IL-3-zetakine combined with a CD33 costimulatory receptor as a dual CAR approach for safer and selective targeting of AML

Despite substantial progresses in prognostic risk stratification¹ and personalized treatment with novel targeted drugs,² between 60% and 70% of adults with acute myeloid leukemia (AML) still relapse after initial responses and do not survive beyond 5 years.³ The graft-versus-leukemia effect elicited by donor T cells after allogeneic hematopoietic stem cell transplantation (allo-HSCT) is the best postremission treatment to eradicate residual chemoresistant leukemic cells.⁴ However, the majority of patients with AML are not eligible for allo-HSCT because of old age, comorbidities, infectious complications, or refractory disease.⁵ Furthermore, ∼40% of patients with AML relapse even after allo-HSCT⁶ and additional curative treatment options are lacking.

Chimeric antigen receptor (CAR) T cells have dramatically improved the outcome of patients with high-risk B-cell malignancies,⁷ whereas their successful application to other neoplasms, including AML, still remains an unmet need.⁸ A major barrier limiting the safe clinical translation of CAR T-cell therapy in the AML field is the lack of a selective target antigen overexpressed on AML blasts and leukemic stem cells but absent on normal tissues.⁹ CD33 and CD123 are 2 of the most attractive AML targets, because of their overexpression on both leukemic bulk and chemoresistant leukemic stem cells in most patients with AML, especially in NPM1 and/or FLT-3 mutated AML.^10,11 Notably, CD123 overexpression has been associated with a negative prognosis with decreased overall survival and lack of clinical remission. Nevertheless, CD33 expression on normal hematopoietic stem/progenitor cells (HSPCs)¹² and CD123 expression on both HSPCs and endothelial cells¹³ raise concerns for life-threatening toxicities. Dual targeting of AML with CAR T cells directed against both CD33 and C-type lectin-like molecule-1 resulted in prolonged grade 4 pancytopenia in all patients, requiring an allo-HSCT to rescue the induced myeloablation¹⁴ whereas a trial of CD123 CAR T-cell therapy was placed on hold after the occurrence of severe capillary leak syndrome (1 lethal) in 2 patients.¹⁵ 

Several strategies have been proposed to minimize off-target toxicity risks. CAR T cell–mediated myelotoxicity could be overcome by a subsequent allo-HCT, although such approach cannot be adopted in elderly or unfit patients.¹⁶ Infusion of messenger RNA “biodegradable” engineered CD123 CAR T cells has been used but is limited by inadequate efficacy.¹⁷ Alternative approaches include generation of CD33 knockout (KO) HSCs resistant to CD33 CAR T-cell myeloid toxicity,¹⁸ development of switchable CAR platforms,¹⁹ and identification of single or combined AML-restricted antigens.^20-22 In this study, we fully exploited the therapeutic potential of the simultaneous targeting of CD33 and CD123, while limiting toxicity, by adopting and optimizing a dual CAR (DC)–trans-signaling strategy²³ in which efficient T-cell activation was only granted by the simultaneous engagement of both antigens.

The total RNA from 12 samples (triplicates of Ontario Cancer Institute (OCI)-AML3 wild type [WT], CD33KO, CD123KO, and CD33/CD123KO) was sequenced on an Illumina Novaseq 6000 platform, generating 150 base-paired end reads, and then analyzed.²⁴ 

The interleukin 3 (IL-3)-zetakine (IL-3z) (immunoglobulin G1 [IgG1]-Fc spacer and CD3z endodomain) and the CD33 chimeric costimulatory receptor (CCR) (IgG1-Fc spacer and CD28 and 4.1BB endodomains) were cloned in an SFG-retroviral construct (In-Fusion HD Cloning Kit). Then, the dual-CAR vector was cloned using a P2A system. Cytokine-induced killer (CIK) cells were subsequently transduced following a previously described protocol.²⁵ For the nonviral CIK cell modification, the Sleeping Beauty (SB)-pT4 vector and SB100X-transposase were kindly provided by CoImmune, Inc (Durham, NC) and DC CIK cells were generated as previously described.²⁶ 

Cytotoxicity and cytokine detection assays were performed as previously described^26,27 (see supplemental Data).

