A Focus on Intermediate-Risk Acute Myeloid Leukemia: Sub-Classification Updates and Therapeutic Challenges

Simple Summary Risk stratification models, including the European LeukemiaNet 2017 and 2022 guidelines, categorize newly diagnosed acute myeloid leukemia (AML) patients into several subgroups of distinct genetic characteristics and disease outcomes. The intermediate-risk group remains the most heterogenous group, as most AML patients fall into it (i.e., a basket category) by virtue of not fulfilling criteria that identify specific entities (e.g., core-binding factor AML, TP53 mutations, complex karyotypes) of well-recognized prognostic significance. In this review, we aim to discuss the latest updates on intermediate-risk definition and highlight the therapeutic advances and challenges that warrant refining the prognostic classification of this category. Abstract Acute myeloid leukemia (AML) represents a heterogeneous group of hematopoietic neoplasms deriving from the abnormal proliferation of myeloid progenitors in the bone marrow. Patients with AML may have highly variable outcomes, which are generally dictated by individual clinical and genomic characteristics. As such, the European LeukemiaNet 2017 and 2022 guidelines categorize newly diagnosed AML into favorable-, intermediate-, and adverse-risk groups, based on their molecular and cytogenetic profiles. Nevertheless, the intermediate-risk category remains poorly defined, as many patients fall into this group as a result of their exclusion from the other two. Moreover, further genomic data with potential prognostic and therapeutic influences continue to emerge, though they are yet to be integrated into the diagnostic and prognostic models of AML. This review highlights the latest therapeutic advances and challenges that warrant refining the prognostic classification of intermediate-risk AML.


Introduction
Acute myeloid leukemia (AML) is a diverse group of hematopoietic clonal disorders characterized by the accumulation of immature myeloid progenitors [1]. Despite recent advances in genomics and therapeutics, long-term outcomes are dismal in many AML subsets. Patient characteristics such as age and performance status (overall, summarized by the concept of fitness [2]) and heterogeneous cytogenetic and genomic underpinnings constitute the main determinants of outcomes [3]. Recent advances in genome scanning techniques have increased our understanding of AML pathogenesis, unveiling a plethora of driver gene lesions and possible disease-outcome modifiers [4]. Despite the abundance of genomic information on AML, current diagnostic and prognostic classifications somehow remain inexact. This is particularly true for the subgroup of patients classified as intermediate-risk according to the standard prognostic criteria used for AML clinical management (the European LeukemiaNet, ELN 2017 [5] and 2022 revision [6]). Apart from FLT3 mutants, most patients usually fall into this group (an umbrella category) by virtue of not fulfilling criteria that identify specific entities (e.g., core-binding factor AML, TP53 mutations, complex karyotypes) with well-recognized prognostic significance.

