples from patients failing IM, three of 15 specimens from individuals classified as “warning,” and five of 18 samples from intolerant patients. Two hundred five and 214 blood samples, respectively, were available for molecular analyses performed at the 3- and 6-month time points (Table 3). For analyses requiring both baseline and 3-month samples, blood specimens were available from 200 patients. Similarly, blood samples from 210 patients were assessable at diagnosis and after 6 months of IM (Table 4). Major molecular response and MR⁴ have been defined as previously reported (36).

Univariate probabilities of overall survival (OS), transformation-free (TFS), failure-free (FFS), and event-free (EFS) survival were calculated using the Kaplan–Meier method. Statistical significance of Kaplan–Meier curve differences was evaluated using the Mantel–Haenszel test as previously described (37). TFS was defined as survival without disease progression to accelerated or blast phases. FFS was defined as survival without evidence of drug failure according to the current ELN criteria. Events included in our definition of EFS were death from any cause, progression from chronic phase, IM failure according to the 2013 ELN recommendations, and development of intolerance. Probabilities of cytogenetic and molecular responses were calculated using cumulative incidence function within the Kaplan–Meier method, in which maintained responses were the events of interest. Significance of BCR–ABL transcript levels at diagnosis for the achievement of OS, TFS, FFS, EFS, and OR was evaluated using the Wilcoxon rank-sum test. Optimal threshold in BCR–ABL/GUS^IS levels associated with specific outcomes after IM were calculated using receiver-operating characteristic (ROC). These thresholds maximize true positive and false positive rates in a binary classification according to the clinical phenotype of interest. P values were two-sided with 95% confidence intervals (CIs). Analyses were performed using the R software (38). To assess the robustness of the association between BCR–ABL^IS levels at diagnosis and FFS, the database was randomly shuffled and divided into a training set (with 80% of patients) and a test set (with 20% of patients) using a simple perl script. The sampling procedure was iterated in order to select more than 50 combinations of randomly selected pairs of training and test subsets with a comparable distribution of patients with BCR–ABL levels above and below the specified threshold.

Patient characteristics are summarized in Supplementary Table S1. All patients achieved a CHR. Two hundred twenty-nine patients (84.2%) attained a CCyR (median time, 6 months; range, 6–24 months), 206 (75.7%) obtained an MMR (median time 12 months; range, 3–57 months), and 68 (25%) displayed an MR⁴ (median time 34.5 months; range, 6–57 months). Median follow-up of the accrued population was 60 months (range, 12–108 months). No patients were lost to follow-up. Eight-year estimates of OS (89%) and TFS (95%) on an intention-to-treat basis are shown in Supplementary Fig. S1. Comparison with 5- and 8-year results reported for the IRIS (39) and German CML IV studies (40, 41) are summarized in Supplementary Table S2. Using the 2013 ELN recommendations, at the time of data lock 178 patients (65.4%) displayed an OR, 61 (22.4%) had failed IM, 15 (5.6%) were classified as “warning,” and 18 (6.6%) discontinued IM due to intolerance. Of the 61 patients failing treatment, one underwent allogeneic stem-cell transplantation, while the remaining 60 received 2G TKIs. Twenty-three of the 60 patients that eventually switched to dasatinib or nilotinib transiently received increased (600/800 mg/daily) IM doses for a median of 13 months (range, 2–39). All 15 intolerant patients received 2G TKIs.

Median BCR–ABL/GUS^IS levels at diagnosis in the entire population were 13.6% (range, 2.1%–66.7%), while median BCR–ABL/ABL expression was 67.2% (range, 5.6%–349.5%). To establish if quantification of BCR–ABL transcripts at diagnosis would correlate with IM response, we stratified our evaluable patient cohort in optimal responders (n = 161, 178 minus 17 unavailable samples as described in Materials and Methods) and individuals failing IM (n = 54, 61 minus 7 unavailable samples), and we then analyzed the amount of BCR–ABL/GUS^IS expression in each group. Subjects failing IM presented significantly higher amounts of BCR–ABL transcripts when compared with optimal responders (20.39% vs. 11.97%; P < 0.001; Fig. 1). We then repeated the same analysis using ABL as a control gene and again found that patients failing IM displayed higher BCR–ABL transcripts when compared with optimal responders (101.69% vs. 61.35%; P < 0.005; Supplementary Fig. S2). While the low number of evaluable “warning” (n = 12, 15 minus 3 unavailable samples) or intolerant (n = 13, 18 minus 5 unavailable samples) patients did not allow any meaningful statistical analyses, median baseline BCR–ABL/GUS and BCR–ABL/ABL levels observed in these groups were similar to those detected in optimal responders.