Time-dependent apoptosis mediated by IL-3z WT CAR and IL-3z mutant CAR CIK cells in endothelial telomerase-immortalized human microvascular endothelium (TIME) cell line was evaluated using Operetta CLS (PerkinElmer). CIK cells were cocultured for 8 hours with the target cells at an effector-to-target ratio (E:T) of 5:1 and images were taken every 5 minutes. Custom-made MatLab-based (Mathworks Inc) software was used for image analysis to extract the populated area of the different dyes.

Peripheral blood (PB)-mobilized CD34⁺ cells were cocultured for 24 hours with or without CAR CIK cells (E:T, 1:1). Afterward, 1000 CD34⁺ cells were sorted on a fluorescence-activated cell sorter Aria II, resuspended in 3 mL MethoCult H4435, and 1 mL duplicates plated in 6-well SmartDish at 5 × 10² cells per well. Plates were incubated at 37°C and 5% CO₂ for 14 days, after which different subtypes of CFUs were counted using an automated StemVision machine (STEMCELL Technologies).

For the OCI-AML3 model, 1 × 10⁶ OCI-AML3 WT or KO clones expressing luciferase were transplanted in NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice. For the KG-1 model, NSG mice received a radiation dose of 0.9 Gy followed 24 hours later by an IV injection of 2.5 × 10⁶ KG-1 cells expressing luciferase. CAR CIK cells were then injected via tail vein at a time and dose provided in the figure legends. For the patient-derived xenograft (PDX) model, 1 × 10⁶ cells from a previously described²⁶ tertiary AML PDX sample were injected IV. Procedures involving animal handling and care were conformed according to protocols approved by the Service center of Preclinical Research of Perugia’s animal house facility and by the Milano-Bicocca University, and were in accordance with national and international law.

CD123 overexpression in AML has already been associated with increased proliferative potential and resistance to apoptosis,²⁸ whereas the implication of CD33 overexpression remains elusive. To further define the functional contribution of these 2 antigens to AML, we analyzed the effects of their CRISPR–CRISPR-associated protein 9 KO (single and double) in the NPM1-mutated OCI-AML3 cell line (Figure 1A; supplemental Tables 1 and 2), as characterized by CD33 and CD123 overexpression^29,30 and thus constituting a suitable model to investigate the effects of their genetic loss. As compared with the unedited parental cell line (OCI-AML3 WT), all the KO clones exhibited impaired proliferation in vitro (Figure 1B). Moreover, in growth competition assays against CD33- and/or CD123-edited clones, OCI-AML3 WT cells rapidly became the predominant population within 9 days of coculture (Figure 1C, P value <.0001 of each KO condition vs OCI-AML3 WT at day 9). Next, we assessed whether this in vitro evidence translated into different kinetics of tumor growth in vivo, exploiting luciferase-expressing genes in all OCI-AML3 clones. Although all the KO clones were still able to engraft in NSG mice (Figure 1D), the overall survival of mice injected with CD123 or CD123/CD33 KO cells was significantly longer than that of mice that had received transplantation with parental OCI-AML3 or CD33 KO cells (Figure 1E, P value <.05). To reveal which pathways linked to CD33 and CD123 overexpression could be involved in leukemic proliferation, we performed unbiased bulk RNA sequencing comparison between OCI-AML3 WT and KO clones. Gene ontology–term enrichment analyses of expressed genes in OCI-AML3 KO clones unveiled a massive downregulation of genes involved in cancer pathways (supplemental Figure 1). Among gene expression profiles (supplemental Tables 3-5), we found a significant upregulation of the cyclin-dependent kinase inhibitor 2A, a tumor suppressor gene fundamental for cell cycle entry,³¹ in all OCI-AML3 KO clones but especially in the CD123 KO ones. Moreover, CD33 KO clones showed an increased expression of major histocompatibility complex–related genes (HLA) and downregulated HOX genes (Figure 1F). Notably, HOX genes are characteristically upregulated in NPM1-mutated AML³² and represent a hallmark of this AML subgroup.³³ Although restricted to a single, yet common, AML subgroup, these findings further support CD123 and CD33 targeting by CAR T cells as a viable approach not only for their overexpression profile but also for the major role played in leukemia biology, which potentially decreases the likelihood of antigen escape mechanisms.

