Thiotepa-busulfan-fludarabine compared to busulfan-fludarabine for sibling and unrelated donor transplant in acute myeloid leukemia in first remission

Background A preparatory regimen consisting of thiotepa-busulfan-fludarabine (TBF) has been associated with reduced relapse in patients with haematological malignancies after haploidentical and cord blood transplants; however, few data exist regarding TBF conditioning in sibling (MSD) and unrelated donor (URD) transplants for AML. Results Among patients receiving a myeloablative (MAC) regimen, TBF-MAC was associated with significantly lower relapse (HR 0.47, p = 0.005) however higher non-relapse mortality (NRM, HR 2.69, p < 10–4) as compared to BF. This led to similar leukemia-free (LFS) and overall survival (OS) between the two regimens (LFS: p = 0.6; OS: p = 0.27). When we selected TBF-MAC patients receiving busulfan 9.6 mg/kg, NRM resulted still higher but no more significantly different as compared to BF-MAC with busulfan 12.8 mg/kg (HR 1.53, p = 0.12); despite the lower busulfan dose, relapse remained inferior with TBF-MAC (HR 0.45, p = 0.01), however no difference in survival could be demonstrated (LFS: p = 0.31; OS: 0.82). Among patients receiving a reduced-intensity (RIC) regimen, similar outcome was observed with TBF-RIC and BF-RIC (LFS: p = 0.77; OS: p = 0.88). Conclusions TBF-MAC as conditioning regimen for transplant from MSD and URD in AML patients in first remission provided stronger anti-leukemic activity but higher NRM as compared to BF-MAC, thus leading to similar survival. TBF-MAC with busulfan 9.6 mg/kg was associated with low relapse and acceptable NRM, however again with no survival benefit. TBF-RIC and BF-RIC resulted in comparable outcome. Methods We conducted a registry-based study comparing outcomes of patients with AML in first remission undergoing transplant from MSD or URD prepared with either TBF (n = 212) or BF (n = 2698) conditioning.


INTRODUCTION
In recent years, the paradigm of conditioning regimens for allogeneic stem cell transplant (allo-SCT) in AML has shifted from heavy myeloablation to reduced intensity protocols, with the aim of minimizing regimen related toxicity while exploiting the graft-versus-leukemia effect [1]. An important step towards this goal has been made with the replacement of cyclophosphamide with fludarabine [2]. A preparatory regimen consisting of busulfan and fludarabine (BF) was initially described in the early 2000s; [3] subsequently different studies were carried out to challenge BF with standard BuCy, reporting conflicting results on the relative merits of the two regimens in terms of non-relapse mortality (NRM), relapse and survival [4][5][6][7]. More recently, different prospective randomized trials have been conducted in an attempt to answer the question whether BF could be a valid substitution for standard BuCy, reporting inconsistent results [8][9][10][11]. A recent meta-analysis of 15 studies [12] concluded that the BF conditioning regimen is associated with inferior toxicity and lower 100-days NRM with no increase in relapse risk as compared to BuCy; however, no survival benefit could be demonstrated. In fact, although the substitution of cyclophosphamide with fludarabine results in decreased toxicity, leukemia relapse still represents the main cause of transplant failure, accounting for about 30% at 2 years when transplant is performed in remission.
Thiotepa is an alkylating compound with immunosuppressive properties which holds favourable characteristics as the capability of penetrating the bloodbrain barrier combined with a reduced non-haematologic toxicity, which led to a widespread use of this compound within transplant preparative regimens [13][14][15]. A conditioning protocol consisting of thiotepa, busulfan and fludarabine (TBF), initially designed for cord blood transplant, [16,17] has recently been associated with low relapse rates and satisfactory outcome in haploidentical transplantation (haplo-SCT) [18][19][20][21][22] and in single center reports of allo-SCT from MSD and URD-SCT for different haematological malignancies [23,24]. However, no data are available about outcome of AML patients undergoing MSD or URD-SCT prepared with this protocol, and no comparison of this novel regimen with standard regimens has been published yet. We therefore took advantage of the EBMT dataset, and retrospectively compared outcome following TBF versus BF as conditioning regimens before allo-SCT from MSD or URD-SCT in a large homogeneous population of AML patients undergoing transplant in first complete remission (CR1).

