NK cell recovery after haploidentical HSCT with posttransplant cyclophosphamide: dynamics and clinical implications

Conceiving strategies to render allogeneic hematopoietic stem cell (HSC) transplantation (HSCT) from HLA-haploidentical family donors safe and feasible has been one of the most challenging efforts faced by the HSCT community over the past several decades. Besides having cured numerous patients that lacked a suitable donor, haploidentical HSCT provided fascinating scientific insights into how the immune system operate upon transfer into an allogeneic environment.^1,2 

One the most remarkable discoveries that originated from early trials of haploidentical HSCT was the description of the principles according to which natural killer (NK) cell alloreactivity ensues, and the observation that, when unleashed, it is accompanied by beneficial effects on HSCT outcome, including protection from relapse.^3-5 

In more recent years, another game-changing discovery stemming from haploidentical HSCT has been the demonstration that high-dose posttransplant cyclophosphamide (PT-Cy) can selectively eliminate the most alloreactive donor T-cell clones in vivo.^6-8  This fostered a true revolution in the field, and haploidentical HSCT platforms based on PT-Cy are increasingly being used worldwide,^9,10  not only because of the impressive abatement of graft-versus-host disease (GVHD) incidence they can convey, but also of their very limited requirements in terms of graft processing and specific expertise from the transplant team. It is largely unknown, however, whether the models that were developed in T cell–depleted haploidentical HSCT still hold true in this setting.

The aim of this study is to trace the dynamics of posttransplantation NK cell recovery in 2 independent series of patients who received haploidentical HSCT with a GVHD prophylaxis based on PT-Cy, and to investigate whether NK cell alloreactivity is preserved in this innovative and increasingly used transplant modality.

Absolute quantification of NK (CD3⁻CD56⁺) and T (CD3⁺) cells was performed in fresh whole blood samples as previously described.¹¹ 

For extended phenotypic analyses, mononuclear cells were isolated from peripheral blood (PB) or bone marrow (BM) by density gradient separation (Lymphoprep; Fresenius). Details on antibodies and panel assembly are provided in the supplemental Methods on the Blood Web site. Acquisition was performed on an LSR Fortessa and an LSRII instrument (both from BD Biosciences). Analysis was performed using FlowJo (TreeStar) and visualized as heatmaps using the pheatmap function in R. Data were further analyzed using the Barnes-Hut stochastic neighbor embedding (bh-SNE) algorithm (using the CYT tool and the MatLab software as described previously¹² ). The input dataset was resampled to obtain an equal number of NK cell events for all of the samples analyzed; the bidimensional maps obtained from this analysis were then processed with FlowJo software to highlight the spatial distribution of NK cells at each time point. Upon bh-SNE analysis,¹²  the k-means algorithm was used for unsupervised clustering of data to identify and quantify in an unbiased manner memory-like NK cells in patient and healthy donor samples.

An in vitro assay to test the effect of mafosfamide on NK cells was adapted from a published protocol.¹³  Briefly, cryopreserved PB mononuclear cells obtained from a donor’s graft were thawed, marked with Cell Trace Violet (CTV, Life Technologies), and cultured in Iscove modified Dulbecco medium (Lonza) supplemented with 10% human serum (Euroclone), 1% penicillin streptomycin, 1% glutamine (Lonza), and 5 μg/mL interleukin-15 (IL-15, Miltenyi). Mafosfamide l-lysine salt (Niomech) was diluted in distilled water and added to the culture after 3 days. After 1 hour of incubation, cells were centrifuged twice and then plated in fresh medium supplemented with IL-15. CTV dilution and quantification of live nonapoptotic cells (Annexin V⁻) were assessed by flow cytometry, before (day 3) and 2 days after (day 5) adding mafosfamide. The same experiment was performed in the absence of CTV labeling to analyze proliferation using Ki-67 intracellular staining.

Cryopreserved PB mononuclear cells obtained from HSC donors before granulocyte colony-stimulating factor mobilization, from patients at day 30 after HSCT, and from healthy subjects, were thawed and plated in Iscove modified Dulbecco medium supplemented with 10% human serum in the presence of 20 IU/mL of recombinant human IL-2 (Novartis). After overnight incubation, NK cells were isolated by negative selection with anti-CD3, anti-CD14, and anti-CD19 magnetic beads (Miltenyi), labeled with CTV and plated in a 50:1, 10:1, or 1:1 effector:target ratio in the presence of the following targets: K562 cell line, OCI/AML3 cell line, and primary AML blasts collected from a patient at diagnosis (90% purity). After a 6-hour incubation, apoptosis of CTV-negative target cells was analyzed by assessing Annexin V (Biolegend) expression by flow cytometry.