Single targeting of CD123 or CD33 by CAR T cells have raised safety concerns for potential severe on-target/off-tumor effects.^34-36 Therefore, we conceived a DC design with transacting costimulation, in order to reduce off-target toxicities while retaining full functionality against AML. Target cells expressing different levels of CD123 and CD33 were used to test the efficacy and safety profile of DC CIK cells, namely the AML cell line KG-1, bone-marrow (BM) derived-CD34⁺ HSPCs, and the TIME endothelial cell line (Figure 2A-B).

DC CIK cells were generated by retroviral transduction with a first-generation anti-CD123 IL-3-zetakine³⁷ (IL-3z CAR) linked to an anti-CD33 CCR (CD33 CCR) carrying CD28-4-1BB costimulatory moieties, through a 2A-sequence peptide. Single-transduced IL-3z CAR, CD33 CAR with CD28-4-1BB-z, and CD33 CCR CIK cells were included as controls (Figure 2C). In particular, CD33 CCR CIK cells were included to confirm the inability of the CD33 CCR to induce CIK cell effector functions per se (supplemental Figure 2A-D). A high transduction efficiency was achieved in all the conditions with a stoichiometric expression of the 2 receptors in the DC (Figure 2D-E). CD4, CD8, and memory phenotypes were similar in all CIK cell conditions (supplemental Figure 2B). All CAR CIK cells specifically induced in vitro target cell lysis (in short- and long-term cytotoxicity assays) and cytokine production when challenged with THP-1, KG-1, and OCI-AML3 CD123⁺/CD33⁺ AML cell lines (Figure 2F-H, P value <.0001 DC vs not-transduced (NT) CIK cells in THP-1 and KG-1; P value <.001 DC vs NT CIK cells in OCI-AML3). However, although the DC exhibited higher IL-2 production after antigen stimulation (Figure 2H), the benefit of a transacting stimulation has been blunted by the high in vitro activity displayed by the single IL-3z CAR. Therefore, in order to achieve limited T-cell activation upon single IL-3z CAR engagement and thus reducing potential DC CIK cell toxicities against CD123^low healthy cells, we considered generating an IL-3z CAR with reduced binding affinity, to obtain a suboptimal signal 1.

MD simulation is a well-established computational method for investigating and understanding biomolecular behavior in atomic detail.³⁸ Four IL-3 mutations (N18K, E22R, E43N, and F113A) known to have effects in the docking of the ligand to CD123,^39,40 were tested alone or together (Mut4) by MD to predict IL-3 mutant binding affinity compared with IL-3 WT. Initially, the stability of the complexes was determined through 30 ns MD simulations by plotting the backbone (Cα atoms) root mean square deviation during the simulation, without significant differences between mutants and WT (supplemental Figure 3A). Moreover, hydrogen bond (HB) analysis did not show any significant changes (supplemental Figure 3B; supplemental Table 6). Together, these observations indicated that single and combined mutations do not affect the conformation of the complex, suggesting that unspecific binding to other molecules is unlikely to occur. Next, we investigated free binding energy (ΔG_binding) between IL-3Rα and IL-3 in WT complex and mutants of the last 10 ns of the trajectories, to envisage possible changes in binding affinity. Mut4 complex showed a lower calculated ΔG_binding value of approximately −24 Kcal/mol, indicating that the Mut4 is the least stable complex among all the mutants (supplemental Figure 3B). This result was confirmed also in extended simulation of 100 ns (supplemental Figure 3C), highlighting differences between WT and Mut4 systems in the HB occupancy analysis (supplemental Table 7). In particular, WT IL-3 showed more prolonged HBs during the simulation time, whereas the Mut4 complex lost Lys54/Glu43 and Arg234/Glu119 contact (Figure 3A-B). A low ΔG_binding in Mut4 complex was observed compared with WT also in the last 40 ns of the simulation (Figure 3C), in which the 2 systems showed their general stability, exhibiting a reduction of ∼43% (Figure 3D).