We next wanted to determine individual BCR–ABL/GUS^IS and BCR–ABL/ABL thresholds associated with specific outcomes after IM treatment. We used ROC curves to define baseline BCR–ABL values that would correlate with lower rates of OS, TFS, FFS, and EFS and with lower probabilities of OR (Table 1 and Supplementary Table S3). While we identified a BCR–ABL/GUS^IS threshold for OS (18.55%), this value did not achieve statistical significance (P = 0.132; Fig. 2A). However, we found that BCR–ABL/GUS^IS levels >18.79% and >14.89% were associated with a lower likelihood of achieving a TFS (P = 0.029) and FFS (P < 0.001), respectively (Fig. 2B and C). Likewise, we observed that a BCR–ABL/GUS^IS value >15.94% discriminated patients with significantly lower probabilities (P < 0.001) of obtaining both an EFS and an OR (Fig. 2D and Table 1). When we used BCR–ABL/ABL transcripts to define outcome-related thresholds, the values calculated for OS (45.07%) and for TFS (44.50%) failed to achieve statistical significance (P = 0.052 and P = 0.167, respectively; Supplementary Table S3 and Supplementary Fig. S3A and S3B). On the contrary, BCR–ABL/ABL levels >92.07%, >102.41%, and >97.36% were significantly associated with lower probabilities of FFS (P = 0.001), EFS (P = 0.005), and OR (P < 0.03; Supplementary Table S3 and Supplementary Fig. S3C and S3D).

We next wanted to validate the performance of the GUS-based and the ABL-based models in accurately discriminating patients unlikely to achieve specified treatment outcomes with standard dose IM. We chose an internal validation method that uses repeated unbalanced splits to subdivide the overall population in training and validation sets (42, 43). The principle behind this approach is that by randomizing patients according to the same normalization parameter using numerically unbalanced splits (80/20), the smaller subgroup will highlight possible inconsistencies, which may be masked in the general population of the study. We therefore randomized our 272 patients for 50 times in different 80/20 groups using the FFS ROC values calculated using either GUS (Table 1) or ABL (Supplementary Table S3) as reference genes. We found that GUS clearly outperformed ABL as a control gene for samples measured at diagnosis (Table 2). Indeed, GUS appropriately classified patients at risk of failing IM in 100% of the 80% splits (training set) and in 96% of the 20% splits (validation set). On the contrary, ABL was reliable in the 80% splits (training set) appropriately classifying 94% of the patients, but generated an unacceptable 80% misclassification rate (20% correct classification) in the smaller 20% splits (validation set). These findings confirm that, unlike GUS, ABL cannot be used as a reference gene for samples collected at diagnosis.

Consolidated evidence has established that BCR–ABL/ABL^IS values >10% after 3 months and >1% after 6 months of TKI therapy are associated with inferior rates of OS, PFS, EFS, and CCyR (44, 45). We wished to compare these two molecular parameters with the different BCR–ABL/GUS^IS thresholds that we had previously identified as, unlike BCR–ABL/ABL transcripts, BCR–ABL/GUS^IS levels can be reliably measured before commencing TKI treatment and may therefore guide the choice of the most appropriate first-line drug (32). We found that baseline BCR–ABL/GUS^IS thresholds for OR, EFS, and FFS were equally effective in predicting the probability of achieving these outcomes as compared with the 10% or 1% BCR–ABL/ABL^IS cutoffs (P < 0.001; Table 3). In our cohort, only the 18.79% BCR–ABL/GUS^IS threshold at diagnosis (P = 0.029) and the 1% BCR–ABL/ABL^IS value after 6 months of treatment (P = 0.021) could significantly discriminate the likelihood of obtaining a TFS (Table 3). This was not the case for the 10% BCR–ABL/ABL^IS level after 3 months of therapy (P = 0.134). As for OS, only the 1% BCR–ABL/ABL^IS threshold after 6 months of IM correlated with higher survival rates (P = 0.025). Neither the baseline 18.55% BCR–ABL/GUS^IS value (P = 0.132) nor the 10% BCR–ABL/ABL^IS cutoff after 3 months of therapy (P = 0.132) were associated with higher survival probabilities (Table 3).