Intermediate-Risk Definition and Prognosis
AML classification and prognostic criteria are based on cytogenetic and molecular features at the time of diagnosis. At least 50% of cases exhibit karyotypic abnormalities by conventional cytogenetic analysis; in addition, mutational screening reveals at least one somatic alteration in virtually all patients [7,8]. Traditionally, AML has been categorized into favorable-, intermediate-, and adverse-risk groups [9]. While the favorable-and adverse-risk groups represent well-defined categories, the intermediate-risk group is generally characterized by the absence of favorable or unfavorable cytogenetic and molecular abnormalities and encompasses variable outcomes with standard-of-care therapies.
Modern genomic profiling using high-throughput technologies has led to the identification of molecular alterations reflective of disease biology and predictive of outcomes. For instance, identifying somatic mutations in the FLT3, NPM1, CEBPA, IDH1, IDH2, DNMT3A, TET2, and KIT genes provided important information that improved AML diagnosis, treatment stratification, and monitoring [10][11][12]. The ELN AML risk stratification in 2017 [5] (Table 1) integrated additional cytogenetic and molecular abnormalities [5] not present in the 2010 version. A subsequent study of 1116 AML patients validated the ELN 2017 prognostic stratification and showed that it assigned more patients to the favorableand adverse-risk groups than the 2010 criteria, thereby shrinking the intermediate-risk group [13]. Nevertheless, this remained an ill-defined "basket category" for patients not fulfilling the criteria for the favorable-or adverse-risk groups. Moreover, with regard to clinical management, while current indications for consolidation treatment with allogeneic hematopoietic stem cell transplantation (HSCT) are well-established in favorable-and adverse-risk patients, information derived from disease monitoring is crucial in clinical decision-making in intermediate-risk patients because of their heterogeneity [14,15].  [5].
Interestingly, in a retrospective study of 863 de novo AML patients < 60 years, Eisfeld et al. found additional gene mutations able to refine the prognostic classification of ELN 2017 [16]. In this study, the authors found a total of 2354 mutations, averaging 3 per individual. WT1/NPM1 co-mutant, DNMT3A, and ZRSR2 mutated patients had very poor survival outcomes, similar to those of the adverse-risk group. Of note, BCORand SETBP1-mutated AML patients were categorized as favorable-risk, whereas IDH mutants were associated with adverse-risk patients and had outcomes similar to those of intermediate-risk patients. They also found that FLT3 internal tandem duplication (ITD) with a high allelic ratio had survival outcomes similar to those of the adverse-risk group rather than the intermediate-risk category, regardless of NPM1 mutational status. Based on these findings, 9% of favorable-risk, and 53% of intermediate-risk AML patients per ELN 2017 could be reclassified as belonging to the adverse-risk category, and 9% of adverse-risk and 4% of favorable-risk patients could be reclassified as intermediate-risk. Thus, the current AML prognostic criteria do not encompass the molecular heterogeneity inherent to the disease, and the inclusion of mutations in additional myeloid genes may refine classification [17]. However, identifying genomic alterations and their prognostic significance is still an evolving field, and guidelines will require ongoing refinements.
Given the expansion of the AML therapeutic arsenal, consideration of the mutational status of additional genes is of utmost importance also in older patients. Recently, the ALFA-1200 study evaluated genomic predictors of outcomes in 471 newly diagnosed AML patients ≥ 60 years treated with standard "7 + 3" induction chemotherapy [18]. The group developed a decision tool to predict survival outcomes in older AML patients using mutations in seven genes, NPM1, NRAS, KRAS, ASXL1, DNMT3A, TP53, and FLT3-ITD. In patients with poor-risk cytogenetics, TP53 and KRAS mutations were independently associated with poor survival, whereas SETBP1, NRAS, ASXL1, and RUNX1 status predicted lower complete remission rates even in patients with favorable-and intermediate-risk cytogenetics. Additionally, mutations in DNMT3A, ASXL1, NRAS, NMP1, and FLT3-ITD (regardless of allelic ratio) were independent predictors of overall survival (OS). When cytogenetic risk and mutations in the seven genes were combined, 39.1% of patients had a 2-year OS of 66.1% (assigned to "go-go" group), 7.6% of patients had a 2-year OS of 2.8% (assigned to "no-go" group), and 3.3% of patients had a 2-year OS of 39.1% (assigned to "slow-go" group). This genetic-based predictive model was validated in three independent cohorts.
Besides genomic characteristics at the time of diagnosis, host-related factors such as poor baseline functional status, multiple comorbidities, and advanced age are associated with poorer prognosis (Figure 1) [19].
Interestingly, machine learning (ML) techniques have been applied in leukemia research and represent a promising auxiliary tool for developing new approaches that aim to improve AML risk stratification. Several studies have implemented unsupervised and unbiased ML methods that demonstrated high accuracy in terms of genomic classification [20]. These approaches may be further employed to define ambiguous definitions by unmasking unexplored clinico-morphologic and genomic characteristics unique to intermediate-risk AML patients. Indeed, ML-based methods demonstrated a 97% accuracy rate when reclassifying AML patients according to four genomic signature clusters [4]. While the clusters exhibited a certain degree of overlap with the 2017 ELN groups, significant differences in survival across the two classification systems emphasized the need for a more subtle distinction of AML risk groups.
Most recently, the 2022 ELN guidelines for diagnosis and management of AML were published [6]. In this newest version, the risk classification of AML has changed to using a hierarchical model that depends on molecular aberrations detected rather than morphologic or historical criteria (e.g., therapy-related, secondary AML). Moreover, the intermediate-risk AML group was redefined with the abrogation of the value given to FLT3 allelic ratio, as demonstrated in Table 2. Table 2. 2022 European LeukemiaNet (ELN) risk classification by genetics at initial diagnosis. Adapted from [6].