Finally, we biologically probed whether the low affinity IL-3z mutant CAR could reduce CIK cell activation upon challenge with the TIME endothelial cells by performing a dynamic live-cell imaging assay assessing microvascular cytotoxicity over time and space. Using caspase-3 as specific marker of early apoptosis, a lower caspase-3 activation was detected on target TIME cells when cocultured with NT and IL-3z mutant CAR-engineered CIK cells as compared with IL-3z WT CAR (Figure 3E-F; supplemental Video). Multiplexed cytokine profiling after long-term coculture showed that KG-1 cells but not TIME cells stimulated IL-3z WT CAR and IL-3z mutant CAR CIK cells to significantly release more Th1/Tc1 (granulocyte-macrophage [GM]-colony-stimulating factor, interferon gamma, tumor necrosis factor α, IL-2) cytokines (supplemental Figure 3D).

IL-3z mutant CAR was then coupled to the CD33 CCR in the newly generated low affinity DC (DC mutant) to test safety and antileukemic efficacy (Figure 4A). As compared with the DC WT, the DC mutant displayed less lytic activity, and therefore superior safety, when cocultured with the low–CD123-expressing TIME endothelial cell line at 2 different E:T ratios (10:1 and 5:1) (Figure 4B, P value <.05 DC WT vs NT; DC mutant vs NT, not significant). This low endothelial toxicity was further sustained by comparable low cytokine levels released by NT and DC mutant CIK cells after long-term cocultures with CD123^low TIME endothelial cell line (Figure 4C). In contrast with cytotoxicity assays, no differences in cytokine production were observed between DC WT and mutant DC, probably because of very low CD123 density on TIME cells. This result is in line with previous studies^27,41 reporting different levels of activation threshold for early cytotoxicity and later functions such as cytokine release. In contrast, in the presence of KG-1, a lower cytokine production (still higher than NT) by low affinity DC CIK cells has been observed, confirming prior findings with low affinity CARs.²⁷ 

Considering the potential severe off-tumor effects on the hematopoietic compartment, we hypothesized that low affinity DC CIK cells would spare HSPCs by 2 distinct but complementary mechanisms: (i) avoiding cell activation against the majority of CD33⁺ hematopoietic cells because of a lack of stimulation through anti-CD33 CCR and (ii) decreasing reactivity toward low expressing CD123⁺ HSPCs because of reduced affinity of the anti-CD123 suboptimal CAR (Figure 4D). In vitro clonogenic efficiency and differentiation capacities of PB–derived healthy CD34⁺ HSPCs were tested after 24-hours coculture assay with different CIK cell conditions with similar CAR expression (supplemental Figure 4), through automated and standardized counting of all hematopoietic colony types after 14 days (Figure 4E). No significant differences have been observed between CD34⁺ HSPCs not exposed or preexposed to DC mutant CIK cells, suggesting that lowering the IL-3z affinity does not hamper the clonogenic capacity of CD34⁺ HSPCs and confirming the hypothesis that the CD33 CCR is not associated with enhanced myelotoxic potential (Figure 4F; DC mutant vs NT; P value, not significant). A significant reduction in clonogenic capacity was observed in IL-3z WT CAR and DC WT conditions, in particular in CFU-GM (Figure 4G, P value <.0001 for IL-3z vs NT; P value <.001 for DC WT vs NT). Moreover, the analysis of CD34⁺CD38⁺ myeloid progenitor subsets confirmed that common myeloid progenitors, granulocyte–monocyte progenitors, and megakaryocyte-erythroid progenitor proportions were less significantly affected by exposure to DC mutant CIK cells as compared with DC WT CIK cells (Figure 4H). Consistent with CD123 expression levels on different subsets of HSPCs,³⁶ only IL-3z WT and DC WT CIK cells significantly reduced absolute counts of CD34⁺CD38⁺ common myeloid progenitors subpopulation (Figure 4I, P value <.0001 DC WT and IL-3z vs NT; P value <.05 DC mutant vs NT; and supplemental Figure 5). In conclusion, these data support that low affinity DC exhibits a lower off-target vascular and hematopoietic toxicity.