We next wanted to establish whether patients with high BCR–ABL/GUS^IS transcripts at diagnosis would also display higher BCR–ABL/ABL^IS values after 3 or 6 months of IM. We compared the populations defined by BCR–ABL/GUS^IS thresholds calculated for each outcome with those identified by the 10% (at 3 months) and 1% (at 6 months) BCR–ABL/ABL^IS levels. Our analysis showed high concordance between the three molecular parameters (Table 4). Specifically, patients failing to achieve an OR (>15.94% BCR–ABL/GUS^IS) or exhibiting lower rates of EFS (>15.94% BCR–ABL/GUS^IS), FFS (>14.89% BCR–ABL/GUS^IS), TFS (>18.79% BCR–ABL/GUS^IS), and OS (>18.55% BCR–ABL/GUS^IS) mostly displayed BCR–ABL/ABL^IS levels >10% at 3 months and >1% at 6 months, with concordance ranging from 61.5% to 72.3% (Table 4). All these concordance rates were statistically significant.

In this first account of the SCREEN study, we describe the correlation between BCR–ABL/GUS^IS levels measured at diagnosis and the response to IM of 272 patients newly diagnosed with CP-CML. The main aim of the study was to determine if the quantitative evaluation of BCR–ABL transcripts before exposure to any therapy could identify patients unlikely to benefit from standard-dose IM. In our analysis, higher levels of BCR–ABL expression were associated with unsatisfactory responses to the drug.

These findings pose several interesting questions. First of all, our data are apparently in contrast with what has been previously published by Hanfstein and colleagues on behalf of the German CML Study Group (46). In their report, the authors investigated the possible predictive value of BCR–ABL transcripts measured—at diagnosis—using GUS as a reference gene in a subgroup of 301 patients from the IM-treated CML cohort recruited to the German CML Study IV. The authors correctly excluded from their analysis samples collected after exposure to HU or IM. They found no correlation between high BCR–ABL expression and OS and PFS and therefore argued for the limited value of quantifying BCR–ABL per se at diagnosis. Just like Hanfstein and colleagues, we failed to detect any significant correlation between baseline BCR–ABL/GUS^IS levels and OS or PFS, respectively (Table 1 and data not shown). However, a careful analysis of our results suggests that this is due to the limited number of events detected in our patient cohort. Specifically, of the 19 deaths observed—to date—in our patient population, only seven were due to disease progression. Hence, most of these events were probably unrelated to CML. This explains why high BCR–ABL/GUS^IS expression was associated with TFS but not with PFS (that includes deaths from any cause in addition to those due to disease transformation) and also explains why the BCR–ABL/GUS^IS threshold calculated for OS (18.55%) was inferior to that found for TFS (18.79%). It should also be noted that, while overall efficacy of IM-based treatments in the German CML Study IV were superimposable to those observed in our series with standard dose IM (Supplementary Table S2), there were significant differences in the baseline BCR–ABL/GUS^IS quantification of the two patient cohorts. Indeed, both median BCR–ABL/GUS^IS expression (13.6% in the SCREEN study vs. 33% in the German CML Study IV) and the observed BCR–ABL/GUS^IS ranges (2.1%–66.7% in our series vs. 0.1%–230% in the German cohort) argue for a higher degree of heterogeneity and, possibly, for more patients with unfavorable disease characteristics in the German population. It should also be noted that, unlike our cohort, only 75 (25%) of the 301 patients analyzed by the German group received IM 400 mg daily. This observation is of critical importance as our preliminary data suggest that using higher IM doses or more potent 2G TKIs raises the thresholds identified in this study for standard-dose IM. Finally, the lack of correlation that we report between the 10% BCR–ABL/ABL^IS threshold and TFS and OS rates (Table 2) should not be surprising because the German group observed the same result when they limited their analysis to a relatively small (i.e., 301 subjects) patient cohort (46).

