NK-cells are a subset of peripheral blood lymphocytes that are innately able to kill malignant cells through different mechanisms based on the balance between activatory and inhibitory signals. Interaction between major histocompatibility complex class I (MHC-I) molecules with killer immunoglobulin receptors (KIRs) on NK cells plays a major role in regulating NK cell function and activity. NK-cells kill leukemia cells when a mismatch between KIRs and their ligand on target cells is present (46). Such activity was demonstrated in T cell-depleted HLA-haploidentical transplant setting by the Perugia group and referred to as “NK cell alloreactivity” (47). The absence of any sort of pharmacologic immune suppression in TCD haplo-HSCT allowed for leukemia killing by alloreactive NK cells. On the other hand, the use of conventional immune suppressives to prevent GvHD in other transplant platforms may limit NK cell alloreactivity and its clinical effect (48). Further, NK cells may be dysfunctional and fail to kill AML blasts in case of abnormal phenotype, decreased degranulation level, and low INF-gamma and TNF-alfa production (32, 49, 50). For many years, NK cell adoptive transfer has been investigated as a possible approach to treat HR-AML. Studies showed donor-derived allogeneic NK-cells achieved durable complete remission in ~33% of HR-AML patients. Such studies proved infusion of a high number of NK-cells to be safe and well-tolerated. Indeed, donor NK cells appear not to cause any GvHD (33–35, 51). Moreover, donor NK-cells are able to persist and expand in vivo after infusion. On the other hand, while promising, allogeneic NK cell adoptive transfer has still limited efficacy, with generally low overall response rate. To overcome such limitations and boost NK cell in vivo function, different protocols and schemes that aim to generate and activate NK-cells are under evaluation and further studies are needed to establish the most effective approach.

CIK cells derived from cytotoxic T lymphocytes (CTL) that are in vitro activated by anti-CD3, OKT3, INF-gamma and subsequently expanded with IL-2. Other than T cell markers, they express surface protein similar to NK-cells, such as CD56, the inhibitory NK receptors, and the natural killer group 2 member D (NKG2D) receptor, one of the most important receptors involved in NK-mediated cancer cell killing (45). In clinical trials, CIK cells have been generated both from autologous and allogeneic lymphocytes and have been infused in combination or not with different strategies of allo-HSCT. Even if the results of early trials were disappointing (37), last studies are more encouraging (36, 38, 39, 52, 53). CIK cell transfer resulted in stable complete remission in ~60% of patients with AML. No significant infusion-related toxicities and a very low rate of acute GvHD were observed after CIK cell infusions. No studies focused on HR-AML, so that the efficacy in this setting still remains to be determined.

CAR-T cells are genetically engineered T cells to express a variable heavy and light chains (V_HL) on cell surface with high specificity for malignant cell antigens (54). Despite the great enthusiasm that followed the CAR-T cell success in the treatment of acute B-lymphoblastic leukemia/lymphoma (B-ALL/LBL) and forms of B-cell lymphoma, generation of CAR-T cells against myeloid leukemic blasts is challenging because of the absence of leukemia-specific target antigens. In fact, AML antigens are often widely expressed by other hemopoietic cells or tissues. While in vitro studies and xenografts demonstrated the effectiveness of anti-CD33 and anti-CD123 CAR-T cells (42, 55), clinical efficacy on AML is still to be confirmed. CD33 is a transmembrane receptor expressed on >90% of blasts, but unfortunately also on multilineage hematopoietic progenitors and myelomonocytic precursors. It was still validated as therapeutic target based on the efficacy of gemtuzumab ozogamicin, a drug-conjugated monoclonal antibody against CD33. Preliminary data of anti-CD33 CAR-T cells are not encouraging (41). CD123 is a transmembrane subunit of the IL-3 receptor expressed on 100% of AML cells, and its expression is increased in FLT3-mutated AML. In vitro ed in vivo (xenograft) preliminary data showed an increased cytokine release and decreased tumor burden using anti-CD123 CAR-T cells. FLT3 receptor is typically expressed on myeloid blasts, independent of FLT3 mutational status. Anti-FLT3 CAR-T cells showed in vitro ed in vivo promising antileukemic effect (56, 57). Moreover, these seem to be less toxic on normal hematopoiesis than the anti-CD33 counterpart. Many other potential targets are now under evaluation. Lewis antigen (LeY) is overexpressed on myeloid blasts in comparison to normal tissues. A trial of autologous CAR-T cells targeting LeY showed a biological response (~60% of patients), but relapse occurred within 2 years (40). An ongoing clinical trial (NCT03222674) evaluates the feasibility, safety, and efficacy of multi-CAR-T cell therapy that targets different AML surface antigens (Muc1/CLL1/CD33/CD38/CD56/CD123) in patients with relapsed/refractory AML. Another phase I study (NCT04156256) evaluates the safety and tolerability of CD123-CD33 dual CAR-T in patients with relapsed and/or refractory AMLs.