Patient characteristics
Patients from 204 transplant centers were included. The total number of patients who received either TBF or BF in CR1 between January 2007 and June 2015 and reported to the EBMT was 4995 (265 TBF and 4730 BF); two thousand and eighty-five patients were excluded from the analysis as they have received a busulfan dose <6.4 mg/kg, thus leading to a final number of 2910 patients; among them, 212 received TBF and 2698 BF, respectively. One thousand six hundred and six patients received a MAC and 1304 a RIC regimen. One thousand six hundred and fifty-three (57%) patients were transplanted from MSD, 987 (34%) from 10/10 URD and 9% (270) from 9/10 URD, respectively. Data about cytogenetic risk were available for 55% of patients, while information on molecular genetics and minimal residual disease (MRD) status was missing for most patients and was therefore not included in the analysis. The characteristics of patients are summarized in Table 1

Engraftment, NRM and GVHD
Engraftment rate was 98% following both regimens (p = 0.88). The median time to neutrophil engraftment was 18 (10-47) days and 15 (5-45) days for TBF-MAC and BF-MAC, respectively (p < 10 −5 ). The 2-year non-relapse mortality rate in the overall population was significantly higher after TBF-MAC compared to BF-MAC in both univariate (27 ± 8% vs 16 ± 2%, p = 0.006) and multivariate analysis (HR 2.7, p < 10 −4 ). When analyzing separately NRM within and after day 100, NRM rate following TBF-MAC was significantly higher as compared to BF-MAC within 100 days after transplant (8 ± 4% vs 5 ± 1%, p = 0.008), while a trend for higher NRM was observed for the time period between 100 days and 2 years after transplant (19 ± 8% vs 12 ± 2% for TBF-MAC and BF-MAC respectively, p = 0.08). Karnofsky performance score at transplant (≥90% vs <90%) did not affect risk of NRM. Independent predictive factors for higher NRM risk were older age, higher busulfan dose, URD transplant compared to MSD and PBSC compared to BM graft.
Main causes of NRM were GVHD and infectious complications ( Table 2). When analyzing separately the incidence of infectious-related and GVHD-related deaths no difference was observed between the two study cohorts.

Transplant outcome
The 2-year relapse incidence was significantly lower in TBF-MAC group (14 ± 6%) compared to BF-MAC group (27 ± 2%, p = 0.002) ( Figure 1). This result was strongly confirmed in multivariate analysis (HR 0.5, p = 0.005). Secondary AML was the only additional factor associated with higher relapse risk in multivariate analysis ( Table 3).
Leukemia-free survival (LFS) at 2 years was not statistically different between TBF-MAC and BF-MAC groups, being 59 ± 10% in TBF-MAC and 57 ± 3% in BF-MAC, respectively (p = 0.5) ( Figure 2). Multivariate     severe cGVHD and with a trend for better GRFS. In the population of patients who relapsed after transplant, there was no difference in overall survival after relapse between the two study cohorts. In MAC group 2-year after relapse OS was 14% vs 10% in the TBF and BF groups, respectively (p = 0.74). In RIC cohort OS was 17% vs 13% one year after relapse (p = 0.86).

Subgroup analysis
In multivariate analysis conducted in the global population, busulfan dose (12.8 vs 9.6 mg/kg) emerged as independently associated with NRM, while had no impact on relapse risk. Given this findings, and according to the regimens schedules as previously published, [10,  17] we conducted a subgroup analysis of TBF-MAC patients who received busulfan 9.6 mg/kg, which were compared to BF-MAC patients receiving busulfan 12.8 mg/kg (Table 4). With the reduced dose of busulfan (9.6 mg/kg) the NRM risk following TBF-MAC resulted still higher however not significantly different as compared to BF-MAC (HR 1.53, p = 0.12). Notably, despite the reduced busulfan dosage, relapse risk following TBF-MAC remained significantly lower as compared to BF-MAC (HR 0.45, p = 0.01) (Figure 3). Finally, no impact  Table 4.