Next, we evaluated whether low affinity DC CIK cells could match safety with an efficacy profile comparable with WT DC CIK cells. Phenotypic analysis performed after 21 days of culture showed similar CAR expression levels and percentage of T and natural killer (NK) cell–like populations as well as memory phenotype between all CAR CIK cell conditions (supplemental Figure 6A-B). Low affinity DC CIK cells retained a high in vitro cytotoxicity against CD123⁺/CD33⁺ KG-1 AML cell line in short- and long-term assays (Figure 5A-B, P value <.001 vs NT). DC mutant CIK cells also showed a significant control of primary blast growth in long-term cytotoxicity assays, comparable with DC WT CIK cells (Figure 5C, P value <.001 vs NT, supplemental Table 8).

The antileukemic activity of DC mutant CIK cells was then confirmed in vivo in an early-treatment model of mice engrafted with the KG-1 AML cell line (Figure 5D). In the groups treated with IL-3z constructs, disease growth was delayed by the treatment without relevant impact on survival, whereas in the DC groups there was better disease control in terms of bioluminescence imaging (BLI) signal (Figure 5E-F) and significantly prolonged survival (Figure 5G, P value <.05 vs KG-1 only). This result correlates with the higher in vitro expansion exhibited by DC CIK cells in a long-term coculture assay with KG-1 cells, as compared with IL-3z CAR CIK cells (supplemental Figure 6C-D). The potency of DC mutant CIK cells was further confirmed in a more challenging AML treatment model, in which the treatment started 2 weeks after KG-1 injection (supplemental Figure 7A-D, P value <.05 vs KG-1 only). Together, these findings suggest that a low affinity DC CIK cell targeting approach preserves a high antileukemic activity.

Because both IL-3z CAR and CD33 CCR carry the same spacer/transmembrane domains (IgG1-Fc spacer and a CD28 transmembrane), we aimed to refine the DC design to avoid any unwanted unspecific activation induced by a potential CAR-CCR dimerization. Indeed, we found that DC mutant CIK cells were significantly activated and produced cytokines not only when challenged with the WT KG1 (expressing both CD123 and CD33 antigens) but also in response to CD123- KG1 (supplemental Figure 8). This result is in line with a recent study by Hirabayashi et al⁴² that demonstrated that a single CD3z can mediate DC activation if both CAR constructs share the same spacer and/or transmembrane domain. Therefore, we generated a novel DC system with distinct spacers and transmembrane motifs (CD8 spacer and transmembrane for IL-3z CAR and CH3 spacer with CD28 transmembrane for CD33 CCR). Moreover, considering our extensive experience with the SB transposon system for the nonviral engineering of CIK cells, both at the preclinical and clinical level,⁴³ we successfully cloned the DC construct within a pT4-transposon plasmid,²⁶ reaching up to 50% of DC expression on CIK cells (Figure 6A).

To better investigate the target-specific transactivation of the CIK cells engineered with the optimized DC design, we used dynamic acoustic force measurements to evaluate binding avidity. WT, CD123 KO, and double CD123/CD33 KO OCI-AML3 cell clones were used as targets to check the contribution of the CD33-mediated engagement of the CCR in the synapse formation. We found that optimized DC CIK cells had a lower binding avidity when challenged with CD123-KO OCI-AML3 and a negligible avidity when challenged with double-negative targets, comparable with the binding avidity of NT CIK cells to all targets (Figure 6B-C). In long-term cytotoxic assays against WT and edited KG-1 cells, the newly optimized DC conferred CIK cells a high, specific antileukemic activity against CD123⁺ KG-1 cells (Figure 6D, P value <.001 vs NT in KG-1; P value <.01 vs NT in KG-1 CD123⁺CD33⁻). Notably, a preferential expansion was observed only when DC CIK cells were challenged against WT KG-1, demonstrating the gain in stimulation provided when the CCR is engaged (Figure 6E, P value <.05 vs NT). The newly designed DC also ensured tumor control and superior overall survival of engrafted KG-1 (Figure 6F-H, P value <.05 vs KG-1 only).

