The anti-CD123 scFV derived from the clinically tested CSL362 MoAb was generated and cloned into the pCCL lentiviral-based second-generation CAR backbone containing a human CD8 transmembrane domain, a human costimulatory domain (either 41BB or CD28), CD3z endodomain, and a T2A–green fluorescence protein (GFP) cassette. The pCCL vector expressing GFP alone (MOCK vector) was used as a control. Chimeric antigen receptor (CAR)-expressing viral particles pseudotyped with vesicular stomatitis virus-G (VSV-G) were generated using HEK 293T cells with a standard polyethylenimine (PEI) transfection protocol. For each production, plasmid transfection was carried out using a 3:1 PEI:DNA ratio using 16 µg transfer vector, 16 µg of pSPAX2, and 8 µg VSV-G per plate, and viral particles were concentrated by ultracentrifugation as previously described.32 Peripheral blood mononuclear cells (PBMCs) were isolated from buffy coats from healthy volunteers by Ficoll-Hypaque gradient centrifugation. Buffy coats were obtained from the Barcelona Blood and Tissue Bank (BST) on institutional review board (IRB) approval (HCB/2018/0030). T-cells were activated by plate-bound OKT3 and anti-CD28 antibodies (BD Biosciences) for 2 days in the presence of interleukin (IL)-7 and IL-15 (10 ng/mL, Mitenyi Biotec).33 34 Surface expression of CAR123 was traced by fluorescence-activated cell sorting (FACS). CAR detection was confirmed by GFP expression and by using an AffiniPure F(ab')₂ Fragment Goat Anti-Human IgG (H+L) (Jackson ImmunoResearch). Activation and subsetting of lentivirally transduced T-cells were confirmed by surface staining with CD25/CD69 (data no shown) and CD3/CD4/CD8, respectively.

Diagnostic immunophenotyping data for the most commonly expressed antigens in AML (CD123, CD33, CD13, CD34, CD15, c-kit, and CD66) were obtained for 97 patients diagnosed at local hospitals: Germans Trias i Pujol (Barcelona, Spain), Hospital Clínico (Madrid, Spain), Hôpital Armand Trousseau (Paris, France) and Santa Creu i San Pau (Barcelona, Spain). CD123 expression was also compared in diagnostic-relapsed paired samples (n=68 patients) and in paired bulk leukemia–leukemia stem cells (LSCs) (n=37 patients).35 Cell lines were stained with CD123-APC, CD33-BV-421, CD14-PerCP-Cy5.5 and CD19-APC. The expression of CD123 and CD33 antigens was prospectively compared in CD34+ HSPCs derived from healthy cord blood (CB, n=22), mobilized peripheral blood (PB, n=10) and diagnostic primary AML samples (n=24). For HSPC subsetting, CD34+ cells were stained with CD34-PE or CD34-PE-Cy7, CD133-PE, CD19-FITC, CD90-APC, CD13-PE-Cy7 and CD71-APC-Cy7, which allow for the identification and quantification of immature HSCs (CD34++CD133+CD90+), myeloid progenitors (CD34+CD13++CD71−/low), erythroid progenitors (CD34+CD71++CD13 low), and B-cell progenitors (CD34+CD19+CD71−CD13−). Isotype-matched, non-reactive fluorochrome-conjugated MoAbs were always used as a fluorescence reference. All antibodies were purchased from Beckton Dickinson. Cells were incubated with MoAbs (30 min at 4°C in the darkness), then washed in phosphate-buffered saline (PBS) and analyzed in a FACSCanto-II flow cytometer equipped with FACSDiva software (Becton Dickinson).36–38 Determination of antigen density for CD33 and CD123 was performed using BDQuantibrite-PE (Becton Dickinson) according to the manufacturer’s instructions.

The cell lines THP-1, MOLM-13 and 697 were purchased from DSMZ (Germany) and expanded according to DSMZ recommendations. Primary AMLs and healthy CD34+ cells were obtained from the aforementioned hospitals and the BST, respectively (IRB approval: HCB/2018/0030). Target cells were incubated with CAR123 or MOCK T-cells at different effector:target (E:T) ratios for the indicated time periods. CART-mediated cytotoxicity was determined by analyzing the residual alive (7-AAD-) target cells at each time point and E:T ratio. For absolute cell counting, Trucount absolute count beads (Becton Dickinson) were used. Furthermore, FACS-sorted CD3+ mature T-cells from the bone marrow (BM) of CD123+ patients with AML were activated, transduced with CD123 CAR and tested against their autologous-matched CD123+ AML blasts. Target cells (1×10⁵) were used for all cytotoxicity assays unless stated otherwise. Table 1 shows the clinical–biological features of the CD123+ AML samples used for in vitro experiments. The production of the proinflammatory cytokines IL-2, tumor necrosis factor alpha (TNF-α) and interferon gamma (IFN-γ) was assessed by ELISA (Human ELISA SET, BD Biosciences) using in vitro supernatants harvested at 16 hours post-T-cell exposure, and sera were collected from mice 10 days after CART infusion.