A detailed description of statistical analyses performed to compare experimental data and patient outcome is provided in the supplemental Methods.

To assess NK cell dynamics after haploidentical T cell–replete HSCT with PT-Cy, we analyzed PB and BM samples collected longitudinally in time from 17 patients treated over the same period in 2 transplantation centers (Ospedale San Raffaele, Milano, Italy [OSR]; John Hopkins University, Baltimore, MD [JHU]). Patient and transplant characteristics are provided in the supplemental Methods and summarized in Table 1; HSCT platforms are outlined in Figure 1A.

After infusion of the graft, which for all patients included an elevated content of mature donor NK cells (median, 9.18× 10⁶/kg; range, 1.66-26.2 × 10⁶/kg), absolute T-cell and NK cell counts were measured in PB before (day 3) and after (days 5, 8, 15, and 30) cyclophosphamide administration (Figure 1A-B). As previously reported,^14-16  T and NK cells were detectable at very low levels even immediately after graft infusion. PT-Cy treatment resulted in a further decrease of cell counts, with residual NK cells barely detectable at day 5. After that, T and NK cells progressively increased at days 15 and 30 after HSCT. Patients treated in the 2 transplant centers displayed similar in vivo dynamics, with a slight delay in recovery in JHU patients (Figure 1C). To assess the origin of the circulating NK cells, we analyzed lymphocyte chimerism in 3 donor-recipient pairs mismatched for HLA-A*02. In both the T- and NK-cell compartments, donor-derived lymphocytes accounted for the majority of cells even at early time points (Figure 1D); host-derived lymphocytes became undetectable by day 30. A minor deflection of chimerism was observed at day 8, suggesting a preferential sensitivity of donor-derived cells to PT-Cy administration.

Cyclophosphamide is an alkylating agent known to be particularly active on dividing immune cells.^7,17  We thus used Ki-67 staining to investigate the effects of PT-Cy on NK cell proliferation. At day 3 after transplantation, NK cells robustly proliferated, even to a higher extent than T cells (Figure 2A-B). As previously described for T lymphocytes,^14,15  we found that NK cell proliferation was completely abrogated by day 8, suggesting selective elimination of cycling NK cells. Importantly, similar NK cell proliferation kinetics were observed in patients from the 2 centers (Figure 2C), suggesting that this dynamic is minimally influenced by differences in graft source, conditioning intensity, and subsequent pharmacological GVHD prophylaxis. Interestingly, in patients in which the donor or host origin of circulating NK cells could be assessed through the expression of mismatched HLA-A*02, we found that the in vivo proliferation was largely restricted to donor-derived NK cells (Figure 2D).

These observations suggested that actively proliferating donor-derived NK cells were selectively eliminated upon cyclophosphamide administration. In line with this hypothesis, we failed to record any activity of ALDH, an enzyme known to confer intrinsic resistance to Cy,^13,18,19  in NK cells contained within the graft or in PB samples collected at day 3 after infusion (Figure 2E).

To directly demonstrate that proliferating NK cells are selectively eliminated by Cy, we mimicked in vitro the transplantation protocol by prompting the proliferation of NK cells with recombinant human IL-15 and challenging them after 3 days with mafosfamide, a Cy analog suitable for in vitro use. After 2 additional days, proliferating NK cells had been eliminated by mafosfamide in a dose-dependent manner, as demonstrated by a progressive decrease in the counts of viable cells (Figure 2F) and in the abrogation of proliferation (supplemental Figure 1A-C). Conversely, T cells and CD34⁺ stem cells from the graft did not proliferate in these experimental conditions and were consequently not eliminated by in vitro administration of the drug. Overall, these data demonstrate that NK cells contained in the graft proliferate robustly upon infusion into patients, and that the proliferation state and lack of ALDH of these NK cells jointly cause them to be purged in vivo by PT-Cy.

IL-15 is a “4 α-helix bundle” cytokine primarily produced by monocytes, macrophages, and dendritic cells, and is known to play a pivotal role in lymphocyte homeostasis and especially in NK cell development, maturation, and proliferation.²⁰  In a lymphodepleted environment, such as the one established in patients after conditioning, elevated IL-15 levels can affect the fate and the function of the infused donor NK cells.^21,22  We thus measured the serum concentration of IL-15 longitudinally during the first months after haploidentical HSCT with PT-Cy (details provided in supplemental Methods).