Intermediate-risk category
Mutated NPM1 † ‡ with FLT3-ITD Wild-type NPM1 with FLT3-ITD t(9;11)(p21.3;q23.3)/MLLT3::KMT2A * † Cytogenetic and/or molecular abnormalities not classified as favorable or adverse † Mainly based on results observed in intensively treated patients. Initial risk assignment may change during the treatment course, based on the results from analyses of measurable residual disease. ‡ AML with NPM1 mutation and adverse-risk cytogenetic abnormalities are categorized as adverse-risk. * The presence of t(9;11)(p21.3;q23.3) takes precedence over rare, concurrent adverse-risk gene mutations. Interestingly, machine learning (ML) techniques have been applied in leukemia research and represent a promising auxiliary tool for developing new approaches that aim to improve AML risk stratification. Several studies have implemented unsupervised and unbiased ML methods that demonstrated high accuracy in terms of genomic classification [20]. These approaches may be further employed to define ambiguous definitions by unmasking unexplored clinico-morphologic and genomic characteristics unique to intermediate-risk AML patients. Indeed, ML-based methods demonstrated a 97% accuracy rate when reclassifying AML patients according to four genomic signature clusters [4]. While the clusters exhibited a certain degree of overlap with the 2017 ELN groups, significant differences in survival across the two classification systems emphasized the need for a more subtle distinction of AML risk groups.
Most recently, the 2022 ELN guidelines for diagnosis and management of AML were published [6]. In this newest version, the risk classification of AML has changed to using a hierarchical model that depends on molecular aberrations detected rather than morphologic or historical criteria (e.g., therapy-related, secondary AML). Moreover, the intermediate-risk AML group was redefined with the abrogation of the value given to FLT3 allelic ratio, as demonstrated in Table 2. In the lower panel, current research approaches to ameliorating the outcomes of such patients are illustrated. Specifically, the application of genome scanning techniques and machine learning methods may help better characterize AML populations to assess geno-phenotypic associations and identify therapeutic vulnerabilities.