CB-derived CD34+ cells were exposed for 24 hours to either CD123 CARTs or MOCK T-cells (E:T 1:1) and then plated (2×10³) onto serum-free methylcellulose H4435 (Stem Cell Technologies). CFUs were then counted and scored after 12–14 days following standard procedures.

Non-obese diabetic-Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice (8–12 weeks old, Jackson Laboratory) were bred and housed under pathogen-free conditions in the animal facility of the Barcelona Biomedical Research Park. In experiments addressing CAR123 efficacy, mice were intravenously transplanted with 0.25×10⁶ Luc-mCherry-expressing patient-derived xenograft AML cells (PDX-579)39 5 days before intravenous infusion of 3×10⁶ of either 41BB-CD123 or CD28-CD123 CARTs derived from healthy PBMCs. Tumor burden was monitored at the indicated time points by bioluminescence (BLI) using the Xenogen in vivo imaging system (IVIS) 50 Imaging System (Perkin Elmer).32 In experiments addressing the myeloablative effect of CAR123, CD34+ HSPCs (0.1×10⁶) were intra-BM transplanted in sublethally irradiated (2 Gy) NSG mice, followed by intravenous infusion of 3×10⁶ of 41BB-CD123, CD28-CD123 CARTs or MOCK T-cells either 1 day or 6 weeks after CD34+ transplantation. BM and PB were FACS-analyzed for human chimerism at sacrifice. Cells were stained with anti-HLA.ABC-PE and CD45-BV450. Engrafted mice were assessed for multilineage engraftment using anti-CD123-APC for myeloid cells, anti-CD19-BV421 for lymphoid cells, and anti-CD34-PE.Cy7 for immature cells. Human absolute engraftment in PB and BM was quantified using BD Trucount tubes according to manufacturer’s instructions.

For comparison of CD123 expression between paired ‘diagnostic-relapse’ and ‘bulk leukemic cells-LSC’, the Mann-Whitney U test was used. For differences in antigen density and engraftment among groups, a one-way analysis of variance test was used. For the remaining comparisons, Student’s t-test was used. All p values were considered statistically significant at <0.05 (*).

We first analyzed by FACS the expression levels of the most common diagnostic myeloid markers in a cohort of 97 patients with AML at presentation. We found that CD123 was the most common and homogeneously expressed antigen (86.4%±26.8 of AML blasts) followed by CD33 (77.4%±32.1) (figure 1A). Important, in 82% of the patients with AML analyzed, >80% of the blasts were CD123+, while only 66% of the patients showed positivity for CD33 in >80% of the blasts (figure 1A). A target antigen for immunotherapy in AML should ideally be absent in HSPCs. CD123 and CD33 are both partially expressed in healthy CD34+ HSPCs,22 23 so we next quantified the density (molecules/cell) of both antigens in fresh primary AML blasts (n=24), healthy CB-derived (n=22) and healthy mobilized PB-derived CD34+ HSPCs (n=10). Of note, 67% (16/24) of patients with AML displayed levels of CD123 significantly higher than those found in both CB-derived and PB-derived CD34+ HSPCs, while only 41% (10/24) of patients with AML displayed levels of CD33 that segregate them from CB-derived and PB-derived CD34+ HSPCs (figure 1B). This suggests that CD123 represents, a priori, a less myeloablative target than CD33. Of note, analysis of paired diagnostic (DX)-relapse (RX) AML samples revealed that CD123 expression is maintained at relapse, and in AML-LSC (identified as CD34+CD38−)35 (figure 1C), reinforcing CD123 as a bona fide immunotarget for R/R AML.

We next designed second-generation 41BB-based and CD28-based CD123CARs coupled in-frame with GFP through a T2A sequence (figure 1D and online supplementary figure S1A). The expression of both 41BB-CD123 and CD28-CD123 CAR in T-cells was confirmed through codetection of scFv and GFP (figure 1E and online supplementary figure S1B) and did not affect the CD4:CD8 ratio (figure 1F). Importantly, activated (CD69+CD25+) T-cells continuously expanded ~50-fold over a 10-day period, similar to MOCK T-cells (figure 1G), demonstrating that redirecting T-cells against CD123 does not hamper T-cell expansion.

We then tested the functionality of our 41BB-CD123 and CD28-CD123 CARs in vitro and in vivo (figure 2 and online supplementary figure S1, S2). In vitro, both 41BB-CD123 (figure 2A) and CD28-CD123 (online supplementary figure S1C) CARTs, but not MOCK T-cells, specifically eliminated the CD123+ AML cell lines THP1 and MOLM13 in an E:T ratio-dependent manner (online supplementary figure S2) while sparing the CD123− B-ALL cell line 697. In fact, CD123+ AML cells barely survived exposure to CD123 CARTs in a 48-hour absolute number assay at a 1:1 E:T ratio (figure 2B and online supplementary figure S1C). We then examined in an autologous setting whether CD3+ T-cells deriving from patients with AML can be isolated, modified to express CD123 CAR, expanded and used as cytotoxic effector cells (figure 2C). Patient-derived CD123 CARTs were successfully generated from magnetic-activated cell sorting (MACS)-sorted CD3+ T-cells (>95% purity) and specifically eliminated autologous patient-matched CD123+ AML blasts (figure 2D). Important, both CD123 CARTs produced high levels of the proinflammatory cytokines IL-2, TNF-α, and IFN-γ on coculture with both AML cell lines (figure 2E and online supplementary figure S1D) and primary blasts (figure 2F), confirming their robust cytotoxicity.