Systemic levels of IL-15 rose immediately after conditioning (Figure 3A), possibly favoring the donor NK cell proliferation seen early after graft infusion (Figure 2B). Interestingly, the high levels of IL-15 documented at day 3 displayed a direct correlation with the number of NK cells reconstituting thereafter (Figure 3B) and an inverse correlation with the frequency of NK cells proliferating at the same time point (Figure 3C). A potential explanation is that higher early proliferation of NK cells consumes IL-15, leading to lower measurable levels in these patients. Starting from day 15, a second wave of highly proliferating donor NK cells appeared in the PB of patients, followed by a drop in IL-15 levels at day 30, suggesting again consumption of the cytokine by reconstituting lymphocytes.

Although much remains to be discovered about the factors that regulate NK cell dynamics in the early posttransplantation phase, these findings highlight that the cytokine milieu early after haploidentical HSCT with PT-Cy include high levels of IL-15, and that these are significantly correlated with the metrics of NK cell recovery.

Our analysis of absolute counts (Figure 1B-C) and proliferation kinetics (Figure 2B-C) of NK cells after haploidentical HSCT and PT-Cy demonstrated that a sizeable population of highly proliferating NK cells becomes evident around day 15 after HSCT. To evaluate the ability of these cells to provide an effective protection against pathogens and disease recurrence, we assessed their phenotypical features and maturation status. The expression of maturation markers was assessed through multiparametric flow cytometry on NK cells contained within the donor leukapheresis (shown as day 0 in Figure 4B), and on patient PB NK cells immediately after HSCT (day 3) and after PT-Cy (days 5, 8, 15, and 30). The results for OSR patients are shown in Figure 4A-B; JHU patient results appear in supplemental Figure 2A-B.

As expected, NK cells in the donor graft exhibited high expression of markers associated with a mature phenotype (CD16, KIR, NKG2C, and CD57) and low expression of markers of immature NK cells (NKG2A and CD62L). Although the phenotype of NK cells circulating in the PB of patients immediately after HSCT (day 3, Figure 4A-B) resembled that of infused NK cells, the NK cell phenotype seen after PT-Cy administration was significantly different. By day 8, in fact, expression of maturation markers on NK cells became low or absent, whereas the percentage of CD56^bright NK cells and the overall expression of NKG2A and CD62L increased. Unsupervised clustering of fluorescence-activated cell sorter data demonstrated clear differences between NK cells found in the graft and in patients immediately after transplant compared with patient NK cells at days 15 and 30, both in OSR and in JHU patients (Figure 4C; supplemental Figure 2A-C).

To validate these findings, the same flow cytometry data were analyzed using the bh-SNE algorithm for unbiased high-dimensional analysis.¹²  NK cell events from all samples were plotted together on a bidimensional map based on the similarities of expression of maturation markers (Figure 4D). NK cells harvested from healthy subjects, grafts, patients pretransplant, and at the early time points after HSCT (from days 3 to 8) clustered in the upper portion of the multidimensional map (Figure 4E), characterized by high positivity for CD16, CD57, and KIR molecules and by intermediate expression of NKG2C. Conversely, NK cells harvested from patients at days 15 and 30 after HSCT grouped in the opposite part of the map, denoting high expression of NKG2A and CD62L molecules (Figure 4D-E). Overall, these data suggest that after the rapid purging of mature proliferating lymphocytes operated by PT-Cy, the subsequent NK cell recovery occurs principally through maturation from graft progenitors rather than by homeostatic expansion of mature graft NK cells.

To address whether, and if so at what time, a mature and functional NK cell repertoire is restored, we longitudinally traced the phenotype of circulating NK cells during the first year after transplant in OSR patients. Consistent with previous reports,²³  we documented stable physiological absolute counts of circulating NK cells from 1 month after HSCT onwards (Figure 5A). Because of the relatively low number of circulating T cells in PB during the first 3 months after transplant, NK cells were the most abundant lymphocyte subset during this time (Figure 5B). Many features linked to immaturity, such as increased proportions of the CD56^bright subset and of CD62L⁺ NK cells, decreased over time and regained normal levels after 9 to 12 months posttransplant (Figure 5C). An opposite trend, but with different kinetics, was observed for the expression levels of the maturation markers CD16 and CD57. The expression of KIRs returned to normal levels by day 60 (Figure 5D), but, as previously reported in other HSCT settings,^24,25  the levels of the inhibitory receptor NKG2A diminished only 6 months after HSCT, suggesting that the ability of circulating NK cells to use target recognition through HLA class I-KIR interactions is not fully recovered before this time point. Of notice, we documented higher than normal expression of the activating receptors NKG2C in reconstituting NK cells from our patients, and acquisition of this receptor was even faster in patients who developed cytomegalovirus (CMV) reactivations during their posttransplant follow-up, in line with previous reports^26,27  (data not shown).