Current Therapeutic Approaches
The current AML treatment paradigm recommended by the major societies, including NCCN and ESMO, is heavily based on four factors: age, performance status, organ function (in other words, patients' fitness), and risk classification per ELN 2017 criteria [5]. Of note is that there is a lack of trials solely focusing on patients with intermediate-risk AML. Hence, the current evidence-based therapeutic approaches in this population are mainly extrapolated from large studies that included different risk categories of patients. In this scenario, it is essential to recognize that the heterogeneity in the studied populations makes generalization difficult and might over-or underestimate the actual benefit of any treatment in such subgroup. In addition, none of the current Food and Drug Administration (FDA)approved agents are specific for the intermediate-risk category, thereby configuring an unmet clinical need in AML. As the prognostication of AML continues to evolve and trial inclusion criteria change, therapeutic recommendations from major societies will likely be modified accordingly. Moreover, since the criteria used to define risk groups significantly vary between trials, it is challenging to determine the benefit of treatments and compare it among different studies.
Treatment of intermediate-risk ELN patients eligible for intensive therapy has not substantially changed over the past few decades, and "7 + 3" induction therapy still represents the standard of care. Data from three randomized trials comparing anthracycline doses (daunorubicin 45 mg/m 2 and 90 mg/m 2 ) show a survival benefit with higher doses of anthracyclines in both younger and older patients (at least up to the age of 65 years) [21][22][23]. These regimens lead to complete remission (CR) in 60-80% of younger adults and 40-60% of older patients. The benefit of adding gemtuzumab ozogamicin (GO), a humanized anti-CD33 monoclonal antibody conjugated to calicheamicin, to standard therapy was limited to favorable-and intermediate-risk patients [24][25][26]. Table 3 summarizes therapeutic regimens recommended by the 2021 NCCN and the 2020 ESMO guidelines for use in intermediate-risk AML. In this subgroup, for patients aged <60 years and with intact organ function, current recommendations support induction chemotherapy, as described above [27,28]. Patients who achieve a CR to induction chemotherapy receive consolidation chemotherapy, most commonly with high-dose cytarabine, but this may vary based on the chosen induction regimen. [29]. Cytarabine administered at 3 g/m 2 (high-dose) by 3 h infusion, every 12 h on days 1, 3, and 5, was associated with a lower relapse rate and improved OS compared to intermediate-and low-dose regimens. However, there was no subgroup analysis according to cytogenetics in the original trial [29]. A follow-up study evaluated the outcomes of high-dose cytarabine in patients stratified into three cytogenetic groups, based on the presence of core-binding factor rearrangements, a normal karyotype, and other cytogenetic abnormalities. In the modern era, intermediate-risk AML would constitute parts of the two latter groups. The 5-year relapse-free survival with high-dose cytarabine was 78%, 40%, and 21% in these three groups, respectively, and was superior to results obtained with the intermediate-and low-dose cytarabine regimens (p < 0.05) [30].
Since most studies have focused on stratifying patients to various first-line therapies, little is known about the outcomes of different consolidation regimens in the intermediaterisk AML group. As mentioned, patients who remain in remission may be considered for HSCT as a potentially curative option, when suitable for the procedure. However, randomized trials investigating the benefit of transplant in intermediate-risk AML are lacking. Studies investigating the benefit of allogeneic and autologous HSCT have used donor availability as a criterion for transplant [31,32]; hence, it remains difficult to assess precisely the absolute benefit of this procedure in this group. Moreover, because the definition of the intermediate-risk AML group continues to change along with the advancements in risk stratification and measurable residual disease (MRD)-directed treatment strategies, determining the benefit of consolidative allogeneic HSCT in this population remains challenging and must account for individual patients' characteristics (disease status, MRD, performance status, and donor availability). In a retrospective analysis, AML patients, less than 60 years old, who underwent allogeneic HSCT had a trend toward improved relapsefree survival (RFS), OS, and a lower relapse rate than their non-transplanted counterparts, but the results did not reach statistical significance [33]. HSCT is currently considered an option after remission induction in the intermediate-risk group, per NCCN and ESMO guidelines [27,28]. Nevertheless, we recommend that the treating physician discusses the role of such a procedure with a patient diagnosed with intermediate-risk AML, including the potential benefits and risks. While consolidation with allogeneic HSCT is a potential option in younger patients who are generally fit and less likely to suffer from treatmentrelated morbidity and mortality, enrollment in clinical trials may be a reasonable alternative in cases rendered not medically fit for HSCT.
The role of MRD in treatment stratification is also evolving. MRD refers to the detection of malignant cells by molecular biology or flow cytometry techniques in patients in complete remission, defined as hematologic recovery and an absence of AML blasts by morphological bone marrow examination. Studies have demonstrated the usefulness of monitoring MRD in patients with RUNX1-RUNX1T1 rearrangements and NPM1 mutations [34,35]. Patients with these abnormalities that become undetectable by molecular techniques tend to have better OS and lower cumulative rates of relapse [36]. However, the application of MRD monitoring in clinical settings remains complicated, mainly due to conflicting results with other molecular abnormalities and different sensitivities inherent to the techniques used [37][38][39]. Other challenges include standardizing the approaches by which MRD is analyzed and interpreted, as well as determining the optimal time and specimen source for its measure [40]. Nevertheless, a growing body of evidence suggests a promising role for MRD monitoring in guiding therapy choices in intermediate-risk AML patients. For instance, in the Italian GIMEMA AML1310 trial, in which the NCCN 2009 recommendations were used for risk classification, intermediate-risk patients were allocated into post-consolidation autologous or allogeneic HSCT, in cases with MRD-negative or MRD-positive status using multiparametric flow cytometry (MFC), respectively [41]. No significant difference was seen in the 2-year OS (79% vs. 70%, p = 0.713) or disease-free survival (DFS) (61% vs. 67%, p = 0.773) rates between the MRD-negative and -positive groups. Remarkably, MRD-positive intermediate-risk AML patients treated with allogeneic HSCT had OS and DFS rates similar to favorable-risk patients. The same group confirmed the role of MRD when risk categorization according to the ELN 2017 recommendations was applied to the AML1310 trial [42]. Indeed, no significant difference in OS at two years was observed between intermediate-risk MRD-negative patients who received either autologous or allogeneic HSCT (85.7% vs. 77.8%, p = 0.234), whereas allogeneic HSCT did significantly increase survival compared to autologous HSCT in MRD-positive patients (75% vs. 0%, p = 0.0231).
Maintenance therapy is used to decrease the risk of disease relapse. In an open-label phase III study, patients ≥ 60 years with AML or refractory anemia with excess blasts who achieved remission post induction were randomized to receive azacitidine 50 mg/m 2 subcutaneously days 1-5 every four weeks vs. observation. While patients treated with azacitidine had improved 12-month DFS (64% vs. 42%, p < 0.05), OS was similar (84% vs. 70% at 12 months, p = 0.69). Of note, the study design did not require patients to receive consolidation with cytarabine, and no subgroup analysis based on risk stratification was provided in this trial [43]. In another study of patients ≥ 55 years with de novo or secondary AML, with intermediate-or poor-risk cytogenetics according to 2011 NCCN risk classification, who were not candidates for HSCT, oral azacitidine 300 mg daily for 14 days per 28-day cycle improved OS compared to placebo (24.7 months and 14.8 months, respectively; p < 0.05) [44]. A subgroup analysis showed improved 2-year OS (difference 13.6%, 95% CI: 3.9-23.4) in the intermediate-risk group.
In patients who do not achieve remission with induction or relapse after treatment, the current NCCN and ESMO guidelines recommend targeted therapy or clinical trial enrollment whenever possible. Table 3 summarizes commonly used options in this setting. As with first-line regimens, most trials included intermediate-risk group AML in the study populations.    Abbreviations: CR = complete remission; ORR = overall response rate; EFS = event-free survival; OS = overall survival; HR = hazard ratio; mos = months; AML = acute myeloid leukemia; MDS = myelodysplastic syndromes; MRC-AML = acute myeloid leukemia with myelodysplasia-related changes; ITD = internal tandem duplication; TKD = tyrosine kinase domain; ISCN = International System for Human Cytogenetic Nomenclature; Ara-C = cytarabine; GO = gemtuzumab ozogamicin; FLAG = fludarabine, cytarabine, and filgrastim; Ida = idarubicin; ADE = cytarabine, daunorubicin hydrochloride, and etoposide phosphate; HMA = hypomethylating agents; SC = subcutaneously; BID = twice daily; NA = not applicable or not available.