We next compared the cytotoxic activity of 41BB-CD123 and CD28-CD123 CARTs in vivo using Luc-expressing CD123+ AML xenograft (figure 2G). NSG mice were transplanted with 0.25×10⁶ Luc-expressing AML PDX cells 5 days prior to intravenous infusion of 3×10⁶ 41BB-CD123 or CD28-CD123 CARTs, and leukemia establishment was followed up weekly by BLI until disease signs were evident (figure 2H). While control mice increasingly showed aggressive disease and disseminated leukemia, CD123 CART-treated mice showed extensive disease control across the experiment, regardless of the costimulation domain used (figure 2H1). Of note, T-cells persisted in PB and BM at sacrifice, although at higher levels in 41BB-CD123 CART-treated mice, in line with the reported longer persistence/effector function of 41BB-costimulated CARTs (figure 2J).40 Similarly, both CD123 CARTs, especially 41BB-costimulated CD123 CARTs produced high levels of the IFN-γ in vivo (figure 2K). Collectively, both 41BB-CD123 and CD28-CD123 CARTs have similarly potent and specific antileukemic activity against AML cells in vitro and in vivo.

There is controversy on whether CD123-redirected T-cells are myeloablative. To prospectively assess the potential myelotoxicity of CD123 CARTs, we first addressed in vitro whether exposure to CD123 CARTs hampers the viability and clonogenic capacity of CD34+ HSPCs (figure 3A). As compared with MOCK T-cells, both CD123 CARTs induced a massive reduction in CD34+ cell counts in a 72 hours (E:T ratio 2:1) assay (figure 3B). Similarly, CD34+ HSPCs pre-exposed to either CD123 CARTs (E:T ratio 1:1) for only 24 hours showed 50%–80% reduction in their clonogenic capacity (figure 3D).

We next assessed the myeloablative potential of both 41BB-CD123 and CD28-CD123 CARTs in vivo using xenograft models of human hematopoietic reconstitution. In an initial set of experiments, sublethally irradiated NSG mice were reconstituted with 0.1×10⁶ CD34+ HSPCs, and 6 weeks later, when human multilineage engraftment was established, mice received 3×10⁶ 41BB-CD123, CD28-CD123 CARTs or MOCK T-cells (figure 3E). Human engraftment was biweekly analyzed in PB (figure 3F) and BM (figure 3G) over 6 weeks. MOCK T-cell-treated mice consistently showed increased myeloid (HLA-ABC +CD45+CD123+CD33+), B-lymphoid (HLA-ABC+CD45+CD123 CD19+) and immature (HLA-ABC+CD45+CD34+) hematopoietic engraftment than that observed on the day of CART infusion. In contrast, both 41BB-CD123 and CD28-CD123 CART-treated mice showed an impaired multilineage engraftment in both PB and BM (figure 3F). However, in this xenograft model of existing hematopoiesis, CD28-based CD123 CARTs proved less myeloablative than 41BB-CD123 CARTs.

Next, we assessed the capacity of both CD123 CARTs in preventing de novo establishment of normal hematopoiesis by transplanting sublethally irradiated NSG mice with CD34+ HSPC and either 41BB-CD123, CD28-CD123 CARTs or MOCK T-cells 1 day after (figure 3H). Long-term multilineage human engraftment was found in both PB and BM in MOCK T-cell-treated mice; however, human hematopoiesis was barely reconstituted in both 41BB-CD123 and CD28-CD123 CART-treated mice (figure 3I), indicating that both 41BB-CD123 and CD28-CD123 CARTs prevent healthy hematopoietic reconstitution.

Finally, to further characterize the myeloablative effects of CD123 CARTs, we exposed total CD34+ HSPCs to either CD123 CARTs or MOCK T-cells for 48 hours at 1:1 E:T ratio, and quantified afterwards whether the myeloablative effects were CD34+ subset-specific (figures 3A, 4A). We found a significant loss of both immature/early HSPCs (CD34 ++CD133+CD90+) and myeloid progenitors (CD34 +CD13+CD71 low), while B-cell progenitors (CD34 +CD19+CD13-CD71−) and erythroid progenitors (CD34+CD71+CD13 low) were unaffected by CD123 CARTs exposure (figure 4A). Of note, CD123 CART-mediated cytotoxicity correlated well with the expression levels of CD123 in the different CD34+ subsets (figure 4C). Collectively, our results suggest that CD123 CARTs ablate human hematopoiesis by targeting both early/immature HSPCs and myeloid progenitors.