As observed for maturation markers, the expression of DNAM-1, 2B4, KLRG-1, and CD8 also rapidly decreased early after transplant and normalized 2 to 6 months after HSCT (Figure 5E). Finally, in line with other HSCT settings,²⁴  we did not detect any significant deviations from normal reference values in natural cytotoxicity receptor (NKp30, NKp44, and NKp46) expression in PT-Cy–based haploidentical transplant recipients (Figure 5F). Of note, throughout the entire follow-up period, the receptor repertoire appeared nearly identical between NK cells circulating in the PB and those harvested at the same time point from the BM (supplemental Figure 3). Overall, these data indicate that the phenotypic recovery of the NK cell repertoire is a long process, taking up to 1 year after haploidentical HSCT with PT-Cy, similarly to what has been reported in other HSCT settings.^24,25 

In the context of T cell–depleted haploidentical HSCT, numerous studies demonstrated that donor-recipient KIR ligand mismatches can unleash reconstituting donor NK cells against residual tumor cells.^2-5  NK cell–mediated alloreactivity is, according to this model, mediated by “single-KIR⁺” donor NK cells, expressing on their surface a sole inhibitory receptor that can bind ligands present in the donor and absent in the recipient.^24,28,29  We thus chose to explore this specific NK cell subset to address the effect of PT-Cy on NK cell–mediated alloreactivity.

Using multiparametric flow cytometry, we compared the dynamics of potentially alloreactive single-KIR⁺ NK cells with their nonalloreactive counterparts and with total NK cells. We found similarly high levels of proliferation in all 3 groups of NK cells at day 3 after graft infusion (Figure 6A) and an analogous behavior in response to PT-Cy (Figure 6B). These ex vivo observations, limited by the low number of NK cells circulating early after HSCT, were supported by in vitro experiments, in which alloreactive and nonalloreactive NK cells from PBSC grafts proliferated in a comparable manner upon exposure to IL-15, and were equally susceptible to mafosfamide (Figure 6C-D).

Indeed, we documented that single-KIR⁺ NK cells present in the graft became almost completely undetectable in the PB of patients at day 30 (0.26 ± 0.19% of total NK cells, P < .0001, Figure 6E).

To investigate the functional consequences of this marked reduction of single KIR⁺ after PT-Cy, we compared the antileukemic potential of NK cells purified from 8 patients at day 30 after transplant. Compared with their counterparts from the corresponding donor PB, patient NK cells displayed impaired killing of the OCI/AML03 cell line and of primary AML blasts (P = .026 and P = .004, respectively), whereas the ability to kill the HLA class I–negative K562 cell line was less affected (Figure 6F; supplemental Figure 4).

Although these in vitro data suggest that NK cells reconstituting early after haploidentical HSCT with PT-Cy display an impaired antileukemic potential compared with their mature donor counterparts, this small but significant difference might still be overruled by other factors in vivo. To provide more insights into this relevant issue, we analyzed the clinical impact of predicted NK alloreactivity in a cohort of 99 consecutive patients who received haploidentical HSCT at OSR. All patients received a myeloablative chemotherapy-based conditioning and a GVHD prophylaxis based on PT-Cy. KIR ligand mismatches were present in 41 of 99 patients (Table 2). In line with our expectations, and contrary to what has been consistently documented in T cell–depleted haploidentical transplants,^2-4  predicted NK cell alloreactivity did not significantly affect any of the major HSCT end points (including GVHD, relapse incidence, and survival) either in the entire cohort (Figure 6G-H; supplemental Figure 5) or in subgroup analysis (supplemental Table 2). These data suggest that by eliminating the majority of mature alloreactive NK cells transferred as part of the graft, PT-Cy dampens the impact of KIR ligand mismatches on HSCT outcome.