Clinical Trials
As discussed, meticulous clinical research from bench to bedside has led to a better understanding of the pathophysiology of AML and, subsequently, to the discovery of novel therapeutic agents. Nevertheless, the molecular heterogeneity of AML continues to remain a challenge. Indeed, the significant advances in elucidating the genomics of AML have not yet led to ameliorating outcomes, which remain dismal in relapsed and refractory patients (R/R AML). Moreover, the situation is even hazier when considering the treatment of older and frail (unfit) AML patients.
In the United States, more than 300 clinical trials are currently assessing the safety and efficacy of various therapeutic agents in AML patients. A detailed review of ongoing studies is beyond the scope of this review; however, in Table 4   Abbreviations: R/R, relapsed or refractory; AML, acute myeloid leukemia; ALL, acute lymphoblastic leukemia; MDS, myelodysplastic syndrome; Allo-HCT, allogeneic hematopoietic cell transplantation; HD, high dose.

Targeted Agents and Challenges in Intermediate-Risk AML
The entire schema of ELN risk classification currently used in the clinic is derived from the experience of patients treated with intensive chemotherapy. Over the past few years, with the advent of many new drugs for AML, including venetoclax combinations and a molecularly defined class of targeted therapies, IDH1/2 or FLT3 inhibitors, it is uncertain if this classification would still apply [59].