Envisaging a clinical translation of the low affinity DC CIK cells with distinct spacer and transmembrane domains, we verified the preserved safety against endothelial TIME cells in long-term cytotoxicity and cytokine production assays; a high antileukemic activity against KG-1 AML cell line and primary blasts was being maintained (Figure 7A-B). Moreover, in order to test DC mutant CIK cells against primary AML cells featuring an antigen density more reflective of a real-world population, we exploited a tertiary PDX mouse model established in our laboratory²⁶ (Figure 7C). At sacrifice (day 35) the treated mice showed low tumor burden in PB and spleen, and tumor burden was almost cleared from the BM (0.062 ± 0.095, n = 4 vs 65.7 ± 6.7, n = 4 of untreated mice, P value <.05), as well as relevant CIK cell infiltration (1.2 ± 1.1 in the BM, 13.2 ± 18.3 in the PB, and 13.7 ± 15.4 in the spleen) (Figure 7E-H). These results support the potential of the proposed strategy of matching efficacy with safety, further encouraging its development for future clinical translation.

In this study, we probe a novel strategy targeting both CD33 and CD123 without the risk of severe on-target/off-tumor toxicities, to improve the outcome of patients with AML. These 2 target antigens were chosen because of their broad expression in AML,^44,45 especially in NPM1-mutated AML,¹⁰ and from the results of our transcriptome analysis on CD33 and/or CD123 KO AML cells. Indeed, in our model of NPM1-mutated AML, we found a close correlation between CD123 and/or CD33 overexpression and downregulation of cyclin dependent kinase inhibitor 2A. Moreover, we uncovered a correlation between CD33 overexpression and HOX genes upregulation, a hallmark of NPM1-mutated AML, and HLA downregulation. The importance of both CD123 and CD33 for AML pathogenesis and self-renewal is also consistent with the observation that loss of these antigens (antigen escape) has not previously been reported after targeted immunotherapies.⁴⁶ Although NPM1-mutated AML has an overall good prognosis without allogeneic transplantation, AML with the cooccurrence of NPM1-mutation and FLT3-ITD or FLT3-ITD (representing 20%-30% of AML) are European LeukemiaNet 2022–defined as intermediate/adverse risk and frequently relapse after standard chemotherapy.⁴⁷ Because CD123 and CD33 overexpression is characteristic of NPM1 and/or FLT3/ITD AML,^10,30,48 we believe that the proposed DC mutant CIK cell strategy could be useful not only in patients with NPM1-mutated AML relapsing after chemoimmunotherapies and who are not eligible for allo-HSCT but also in the FLT3-ITD AML subset (with or without NPM1 mutations). Moreover, targeting CD123 could have a direct effect on all high-risk AMLs, as many studies have reported a direct correlation between CD123 levels and the number of leukemic blasts at the diagnosis,^49,50 suggesting a role of CD123 in leukemic blast proliferation/survival. Notably, CD123 overexpression has been correlated to poor prognosis, resistance to apoptosis, and lower survival after standard treatments.⁵¹ A similar finding was recently reported in a pediatric setting from a prospective analysis on more than 1000 BM specimens, showing that high CD123 expression correlated with higher prevalence of KMT2A and FLT3-ITD mutations, a higher relapse risk, and lower survival.⁵² 

The simultaneous targeting of these antigens with T cells expressing 2 independent CARs for CD123 and CD33 have already allowed the elimination of either single- or double-positive target cells.⁵³ However, CD33 and CD123 are also expressed on normal cells, including HSPCs and endothelial cells.⁵⁴ Thus, the translation from preclinical studies to the clinic of CAR T-cell therapy directed to these antigens is associated with a serious risk of toxicity, which can be even higher when compared with a single targeting approach.