Finally, to verify whether the proportion of mature NK cells spared by PT-Cy might have a clinically relevant role in determining HSCT outcome, we analyzed the maturation phenotype of NK cells circulating at day 30 after HSCT in 59 OSR patients who underwent haploidentical HSCT with PT-Cy and linked it to disease-related clinical end points. Beside the analysis of single maturation markers, we also enumerated CD57⁺CD16⁺KIR⁺NKG2C⁺ mature memory-like NK cells, based on unsupervised data clustering of bh-SNE data using the k-means algorithm (supplemental Figure 6). In univariate analysis, low expression of CD62L and high expression of KIRs on NK cells at day 30 after HSCT were significantly correlated with lower relapse incidence (P = .03 and P = .0003, respectively), and high absolute counts of total and memory-like NK cells displayed a trend toward significance for the same end point (P = .08 and P = .07), in line with previous reports.³⁰  Moreover, high expression of KIRs on NK cells at day 30 displayed a significant correlation with higher progression-free survival (P = .02, Figure 7A-D; supplemental Table 3). Despite the relatively small size of our patient series, the significant correlation between KIR expression on NK cells at day 30 and lower relapse incidence held true in multivariate analysis (P = .012; supplemental Table 4), suggesting that KIR expression might represent a clinically relevant proxy for the functional competence of reconstituting NK cells in preventing recurrence.

By counterposing the immune systems of 2 individuals with disparate HLA and KIR assets, haploidentical HSCT represents a unique model to investigate the biological determinants of NK cell tolerance and alloreactivity and has provided convincing in vivo evidence of the immunotherapeutic potential endowed in this cell subset.^2,5,22,31  However, with the ongoing development of strategies to render HLA-mismatched transplant more feasible, it is becoming increasingly evident that the complex dynamics of NK cell recovery are dramatically affected by changes in the conditioning regimen and GVHD prophylaxis: for this reason, immunogenetic models to predict NK cell–mediated effects have to be validated in each HSCT platform.

In the present study, we integrated immunogenetic, functional, and clinical data to analyze the metrics and correlates of NK cell repertoire reconstitution in patients who received haploidentical HSCT according to 2 different PT-Cy platforms. We undertook this to address whether the models formulated in T cell–depleted haploidentical HSCT also hold true in this more recent and increasingly common transplantation modality.

We showed that the high numbers of mature NK cells infused as part of the graft in this HSCT setting immediately encounter high levels of homeostatic cytokines released as a consequence of the conditioning regimen–mediated lymphodepletion. As a result, we observed that infused NK cells are prompted to proliferate to an even higher extent than their T cell counterparts, and showed in vitro and in vivo that such proliferation confers NK cell sensitivity to Cy-mediated killing.

After the purging of the majority of mature donor NK cells from the graft by PT-Cy, we observed the appearance of a second wave of donor-derived NK cells. These cells, however, displayed phenotypic and functional features of immaturity for several months after HSCT, strongly suggesting that they might stem directly from HSCs contained in the graft rather than from the mature NK cells infused alongside them.

Although it is extremely difficult to track their complex dynamics during the early posttransplant phase because of their very low numbers, single-KIR⁺ NK cells, considered to include potentially alloreactive NK cells, appeared to behave similarly to all other mature NK cells and were thus almost completely eliminated by PT-Cy. As a consequence, the most widely accepted algorithm used to predict beneficial NK alloreactivity in T cell–depleted transplants³  failed to predict clinical outcome in our patient series. Conversely, and consistently with the role we described for PT-Cy in this setting, the absolute counts and relative proportion of mature NK cells at day 30 after HSCT appeared to be a more reliable predictor of an effective NK cell–mediated immunosurveillance against relapse in this HSCT platform.

Our results have several other practical implications: whereas, for instance, it is becoming widely accepted that transplantation from HLA-mismatched relatives represents a viable alternative for patients who lack a conventional matched donor,^9,10  the question of which donor to select among several possible family members is still debated.^32,33  Here we found that the “classical” KIR ligand mismatch model of NK cell alloreactivity did not correlate with any of the major HSCT end points in PT-Cy–based haploidentical HSCTs, indicating that, in this context, priority should be given to other donor selection criteria, including CMV serostatus, gender mismatch, and presence of donor-specific anti-HLA antibodies in the recipient. It should be considered, however, that different NK cell alloreactivity models, also taking into account the donor-recipient KIR genotype, might be needed to predict NK-mediated effects in this context, as widely accepted for HLA-matched donor HSCT,^34,35  and already proposed by other studies for the haploidentical context.^36,37 

A second observation relates to the finding of a relatively long time window in which most of the circulating NK cells display phenotypic features of immaturity. Previous reports have highlighted that, despite the relatively rapid recovery of CD8 T-cell counts, PT-Cy transplantation is often accompanied by frequent viral reactivations, including those from CMV, polyomaviruses, and human herpes virus 6.³⁸  We can speculate that at least part of this susceptibility might be due to the ineffective protection conferred by the immature donor-derived NK cells, and future studies should verify whether the dynamics of recovery of memory NK cells could predict antiviral immune competence more effectively than currently used biomarkers.