In alternative to CAR-T cells, CAR-NK cell therapy ideally combines the specific targeting provided by CARs with the NK cell ability to kill AML blasts in the absence of relevant systemic toxicity. Indeed, CAR-NK cells showed promising results with no important toxicity in lymphoma patients (58). Mouse preclinical models suggest that CD123 CAR-NK cells may be effective in AML (59). Clinical-grade CAR-NK cells can be manufactured from multiple sources (e.g., peripheral blood mononuclear cells, umbilical cord blood, hematopoietic progenitors, induced pluripotent stem cells), including the recently introduced CAR NK-92 cells, which consist of a modified CAR-engineered form of the NK-92 cell line. Such cell line represents an easily manageable and cost-effective tool for large-scale production of CAR-NK cells. Conversely, few drawbacks should be taken into account when using such strategy for CAR-NK cell manufacturing: i) failure of an in vivo expansion, due to lethal irradiation before infusion; ii) lack of NK-cell activating molecules (CD16 and NKp44); iii) potential in vivo tumorigenicity (60). The first-in-human clinical trial using CD33 CAR-NK cells derived from engineered NK-92 cells on three relapse/refractory extramedullary AML patients had no encouraging results (1/3 reached a transitory complete remission of 4 months) (44). Other trials with CD33 CAR-NK cells are under investigation (NCT02892695, NCT02944162). Such studies will help to clarify whether combinatorial strategies can provide antileukemic activity in the absence of relevant toxicity. While there was no specific focus on HR-AML in these preliminary studies and no clear studies showed HR-AML to be particularly sensitive to immune killing, the development of an effective anti-AML CAR-T or CAR-NK cell approach might provide a potent tool for reducing relapse in this high-risk disease.

T cells can be engineered to target different tumor-associated antigens that are frequently expressed in advanced AML blasts and other hematological neoplasms. Preliminary results of such tumor-associated antigen lymphocyte therapy (TAA-T) for different relapsed hematologic malignancies after allo-HSCT (11 patient, Hodgkin’s lymphoma n=2, B-ALL n=3, AML n=5, and 1 HR-AML post 2^nd allo-HSCT) showed that 80% of patients (4/5) with AML achieved a stable complete remission (61). This study also suggested TAA-T to be safe and tolerable (only one patient showed a liver GvHD; no cytokine release syndrome or neurotoxicity was observed). These preliminary promising data suggest that TAA-T therapy may be a feasible option for preemptive treatment of relapse after allo-HSCT for HR-AML, but further clinical studies are needed to ascertain its feasibility and efficacy in this setting. T-cell receptor (TCR)-modified T-cell therapy is a novel emerging strategy using the anti-tumor effect of genetically modifying T cells through the transduction of TCR genes against several cancer antigens (62, 63). The impact of this therapy against specific leukemic antigens is still under investigation. This therapy seems also very safe (64). The in vitro and in vivo preliminary studies on B-malignancies are very promising (65).