In order to fully exploit the therapeutic potential of simultaneously targeting these 2 antigens while limiting the concern for life-threatening toxicities, we generated and optimized a DC trans-signaling strategy in which efficient T-cell activation is only granted by the simultaneous engagement of both CD123 and CD33 antigens. CIK cells, that is, effector T lymphocytes with NK features acquired during ex vivo expansion,⁵⁵ were genetically modified to coexpress a first-generation anti-CD123 CAR, providing the activation signal 1 (CD3ζ), and an anti-CD33 CCR delivering the costimulatory signal 2 (4-1BB and CD28 double costimulation). In the AML setting, an analogous approach has been preclinically described to enhance selectivity toward CD13 and TIM-3 double-positive AML cells.⁵⁶ However, a moderate, yet significant, level of toxicity toward healthy CD13 single-positive cells, including HSCs, could not be avoided. Similarly, in our first prototype of a DC, a significant level of on-target/off-tumor toxicity toward endothelium and myeloid progenitors was still present after a single CD123 engagement in vitro. This prompted us to generate an anti-CD123 CAR with a reduced binding affinity to minimize the off-tumor CAR-induced activation.^27,57 Instead of an anti-CD123 scFv-based CAR, we exploited the IL-3 sequence as the natural CD123 ligand to build an IL-3z. This allowed us to screen in silico potential low affinity IL-3 mutants by replacing already known IL-3 epitopes predicted to be crucial for CD123 binding, with the intent to minimize toxicities toward CD123⁺ normal cells. Four different point mutations were validated by MD simulation and trajectory analysis. The IL-3 mutant carrying 4 amino acid substitutions displayed the lowest ΔG_binding and was selected for subsequent in vitro and in vivo characterization. The selected IL-3z mutant significantly reduced CIK cell activation after encountering both CD123-low endothelial cells and myeloid progenitors, while allowing efficient AML targeting both in vitro and in vivo, when coupled to a CD33 CCR domain.

Decreasing signal 1 strength did not jeopardize the reactivity toward CD123⁺/CD33⁺ leukemic cells, as efficient the dual-CAR CIK cell activation threshold was reached only after the simultaneous engagement of both receptors, whereas a weaker or null CIK cell reactivity was observed in response to CD123⁺ endothelium and HSPCs, and CD33⁺ hematopoietic cells. This trans-signaling model was further improved by optimizing the hinge/transmembrane regions to avoid any possible homodimerization between the CAR and the CCR bearing the same CD28 transmembrane domains, as recently described.⁴² Different transmembrane domains were therefore introduced (CD8tm for anti-CD123 CAR and CD28tm for anti-CD33 CCR). This tactic allowed the avoidance of unwanted DC CIK cell activation against CD33 single-positive target cells. Binding avidity measurements showed how fine tuning of the DC system with distinct transmembrane motifs avoided CIK cell recognition of CD123⁻CD33⁺ cells, without compromising the binding and the subsequent full activation after double-antigen engagement on AML cells.

Envisaging a clinical translation of this immunotherapeutic strategy using a manufacturing technology more cost effective but still efficient in gene transfection,⁵⁸ we transferred our DC from the retroviral vector to the SB transposon system, exploiting the already optimized nonviral CAR CIK cell platform we had developed for patients with B-cell acute lymphoblastic leukemia.⁴³ The bicistronic construct was successfully expressed by CIK cells, confirming the great payload capacity of the transposon system. The ultimate nonviral-engineered low affinity DC CIK cell product preserved a beneficial safety profile and antileukemic activity in vitro. Furthermore, in a PDX established by our laboratory, expressing CD123 and CD33 antigens at levels more closely resembling physiological levels, high in vivo tumor control and robust CIK cell expansion has been observed.

Using engineered CIK cells as an allogeneic source for the DC approach would also overcome anti-AML CAR manufacture issues. In fact, the intensive induction/salvage chemotherapy schedules in AML can dramatically impair the quality and quantity of autologous T or NK cells at leukapheresis.⁵⁹ Moreover, aggressiveness of relapsed/refractory (r/r) AML requires fast manufacturing to allow CAR T-cell infusion before progression or significant infectious complications. Among 10 patients enrolled in a recent phase 1 trial, evaluating autologous CD33 CAR T cells in r/r AML, apheresis could be performed in 8 patients, product release specification was met in 4, and infusion completed only in 3, who died from disease progression.⁶⁰ Conversely, our first-in-human clinical trial exploring allogenic CD19 CAR CIK cells in r/r B-cell acute lymphoblastic leukemia after allo-HSCT showed feasibility to generate a CAR product from donor PB for all patients enrolled, with no reported graft-versus-host disease or immune effector cell–associated neurotoxicity syndrome, and only occasional grade 2 cytokine release syndrome.⁴³ Such an impressive safety profile and encouraging activity of CAR CIK cells is particularly attractive in the AML setting, in which older age and comorbidities hamper adoptive cell therapy options because of the risk of severe cytokine release syndrome or immune effector cell–associated neurotoxicity syndrome.