Propensity score matched-pairs analysis
Primary and secondary endpoints were further challenged in a propensity score matched-pairs analysis, conducted on 138 TBF-MAC matched with 262 BF-MAC patients, which confirmed the results obtained in the global population ( Supplementary Tables 1 and 2, appendix).

Reduced-intensity conditioning: TBF-RIC versus BF-RIC
One thousand three hundred and four patients received a RIC regimen; among them, 65 patients conditioned with TBF-RIC were compared to 1239 patients who received BF-RIC. Characteristics of patients are summarized in Table 1.
The use of ATG was independently associated with significantly lower probability of severe cGVHD and better GRFS.
Similarly to the MAC cohort, endpoints were further tested by a propensity score matched-pairs analysis, conducted on 61 TBF-RIC vs 118 BF-RIC patients, which confirmed these results (Supplementary Tables 1 and 2, appendix).

DISCUSSION
Despite great efforts in optimizing preparatory regimens and transplant procedures, leukemia relapse remains today the major cause of transplant failure. We hypothesized that the thiotepa-busulfan-fludarabine protocol could ensure a strong leukemia control, reducing relapse and thus leading to improved outcome as compared to busulfan-fludarabine.
In fact, the TBF-MAC conditioning was associated with inferior relapse as compared to BF-MAC; however, the increased non-relapse mortality offset this advantage and no difference could be observed in survival between the two regimens. Notably, the 2-year relapse incidence was as low as 14% following the TBF-MAC; in multivariate analysis, relapse risk was cut by half using the TBF-MAC as compared to the BF-MAC protocol. These results compare favourably with myeloablative regimens currently employed; for instance, 2-year relapse rate following the classical CY/TBI regimen stands at about 20% when transplant is given in remission [25]. In fact, leukemia recurrence following transplant remains an unresolved issue, and the recent tendency to mitigate conditioning regimen intensity raises significant concerns about further increasing relapse risk, as confirmed by recent evidence [26]. In their randomized BF vs BuCy study, Lee and colleagues [8] reported a 2-year relapsefree survival of 55% following BF, which was significantly inferior as compared to BuCy; Rambaldi et al. [10] observed a cumulative incidence of relapse following BF of about 30% at 2 years which however, in contrast to the previously cited study, was not statistically different as compared to BuCy.
The addition of thiotepa to busulfan and fludarabine appears to provide a significantly stronger anti leukemic effect; in fact, our results concur with recent observations reporting encouraging low relapse rates when TBF was employed in cord blood (Sanz et al., [17] 5-year RI: 18%) and haplo-SCT (Bacigalupo et al., [19] cumulative incidence of relapse related death: 11% for patients in CR1). In the study by Di Bartolomeo et al., [21] the use of TBF-MAC was the only factor predicting lower relapse risk in multivariate analysis. It might be speculated that the combined efficacy of two alkylating agents mediating powerful anti-leukemic activity could be responsible of such effective disease control. In addition, thiotepa is well known to have a radiomimetic action and to penetrate sanctuary sites [27].
Nevertheless, the low relapse rate following TBF observed in our study should be interpreted with caution as it could, at least in part, be related to the high NRM observed in the same cohort of patients. Indeed, 2-year NRM following TBF-MAC was 27% in the overall population, which resulted significantly higher as compared to the BF conditioning. Higher NRM following a strong myeloablative protocol (TBF) as compared to a reduced-toxicity regimen (BF) does not represent a surprising finding per se; further, this result is in accordance with the previously cited studies in which TBF-MAC was employed in haploidentical (Arcese et al. [22] cumulative incidence of NRM: 32%) and cord blood transplant (Sanz et al., [17] 5-year NRM: 44%). However, the significant NRM rate reported in our study warrants a deeper reflection. When examining factors associated with NRM, busulfan dose (12.8 vs 9.6 mg/kg) stood out as strongly predicting high NRM risk, while it did not demonstrate any impact on relapse incidence. In view of this observation, and considering the published standard regimen schedules, [10,17] we performed a subgroup analysis comparing TBF-MAC with busulfan 9.6 