A multicenter study of HLA-matched allo-HSCT that employed various immunosuppressive strategies for GvHD prophylaxis (cyclosporine A, tacrolimus, and T-cell depletion) compared a total of 584 patients carrying HR-AML in first complete remission (CR1) from 151 transplantation centers. It showed a median 3-year OS of 45% (range 38–52%), 37% (range 31–44%), and 31% (range 22–41%) in patients undergoing matched sibling donor (MSD), HLA-well-matched and partially-matched unrelated donor (MUD) transplantation, respectively. Myeloablative or reduced-intensity conditioning (RIC) regimens were used. Cumulative incidence (CI) of relapse at 3 years was 37% for MSD, 40% for well-MUD, and 24% for partially-MUD, while 3-year relapse-free survival (RFS) was 42, 34, and 29%, respectively. No significant differences in relapse were observed among the various cytogenetic subsets (66). Another retrospective multicenter study that involved more than 500 transplantation centers reported outcomes of 1,342 patients with CK-AML. Increased risk of relapse correlated with age, secondary AML, active disease at transplant, and the presence of deletion/monosomy 5. High tumor burden before transplant negatively impacted on post-transplantation outcomes. Indeed, 2-year CI of relapse for patients in CR and with active disease at transplantation was 47 and 64%, respectively. A very short OS at 2 years post-transplantation was observed in a subgroup of patients carrying deletion or monosomy 7 and deletion or monosomy 5 (29 and 20% respectively vs 42% in control groups without 7 and 5 deletion/monosomy). No significant survival benefit was observed between fully myeloablative conditioning and RIC regimen for patients with CK AML (34 and 28%). RFS rate was 39.9, 33, and 18.3% for patients ages <40, 40 to 60, and >60 years, respectively (14). Such studies demonstrate that the high relapse incidence after transplant in HR-AML patients is the major limitation of the procedure. Such outcomes urge the development of novel transplantation approaches.

The recent advancements in T-cell manipulation and in GvHD prophylaxis make haplo-HSCT a valuable transplantation strategy to overcome intrinsic chemotherapy resistance of high-risk leukemias. Haplo-HSCT procedures can be mainly divided in two major categories: T-cell depleted (TCD) peripheral-blood stem cells (PBSCs) haplo-HSCT and unmanipulated haplo-HSCT.

Unmanipulated haplo-HSCT relies on pharmacologic GvHD prophylaxis, and it is now adopted worldwide. The use of G-CSF-primed grafts (67, 77), post-transplant high-dose cyclophosphamide (PT-Cy) in combination with other immunosuppressive drugs (78, 79), and post-transplant rapamycin (80), are different approaches that have been tested in this setting. While such strategies help to keep non-relapse mortality (NRM) acceptable, disease relapse remains a major concern, especially when non-myeloablative conditioning regimens are used (81). A study of unmanipulated G-CSF-primed haplo-HSCT showed a 1-year CI of NRM of 36% and a CI of relapse of 21% at 1 year and 28% at 5 years respectively, with a 3-year probability of OS and RFS in 44 and 30%, respectively, in high-risk patients (> second CR or active disease) with hematologic malignancies (including HR-AML) (67). Haplo-HSCT with PT-Cy is now the most widely adopted haplo-HSCT platform, thanks to acceptable rates of acute and chronic GvHD, low NRM, no need of graft manipulation and contained costs. On the other hand, relapse rates are still disappointing in HR-AML patients. In fact, subanalyses showing outcomes of patients with adverse genetic risk AML reported relapse rates up to 50% (14). Because of such limitation and with the goal of reducing leukemia relapse, high-intensity myeloablative conditioning regimens have been employed. Chiusolo P et al. (68) and Devillier R et al. (69) showed a CI of AML relapse of 24% at 4 years and 25% at 2 years, respectively. Further studies will be needed to evaluate if such strategies are effective in subsets of HR-AML patients.