Another critical issue concerns possible correlations between baseline BCR–ABL/GUS^IS levels and different clinical and molecular parameters. In a univariate analysis, we found no associations between BCR–ABL/GUS^IS transcripts at diagnosis and patient age, sex, hemoglobin levels, Sokal risk scores, number of Ph-positive metaphases and white blood cell (WBC) counts. Interestingly, the lack of correlation between WBC counts at diagnosis and BCR–ABL/GUS^IS levels suggests that high BCR–ABL values are independent from the leukemic burden (i.e., from the number of Ph-positive cells at diagnosis). For example, in our series, the patient with the highest WBC count at diagnosis (patient #25: WBCs 758.000/μL) expressed low (5.51%) BCR–ABL/GUS^IS, while the individual (#157) with the highest BCR–ABL/GUS^IS value at diagnosis (66.71%) only had 50.000/μL WBCs. Thus, high BCR–ABL/GUS^IS expression is probably indicative of higher amounts of BCR–ABL transcripts within each leukemic cell, a well-established sign of CML disease progression (47). An additional observation strengthens this hypothesis: the median baseline BCR–ABL/GUS^IS levels of the seven patients deceased after CML progression was 33.85% (range, 23.09%–41.23%), a much higher value than the 20.39% displayed by the 61 patients failing IM.

From a technical standpoint, our results provide further evidence supporting the concept that ABL is not an appropriate reference gene for samples presenting high BCR–ABL transcripts (i.e., >10% BCR–ABL/ABL^IS) as these specimens require a BCR–ABL-indipendent housekeeping gene. Indeed, internal validation experiments in our patient cohort clearly showed that ABL may missclassify up to 80% of patients at risk of failing IM (Table 2). The importance of using the correct reference gene when measuring BCR–ABL expression at diagnosis is highlighted by the data summarized in Supplementary Fig. S4. Both patients depicted in the top panel presented high BCR–ABL/GUS^IS values at baseline (54.9% and 40.2%) but low BCR–ABL/ABL transcripts (33.8% and 15.8%), yet both subjects failed IM treatment after 6 and 12 months, respectively. It will ultimately be up to each individual laboratory to choose between using different reference genes for samples collected at diagnosis (24) or at other time points (ABL) or alternatively switching to GUS for all their molecular determinations.

From a biological standpoint, our findings raise several, yet unanswered, issues. Our data suggest that the fate of the leukemic clone may be already determined at the time of diagnosis. However, it is still unclear if quantitative or qualitative mechanisms underlie the inferior IM sensitivity of CML clones with high BCR–ABL. It is likely, although untested, that higher BCR–ABL transcripts translate to higher expression of the BCR–ABL oncoprotein and increased tyrosine kinase activity. According to the quantitative hypothesis, this would ultimately strengthen canonical BCR–ABL-dependent signaling resulting in a less responsive leukemic population. On the other hand, the qualitative hypothesis assumes that higher BCR–ABL activity leads to “leakage” of BCR–ABL signaling to downstream targets that are usually not involved in BCR–ABL-dependent transformation. It is also possible that both quantitative and qualitative mechanisms contribute to the lower IM responsiveness of patients displaying high levels of BCR–ABL. It also remains to be seen if CD34⁺ progenitors derived from patients presenting high BCR–ABL/GUS^IS levels at diagnosis display higher BCR–ABL transcripts as compared with CD34⁺ cells isolated from patients with low BCR–ABL/GUS^IS.