mg/kg and BF-MAC with busulfan 12.8 mg/kg. With the reduced dose of busulfan, NRM risk after TBF substantially decreased to a level not statistically different as compared to BF; however a trend towards higher NRM for TBF was retained. Importantly, the lower dose of busulfan did not impair the anti-leukemic activity of TBF, as relapse remained significantly inferior as compared to BF. Based on these data, the combination of two alkylators at full myeloablative doses appears excessively toxic, with no apparent benefit in terms of disease control. Nevertheless in this subgroup analysis, similarly to what was found in the global population, the strong advantage of TBF-MAC in terms of reduced relapse was negated by the high (although slightly reduced) NRM. Indeed, despite TBF hazard ratios for LFS, OS and GRFS resulted constantly below 1, no statistical difference in survival could be observed between the two regimens.
When conducting exploratory analyses according to age, donor type and leukemia risk, we could not find any significant difference between TBF-MAC and BF-MAC in terms of survival in any of the subgroups examined.
The main causes of death were infection and GVHD. Lethal infectious complications accounted for one out of four deaths in the TBF group. Our results are in accordance with previous publications reporting high rate of infection following TBF regimen [22]. Thiotepa holds a heavy myeloablative effect combined with a powerful immunosuppressive potential which may have contributed to this. The acknowledgment of such risks following TBF-MAC could help to better select patients suitable for this regimen, and to implement an optimal anti-infectious monitoring, prophylaxis and treatment which may improve outcome.
Notably, despite the higher intensity, TBF-MAC resulted in similar incidence of grade II-IV aGVHD as compared to BF-MAC. The strong immunosuppressive activity of thiotepa in association with fludarabine may have played a role in this. Finally, incidence of severe cGVHD was similar between the two conditioning regimens. Importantly, the inclusion of ATG in both protocols was independently associated with significantly lower probability of severe cGVHD both in patients receiving MAC and RIC, and with better GRFS.
In a separate analysis, we compared reduced-intensity versions of TBF and BF, reporting similar outcome. Interestingly, NRM rate of TBF-RIC was similar to BF-RIC, despite the combination of two alkylators in a cohort of patients with a median age of 60 years. We did not observe a difference in relapse incidence between the two study cohorts; however, due to the limited number of patients receiving TBF-RIC no definite conclusion can be made.
It is important to recognize the limitations of the present study. First, the retrospective design did not allow to study the reason for patient allocation to a specific regimen. Secondly, the two study arms were unbalanced as to the number of patients included. Finally, some of the patient's characteristics varied among the two groups. Nonetheless, the present analysis represents the largest study reporting outcome of TBF conditioning in MSD or URD-SCT for AML; further, we addressed the inherent limitations of a registry-based study performing a confirmatory PS matched-pair analysis, which thoroughly upheld all the results we observed in the overall population.

CONCLUSIONS
In conclusion, our results suggest that TBF-MAC provides significantly lower relapse, which is counterbalanced by increased NRM as compared to BF-MAC, thus resulting in similar survival. The combination of thiotepa with high dose busulfan (12.8 mg/kg) appears excessively toxic, while TBF with lower dose of busulfan (9.6 mg/kg) seems to retain strong anti-leukemic effect in combination with acceptable NRM, however with no statistically significant survival advantage over BF. In elderly patients, TBF-RIC appears a promising reduced-intensity regimen, as similar outcome was reported as compared to BF-RIC. This registry data should be validated by a well-designed randomized trial comparing outcome of TBF with other busulfan-based regimens, both in the MAC and RIC setting. Specifically, designed studies are necessary to identify which patient category could benefit the most from the strong anti-leukemic potential of TBF protocol, without an excess of NRM. Young, fit patients at very high risk of relapse might candidate for TBF regimen, providing a careful patient selection and an optimal supportive care in order to minimize transplant toxicity.