FLT3 Inhibitors
Mutations in the FLT3 gene are among the most common mutations in intermediaterisk AML, most commonly associated with a normal karyotype [60]. These mutations lead to constitutive activation of the FLT3 receptor tyrosine kinase and multiple downstream signaling cascades involved with AML cell survival, proliferation, and differentiation. In the ELN 2017, the allelic burden and the pattern of co-mutations (in particular in NPM1 and DNMT3A) modulated the prognostic significance of the more common and prognostic FLT3 mutation, FLT3-ITD [7,61]. For instance, FLT3-ITD with a high allelic ratio (>0.5) conferred adverse prognosis; however, the simultaneous presence of an NPM1 mutation made it instead recognized as intermediate-risk [4]. The new 2022 revision abated this differentiation by assigning all FLT3 mutants to the intermediate-risk category [6]. The treatment paradigm for FLT3-mutated AML patients has been transformed by the advent of small molecule tyrosine kinase inhibitors (midostaurin, gilteritinib) for both newly diagnosed and R/R AML [54,62]. Furthermore, the combination of azacitidine and venetoclax has changed treatment outcomes for adults who are not candidates for intensive induction chemotherapy [60,63]. Response rates of FLT3-mutated AML patients to the azacitidine/venetoclax combination are comparable to those of patients with wildtype FLT3. In light of response improvements from doublets, the need to study rational combinations of triplets or sequential therapies to continue improving the outcomes of patients with AML with FLT3-ITD is a current topic of research.

IDH1/2 Inhibitors
Characterization of the unique mechanism of leukemogenesis resulting from IDH1 and IDH2 mutations in AML led to the development of targeted oral therapies. This represents a remarkable bench-to-bedside journey. IDH1/2 mutations are present in~20% of AML cases (IDH1:~5%-8% and IDH2: 9%-12% [61]). These mutations frequently occur in normal-karyotype AML, representing a common subset of intermediate-risk diseases. First-in-class, oral, selective inhibitors of mutant IDH1 (ivosidenib) [64] and IDH2 (enasidenib) [65] are FDA-approved for the management of R/R AML, based on phase 1 clinical trials showing overall response rates of 42% and 39%, respectively, along with manageable toxicity profiles. Despite the initial success, several questions were raised based on the preliminary results of the recent randomized trials. The phase 2 study comparing azacitidine with or without enasidenib as a first-line treatment for AML patients who are unfit for intensive chemotherapy showed improved response rates but similar OS as the combination therapy [66]. Similarly, the phase 3 IDHENTIFY study enrolling IDH2-mutant R/R AML patients (after the failure of two or three lines of therapy) reported no survival benefit with enasidenib compared with standard of care [67]. These two studies highlight the challenges in deciding the optimal approach to using targeted therapies in AML and the need for randomized studies. Besides, a recent phase 3 trial evaluating azacitidine with or without ivosidenib in IDH1-mutant newly diagnosed AML unfit for intensive treatment showed a 12-month event-free survival rate significantly higher in the experimental group, along with a benefit in OS (24.0 vs. 7.9 months; p = 0.001), opening new therapeutic opportunities in such a setting. [58] Particularly, the use of ivosidentib plus azacitidine enabled the clearance of detectable IDH1-mutated cells, and less transfusion support was needed, with safety signals similar in the two groups if excepting the higher rate of differentiation syndrome in the experimental arm.

Others
CA-4948 is a potent interleukin-1 receptor-associated kinase 4 (IRAK-4)/FLT3 inhibitor, with potential for anti-tumor activity in AML, that is currently in the early phases of therapeutic development (NCT04278768). In adult patients with FLT3-mutated R/R AML, combinations with experimental oral agents such as TL-895 (tyrosine kinase inhibitor) and KRT-232 (MDM2 inhibitor) are being evaluated for safety and efficacy in an early phase clinical trial (NCT04669067). Minnelide is a potent heat shock protein 70 inhibitor that showed potent preclinical anti-tumor activity in AML [68]. A phase 1 study (NCT03760523) is ongoing to investigate this novel agent in patients with R/R-AML who are ineligible for intensive chemotherapy.