In conclusion, we provide a rational approach to engineering a DC with trans-acting costimulation that provides advantages over single-targeting CARs, such as improved efficacy and high specificity for CD123 and CD33 recognition on leukemic cells while substantially reducing the risk of life-threatening toxicities.

The authors thank the parent committees Quelli Che…Con Luca Onlus, Comitato Maria Letizia Verga, Amici di Duccio, and Stefano Verri for their generous and constant support and CoImmune, Inc. for providing the SB plasmids. The authors thank Giusi Melita for technical support in the in vivo studies, and Simone Naso, Roberta Rossi, and Serena Donnini for technical support in the in vitro studies.

The work was supported by grants Ricerca Finalizzata-Giovani Ricercatori (GR-2016- 02363491), AIRC 5x1000 “Immunity in Cancer Spreading and Metastasis” (#21147), the Ministero della Salute Research project on CAR T cells for hematologic malignancies and solid tumors conducted under the aegis of Alliance Against Cancer network, the European Research Council (ERC Adv grant 2016 #740230 [B.F. and E.R.C.], Cons grant 2016 #725725 [M.P.M.]), and the Italian Association for Cancer Research (AIRC grant #23604 [B.F.]). The authors thank Quelli Che…Con Luca Onlus, which funded the project and Comitato per la vita “Daniele Chianelli.”

Contribution: V.M.P. and M.C.R. were responsible for study conceptualization, methodology, data curation, formal analysis, validation, investigation, and visualization, writing of original draft, and review and editing the final manuscript; I.P. was responsible for data curation, validation, formal analysis, investigation, and methodology; S.G. was responsible for data curation, formal statistical analysis, and methodology; G.A. was responsible for data curation, validation, and methodology; G.P. performed data curation, validation, formal analysis, investigation, and visualization; V.C. was responsible for data curation, validation, formal analysis, investigation, visualization, and methodology; A.M. was responsible for formal analysis, investigation, visualization, and methodology; M. Sabino was responsible for formal analysis, visualization, and methodology; V.T. and G.S. were responsible for formal analysis, visualization, and methodology; F. Mezzasoma and F. Morena performed data curation, validation, formal analysis, and investigation; S.M. reviewed and edited the manuscript; D.S. performed data curation, validation, and formal analysis; J.F.A. was responsible for data curation, validation, methodology, and formal analysis; B.W. was responsible for data curation, validation, methodology, and formal analysis; M. Serafini reviewed and edited the manuscript; M.P.M. and B.F. provided supervision and resources and edited the manuscript; A.B. conceptualized the study, provided resources and supervision, acquired funding, and reviewed and edited the manuscript; and S.T. conceptualized the study, was responsible for data curation, formal analysis, validation, investigation, visualization, methodology, acquired funding, and reviewed and edited the manuscript.

Conflict-of-interest disclosure: Fondazione Tettamanti submitted a PCT application on 6 November 2015 (PCT/EP2015/075980), “Improved method for the generation of genetically modified cells,” Biondi, A., Biagi, E., Magnani, C.F., Tettamanti, S. The technology was licensed to CoImmune, Inc. for further development. M.P.M. reports honoraria from Rasna Therapeutics, Inc. for scientific advisor activities and serves as consultant for scientific advisory boards of AbbVie, Amgen, Celgene, Janssen, Novartis, Pfizer, and Jazz Pharmaceuticals. B.F. licensed a patent on NPM1 mutants (#102004901256449) and declares honoraria from Rasna Therapeutics, Inc. for scientific advisor activities. The remaining authors declare no competing financial interests.

Correspondence: Andrea Biondi, Pediatrics and Tettamanti Center, Fondazione IRCCS San Gerardo dei Tintori; School of Medicine and Surgery, University of Milano-Bicocca, Monza, Italy; e-mail: abiondi.unimib@gmail.com.