Study design and data collection
This is a registry based retrospective study. Data were provided and the study design was approved by the acute leukemia working party (ALWP) of the EBMT group registry, in accordance with the EBMT guidelines for retrospective studies. EBMT is a voluntary working group of more than 500 transplant centers which are required to report all consecutive stem cell transplantations and follow-up once a year. Audits are routinely performed to determine the accuracy of the data. Since 1990, patients have been able to provide informed consent that authorizes the use of their transplant information for research purposes. The ALWP of the EBMT granted ethical approval for this study. We included in the analysis patients with AML older than 18 at diagnosis, who had received either TBF or BF as conditioning regimen for MSD or URD SCT in CR1 as first transplant between January 2007 and June 2015, reported to the EBMT. All unrelated donors were HLA-matched (10/10) or mismatched at one HLA locus (9/10). Patients who received conditioning regimens including oral busulfan, T-depleted grafts, or transplant from <9/10 mismatched unrelated donor were excluded. Myeloablative conditioning regimen (MAC) was defined as ivBusulfan dose ≥9.6 mg/kg (TBF-MAC and BF-MAC), while reduced-intensity conditioning (RIC) was defined as iv Busulfan dose of 6.4 mg/kg www.impactjournals.com/oncotarget (TBF-RIC and BF-RIC) in accordance with the EBMT definitions [28].

End-point definitions and statistical analysis
Primary end-points were overall survival (OS) and leukemia-free survival (LFS). Secondary end-points were relapse incidence (RI), non-relapse mortality (NRM), graft-versus-host free, relapse-free survival (GRFS), engraftment, incidence and severity of acute (aGVHD) and chronic graft-versus-host disease (cGVHD). The severity of acute GVHD was graded on a I-IV scale, while cGVHD was scored as mild, moderate or severe in accordance to EBMT standards [28]. LFS was defined as the interval from transplant to either relapse or death. OS was defined as the time between the date of transplant and the date of death. GRFS was defined as alive with no previous grade III-IV aGvHD, no severe chronic GvHD and no relapse [29]. Probabilities of OS, LFS and GRFS were estimated using Kaplan-Meier curves. Cumulative incidence functions were used to estimate relapse incidence (RI) and non-relapse mortality (NRM) in a competing risks setting. In order to study acute and chronic GVHD, we considered death and relapse as competing events. The main patient characteristics were compared using Mann-Whitney test for quantitative variables, chi-square test or Fisher exact test for categorical variables. Univariate analyses were performed using the log rank test for OS, LFS and GRFS, the Gray test for cumulative incidences. Multivariate analyses were performed using the Cox proportionalhazard model. Factors differing between two groups in terms of distribution and factors significantly associated with the outcome were included in the multivariate analysis. In order to test for a centre effect, we introduced a random effect of frailty for each centre into the model [30]. Finally, a propensity score matched pairs analysis was conducted to corroborate the results obtained in the global population; detailed statistics and results are presented in the appendix. All tests were two-sided and P values < 0.05 were considered as indicating a statistically significant association. Analyses were performed using the R statistical software version 3.2.3 (available online at http://www.R-project.org), and propensity score analysis was performed using the 'MatchIt' [31].

Author contributions
F.S. and A.N. designed the study, the synopsis of which was approved by the acute leukemia working party of the EBMT; M.L. performed all the statistical analysis; F.S. wrote the first draft of the manuscript; A.N. and B.S. reviewed the manuscript; all co-authors contributed data to the EBMT registry, read the manuscript and approved the final version.