Challenges
Many challenges remain unsolved in intermediate-risk AML. A first challenge is to define the precise scenarios for using the plethora of novel targeted drugs [69]. The ELN criteria, including the latest revision, are based on data derived from intensively treated patients, thereby leaving unanswered the question if current schemes may still be valid in patients treated with newer agents [6]. This is particularly true for FLT3-positive cases, now representing a good fraction of the intermediate-risk group according to the latest classification, especially in light of the availability of specific targeted treatment options. A second challenge is represented by AML trial designs, which beg the question of what is the best setting to test these novel compounds. Indeed, since the intermediate-risk group is the most heterogeneous category, issues of clonal evolution and genomic and/or epigenomic complexity, after receiving standard of care therapies, and the emergence of subclones with distinct resistance mechanisms are only some of the challenges to be faced. A third challenge is the incorporation of novel agents and existing targeted therapies in the frontline regimens, which are likely to increase the proportion of patients achieving and maintaining CR for longer periods [70]. As a consequence, and with deeper responses resulting from the use of tailored therapeutic approaches in the induction schemes, the definition of the need for consolidation/maintenance regimens and HSCT will urge further refinements. While guiding treatment strategies by MRD-based approaches could answer some of these questions, individual patients' characteristics and inherent subgroup genomic features (e.g., disease subclassifications exemplified in the recent WHO revision and other classifications [71,72]) may represent other difficult obstacles to overcome.

Immunotherapies in AML
Over the last decade, immunotherapies have shown encouraging anti-tumor activity across most cancer types and may offer a therapeutic opportunity also in intermediate-risk AML, in a way that is mutation-agnostic. Immune-based therapeutics, including monoclonal antibodies, immune checkpoint inhibitors (ICIs), bispecific T cell engagers (BiTE) antibodies, and chimeric antigen receptor T-cell therapy (CAR-T) are being extensively investigated to improve the outcome of AML patients.

The Role of Immune Checkpoint Inhibitors in AML
Immune checkpoints, namely cytotoxic T lymphocyte antigen 4 (CTLA-4), programmed death-1 (PD-1), and programmed death-ligand 1 (PD-L1), play a crucial role in regulating T cell immune response to the tumor cells [73]. ICIs, such as ipilimumab, nivolumab, pembrolizumab, and atezolizumab, have revolutionized the treatment landscape in various solid tumors. ICIs promote anti-tumor activity by enhancing T cell activation and preventing immune escape of tumor cells [74]. The role of ICIs in the management of AML is evolving, and various trials are ongoing to evaluate its efficacy. High expression of PD-1, PD-L1, and T-cell immunoglobulin mucin domain 3 (TIM3) and co-expression of PD-1/CTLA-4, PD-1/PD-L1, PD-1/LAG-3, and PD-L2/CTLA-4 are associated with poor prognosis in AML, particularly in patients with FLT3, RUNX1, and TET2 mutations [75][76][77]. High PD-1, PD-L1, and PD-L2 expression could potentially contribute to resistance to treatment with azacytidine [76]. However, a recent randomized phase II clinical trial of Azacitidine in combination with PD-L1 inhibitor (Durvalumab) in older patients (≥65 years) with AML failed to show clinical improvement, compared to azacitidine alone [78]. Of note is that in this study, the response to treatment was not associated with PD-L1 expression, DNA methylation, and mutational status. In a phase I/Ib clinical trial, the efficacy, and safety of ipilimumab were evaluated in patients with relapsed hematologic malignancies after allogeneic hematopoietic stem-cell transplantation (Allo-HSCT) [79]. Out of 28 patients enrolled, 12 were relapsed AML, and no response was observed in patients who received 3 mg/kg of ipilimumab. However, dose escalation to 10mg/kg demonstrated anti-tumor activity; CR was noted in 5 out of 12 relapsed AML patients (4 with extramedullary leukemia and 1 with secondary AML), with a durable response of more than 1 year in 3 patients. In another phase I clinical trial, PD-1 blockade (Nivolumab) was evaluated in patients with hematologic malignancies after Allo-HSCT [80]. Out of 28 patients enrolled in the clinical trial, 10 were AML. Only minimal response was observed in AML patients, and immune-related adverse events such as graft vs. host disease occurred even with lower doses of Nivolumab.
Ravandi et al. evaluated nivolumab in combination with Idarubicin and cytarabine in newly diagnosed AML patients. In this single-arm, phase 2 clinical trial, 44 patients were enrolled, of whom 17 (39%) were intermediate-risk AML based on ELN 2017 classification [77]. The median overall survival was 18.54 months. Overall, 19 patients had a response and subsequently underwent Allo-HSCT. Of note, the authors reported higher CD4+ T-cell co-expression of PD-1/TIM3 and PD-1/LAG-3 in non-responders compared to responders.