From a clinical standpoint, additional studies will be needed to compare the predictive potential of the ELTS with that of BCR–ABL/GUS^IS quantification at diagnosis. Similarly, the German and the Australian groups have suggested that dynamic measurements of BCR–ABL transcripts at baseline and at the 3-month time-point allow calculation of either a BCR–ABL reduction rate (46) or a BCR–ABL halving time (48). In their patient series, both indicators were reliable predictors of the benefit from various IM-based therapies. At the present time, we have not repeated BCR–ABL/GUS^IS measurements in our samples collected after 3 months of IM. Hence, we cannot establish to what extent patients expressing high BCR–ABL/GUS^IS at diagnosis would also display modest reductions in their BCR–ABL ratio or in their BCR–ABL halving times. Furthermore, while our results strongly suggest that patients expressing high levels of BCR–ABL/GUS^IS at diagnosis are unlikely to achieve optimal outcomes with standard-dose IM, they have not established if these patients are generally unresponsive to TKIs or would benefit from higher (tolerability-adapted) doses of IM (12) or from 2G TKIs. Identification of baseline BCR–ABL/GUS^IS cutoffs for specific endpoints in wide and uniformly treated patient cohorts will be necessary to address this issue that will be tackled by a prospective study beginning enrolment next year. Lastly, we used a previously reported statistical approach (49) to generate nomograms predicting the 8-year likelihood of achieving FFS when receiving standard-dose IM according to BCR–ABL/GUS^IS expression at diagnosis and the Sokal or the Hasford scores (Supplementary Fig. S5). As expected, these nomograms confirm that patients exhibiting elevated BCR–ABL/GUS^IS levels and included in the high Sokal or high Hasford categories display the highest risk of failing IM. These nomograms may be of clinical value for physicians in need of estimating the FFS probability of their patients.

In summary, our findings suggest that patients displaying higher levels of BCR–ABL/GUS^IS transcripts at diagnosis are less likely to benefit from standard-dose IM. As achieving an OR according to the latest ELN recommendations is a mandatory goal for most patients with newly diagnosed CML, baseline quantification of BCR–ABL expression may represent a useful parameter for physicians in need of discriminating patients likely to respond to IM from those that should receive alternative treatments. In the era of “personalized medicine,” a molecular index reliably measurable at diagnosis might therefore prove useful in guiding the choice for the most appropriate first line treatment of CML (50).

P. Vigneri reports receiving commercial research grants from Novartis, and is a consultant/advisory board member for Astra Zeneca, Bristol-Myers Squibb, Incyte, and Pfizer. F. Stagno is a consultant/advisory board member for Bristol-Myers Squibb and Novartis. S. Molica is a consultant/advisory board member for AbbVie, Gilead, Jansen, and Roche. M.C. Müller reports receiving commercial research grants from and is a consultant/advisory board member for Ariad, Bristol-Myers Squibb, and Novartis. A. Hochhaus reports receiving commercial research grants from Ariad, Bristol-Myers Squibb, Pfizer, and Novartis. F. Di Raimondo reports receiving speakers bureau honoraria from Bristol-Myers Squibb and Novartis. No potential conflicts of interest were disclosed by the other authors.

The funders had no role in the design or conduct of the study and the collection, management, analysis, or interpretation of the data.

Conception and design: P. Vigneri, F. Stagno, S. Siragusa, M. Musso, B. Martino, A. Hochhaus, F. Di Raimondo

Development of methodology: S. Stella, M. Massimino, M.C. Müller, A. Hochhaus

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Vigneri, F. Stagno, A. Cupri, A. Antolino, S. Siragusa, D. Mannina, C. Musolino, G. Mineo, C. Tomaselli, P. Murgano, M. Musso, F. Morabito, S. Molica, B. Martino, M.C. Müller, F. Di Raimondo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Vigneri, S. Forte, C. Musolino, F. Morabito, A. Hochhaus, F. Di Raimondo

Writing, review, and/or revision of the manuscript: P. Vigneri, F. Stagno, S. Stella, M. Massimino, S. Siragusa, S.S. Impera, C. Musolino, L. Manzella, M.C. Müller, A. Hochhaus, F. Di Raimondo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Vigneri, F. Stagno, S. Stella, M. Massimino, A. Hochhaus

Study supervision: P. Vigneri, S. Stella, M. Massimino, S. Siragusa, A. Malato, F. Morabito, F. Di Raimondo

This article is dedicated to Alessandra Aloisi and Salvatore Berretta.

This work was supported by AIRC grants (RG1062) to F. Di Raimondo and P. Vigneri (IG12958). Since 2008, the SCREEN network has received support from Bristol-Myers Squibb and PSN, Assessorato alla Salute, Regione Sicilia.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.