In the last 20 years, several strategies of ex vivo T-cell depletion (TCD) have been tested to improve outcomes of acute leukemia patients who underwent haplo-HSCT. While traditional TCD procedure based on positive selection of CD34+ cells was associated with delayed immune reconstitution and increased risk of NRM (70, 82), more recent strategies are directed towards the preservation of immune subsets that improve post-transplant immune recovery for more effective anti-infective and antileukemic activities (82). Among these, αβ T-cell-depleted haplo-HSCT appears to be an effective platform for the treatment of HR-AML. In αβ T-cell-depleted haplo-HSCT the graft is manipulated to eliminate T cells that express αβ T cell receptor and which are demonstrated to be the main T cell population responsible for alloreactions that cause GvHD. In the studies by Shelikhova L et al. (71) and Maschan M et al. (72), children with primary refractory or relapsed AML who underwent αβ T-cell-depleted haplo-HSCT reached hematologic complete remission, despite 9/22 of them carried adverse-risk cytogenetics. However, the relapse rate and OS at 2 years after allo-HSCT were 42 and 52%, respectively. In different studies by Locatelli F et al. (73) and Bertaina A et al. (74), αβ T-cell and B-cell-depleted haplo-HSCT proved to be a safe and suitable approach in high-risk acute leukemias (HR-AL) in children. Indeed, it achieved a 5-year probability of chronic GvHD-free/relapse-free (GRFS) survival of 71% in HR-AL patients (73). A novel TCD haplo-HCT platform employs grafts that have been selectively depleted of naive T-cells. Indeed, depletion of naïve T cells (T_N) from PBSC preserves hematopoietic engraftment and allows for the transfer of donor-derived memory T cells, that can confer immunity against pathogens with low risk of GvHD (75). This approach has demonstrated to improve outcomes of HR-AL patients (the 2-year relapse rate was 21% and the 2-year RFS was 70%) in a single-arm trial (75). Thus, such approaches are promising, but relapse rates still reduce outcomes of HR-AL patients.

We have recently demonstrated that haplo-HSCT combined with regulatory and conventional T-cells adoptive immunotherapy (Treg-Tcon haplo-HSCT) is able to overcome disease-intrinsic chemoresistance (18, 76). We enrolled 50 AML patients in the study; 40% of them (20/50) had HR-AML. An “age-adapted” myeloablative conditioning based on total body irradiation (TBI) for patients up to the age of 50 years and total marrow/total lymphoid irradiation (TMLI) for patients aged 51–65 years was followed by thiotepa, fludarabine, and cyclophosphamide. No pharmacological GvHD prophylaxis was given. Two millions/kg donor regulatory T cells were given at day −4 to allow for their alloantigen-specific in vivo expansion. One million/kg conventional T cells were given at day −1 and were followed by the infusion of a “megadose” of purified CD34+ hematopoietic progenitor cells at day 0. Fifteen/50 patients developed grade ≥2 acute GvHD (aGvHD). Moderate/severe cGvHD occurred in only one patient. Only two patients relapsed (4%). Consequently, at a median follow-up of 29 months, the probability of moderate/severe cGvHD/relapse-free survival was 75% (18, 76). TMLI allowed to safely extend the powerful effect of a myeloablative conditioning to older (>60 years old) patients. Further, when looking at the different genetic signatures of the enrolled AML patients, we found that HR-AML did not have a higher risk of relapse in comparison to more favorable subgroups. Indeed 17 of the 20 HR-AML patients are alive and leukemia-free despite many of them had detectable disease at transplant. Such results demonstrate HR-AML to be sensible to immune-mediated killing. Indeed, the absence of pharmacologic immune suppression in Treg/Tcon haplo-HSCT could have favored a potent GvL effect that was exerted across all the AML subsets and that was not limited by disease burden and previous refractoriness to chemotherapeutic agents. ELN AML genetic risk stratification is considered to retain outcome prediction after allo-HSCT (83). However, our study showed that effect was lost after Treg/Tcon haplo-HSCT in a single series of 50 AML patients. While larger multicentric studies are needed to support such conclusion, the potent GvL activity of Treg/Tcon haplo-HSCT appears to be an effective tool for the treatment of such unfavorable AML.