Monoclonal Antibodies and CAR-T Cell Therapy in AML
CD33 is highly expressed on AML cells, and unconjugated and conjugated CD33 monoclonal antibodies are in the early phase of drug development [81]. Current clinical trials targeting CD33 with GO (NCT03672539, NCT04207190, NCT03839446) are outlined in Table 4. For instance, lintuzumab Ac225, a radiolabeled anti-CD33 antibody, combined with chemotherapy, demonstrated a clinically acceptable safety profile [82]. A phase I/II study (NCT03441048) is underway to assess the safety and efficacy of this novel therapeutic agent in R/R AML. Flotetuzumab is a humanized bispecific antibody-based molecule binding to CD3 and CD123. Results from a phase 1/2 study (NCT02152956) demonstrated acceptable safety and evidence of anti-tumor activity in R/R AML patients [83]. A phase 2 study is ongoing to evaluate its efficacy in AML that has relapsed following allogeneic HSCT (NCT04582864).
Another active area of experimental drug development is CAR-T cell therapy [84]. In AML, targets such as CD33, CD123, CLL1, and CD44v6 are the most frequent (NCT04789408, NCT04219163, NCT03190278). Notably, commonly expressed CAR-T targets are also present in normal myeloid cells, leading to potential myelosuppression [81,85]. Newer generations of CAR-T cell products could potentially overcome these limitations. For instance, CD70 was recently identified as a promising target for CAR-T cell therapy in AML because of its selective expression on both leukemic blasts and leukemic stem cells [85], providing a rationale for targeted therapy in AML without adversely affecting normal hematopoiesis [86]. In a recent study, two CD70-CAR T constructs exhibited significant anti-tumor efficacy in vitro and in vivo [87], effectively eliminating AML cells and sparing normal hematopoietic stem cells, thereby avoiding potentially dangerous on-target/offtumor toxicity.

Conclusions
AML is a disorder characterized by a puzzled molecular landscape. As a result, patients are characterized by different outcomes according to their individual genomic makeup, which, together with host factors such as age and performance status, contribute to the diverse therapeutic and survival scenarios. Intermediate-risk cases, as per ELN 2017, and more recently ELN 2022, represent the most heterogeneous group because of the broad features included in the definition of this category (or, better yet, the handful of characteristics used for the inclusion in the other two groups). While new genomic information is generating a more detailed dissection of the disease, as in the recent two classifications of myeloid disorders [71,72], an urgent need for a redefinition of diagnostic and prognostic criteria is vital for both better trial designing and adequate indications for new targeted treatments. Immunotherapies also offer potential opportunities, but this area needs further refinements and additional basic and translational studies in AML in general. Furthermore, the application of modern machine learning approaches with the current availability of genomic information and big data platforms will potentially unravel specific subgroups of patients, further contributing to identifying discrete differences in AML patients, particularly in the intermediate-risk group.
Perhaps, as our understanding of AML pathogenesis and risk-stratification modalities continues to improve, we anticipate that the intermediate-risk group definition will continue to change. In fact, the exponential growth in the targeted therapeutic options utilized in AML continues to alter the prognostic significance of many molecular derangements. Nonetheless, in harmony with the 2022 ELN guidelines, we recommend judicious incorporation of the patient's fitness for the intensive regimens and disease molecular characteristics used as criteria for treatment selection.

Conflicts of Interest:
The authors declare no conflict of interest.