Besides pointing out these limitations in NK cell defense intrinsic to PT-Cy–based haploidentical HSCTs, our study provides a framework to integrate new strategies to better exploit the immunotherapeutic potential of NK cells for this transplant platform.

For instance, the marked increase in IL-15 levels in the time window that immediately follows PT-Cy administration, reaching levels not commonly detected in physiological conditions, suggests that there is a temporal window here with a very favorable cytokine milieu for the in vivo expansion of alloreactive NK cells and provides the rationale for adoptive infusion of mature donor NK cells in this time frame. Of notice, several investigational trials of PT-Cy–based haploidentical HSCT followed by infusion of donor NK cells have recently been initiated and already reported promising results.^39-41 

In addition, several new agents developed to improve NK cell–mediated antileukemic responses might be integrated into PT-Cy–based haploidentical HSCTs. For instance, the recently developed anti-CD16/anti-CD33/IL-15 TriKe constructs⁴²  might be of benefit during the intermediate phase of recovery, when CD16 expression has been recovered but the KIR/NKG2A system has not yet reached its final equilibrium and IL-15 serum concentration has dropped to levels that might be insufficient to sustain the in vivo expansion and persistence of adoptively transferred NK cells.

An additional ground for improvement relates to the pharmacological GVHD prophylaxis used after PT-Cy administration. In accordance with the majority of current protocols for T cell–replete haploidentical HSCT, both regimens used in our study included mycophenolate mofetil during the first month after transplant: a number of studies have shown that this drug can negatively affect NK cell function^43,44 ; thus, together with the in vivo purging of mature alloreactive NK cells, it might further reduce the antileukemic potential of early reconstituting NK cells. Importantly, the observed effects may also relate to other pharmacological agents and GVHD prophylaxis, an issue to be addressed in more detail in future studies.

In conclusion, our study strongly supports the notion that the complex and not yet fully elucidated dynamics of NK cell recovery after haploidentical HSCT are heavily affected by competition with other immune subsets and by the drugs used in the transplantation protocol. Whereas models of predicted NK cell alloreactivity developed in the original T cell–depleted setting held true in transplants performed with selective depletion of T-cell receptor αβ positive T cells from the graft,⁴⁵  or with coinfusion of donor-derived regulatory and conventional T cells,^46,47  the same algorithms were unable to predict outcome in the context of GIAC protocols⁴⁸  or, as shown in this report, in PT-Cy–based transplants.^36,37  In the dawning era of “personalized” haploidentical transplants, it is necessary to adapt tools to predict outcome to an informed analysis of the dynamics of immune recovery in each HSCT setting and to leverage this knowledge to develop appropriate adjunctive strategies to improve HSCT outcome.

Work at Ospedale San Raffaele was supported by grants from the Italian Ministry of Health (RF-2011-02351998, RF-2011-02348034, and TRANSCAN HLALOSS), the Associazione Italiana per la Ricerca sul Cancro (Start-Up Grant #14162), the ASCO Conquer Cancer Foundation (2014 Young Investigator Award), and the DKMS Mechtild Harf Foundation. G.O. was supported by a Fondazione Matarelli fellowship from the Associazione Italiana Leucemie and by a Fondazione Umberto Veronesi fellowship. Work at Johns Hopkins University was supported by the National Institutes of Health National Heart, Lung, and Blood Institute (grant R01HL110907) (L.L.) and by Wenner-Gren Foundation fellowship (S.B.).

Contribution: A.R., G.O., F.C., L.L., and L.V. designed the study. A.R., G.O., S.B., V.G., N.C., C.T., and L.Z. performed experiments. R.G., S.P., M.M., F.G., A.A., M.T.L.S., J.P., L.L., and F.C. collected and analyzed patient clinical data. A.R., F.L. and S.P. performed statistical analyses, C.B. provided scientific counseling, A.R., G.O., F.C., L.L., and L.V. wrote the paper. All authors read and approved the final version of the manuscript.

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

The current affiliation for N.C. is the University of Milan, Milan, Italy.

Correspondence: Fabio Ciceri, Hematology and Bone Marrow Transplantation Unit, San Raffaele Scientific Institute, via Olgettina 60, Milano, Italy; e-mail: ciceri.fabio@hsr.it.