CD33‐directed immunotherapy with third‐generation chimeric antigen receptor T cells and gemtuzumab ozogamicin in intact and CD33‐edited acute myeloid leukemia and hematopoietic stem and progenitor cells

Immunotherapies, such as chimeric antigen receptor (CAR) modified T cells and antibody‐drug conjugates (ADCs), have revolutionized the treatment of cancer, especially of lymphoid malignancies. The application of targeted immunotherapy to patients with acute myeloid leukemia (AML) has been limited in particular by the lack of a tumor‐specific target antigen. Gemtuzumab ozogamicin (GO), an ADC targeting CD33, is the only approved immunotherapeutic agent in AML. In our study, we introduce a CD33‐directed third‐generation CAR T‐cell product (3G.CAR33‐T) for the treatment of patients with AML. 3G.CAR33‐T cells could be expanded up to the end‐of‐culture, that is, 17 days after transduction, and displayed significant cytokine secretion and robust cytotoxic activity when incubated with CD33‐positive cells including cell lines, drug‐resistant cells, primary blasts as well as normal hematopoietic stem and progenitor cells (HSPCs). When compared to second‐generation CAR33‐T cells, 3G.CAR33‐T cells exhibited higher viability, increased proliferation and stronger cytotoxicity. Also, GO exerted strong antileukemia activity against CD33‐positive AML cells. Upon genomic deletion of CD33 in HSPCs, 3G.CAR33‐T cells and GO preferentially killed wildtype leukemia cells, while sparing CD33‐deficient HSPCs. Our data provide evidence for the applicability of CD33‐targeted immunotherapies in AML and its potential implementation in CD33 genome‐edited stem cell transplantation approaches.

(HSPCs). When compared to second-generation CAR33-T cells, 3G.CAR33-T cells exhibited higher viability, increased proliferation and stronger cytotoxicity. Also, GO exerted strong antileukemia activity against CD33-positive AML cells. Upon genomic deletion of CD33 in HSPCs, 3G.CAR33-T cells and GO preferentially killed wildtype leukemia cells, while sparing CD33-deficient HSPCs. Our data provide evidence for the applicability of CD33-targeted immunotherapies in AML and its potential implementation in CD33 genome-edited stem cell transplantation approaches.

What's new?
In the development of immunotherapy for acute myeloid leukemia (AML), a target of interest is CD33, which is expressed on blast cells in more than 90 percent of AML patients. CD33 is also expressed on healthy myeloid and progenitor cells, however, raising the risk for off-target effects with CD33 therapies. Here, the authors introduce a CD33-directed third-generation chimeric antigen receptor (CAR) Tcell product (3G.CAR33-T). 3G.CAR33-T cells were effective against CD33-positive cells, including AML blasts, and successfully overcame AML drug resistance. Genomic deletion of CD33 in hematopoietic stem and progenitor cells resulted in preferential killing of leukemia cells by 3G.CAR33-T cells.

| INTRODUCTION
Acute myeloid leukemia (AML) is a malignant proliferation of abnormal myeloid cells in the bone marrow, which interferes with normal hematopoiesis. Although treatment has improved the prognosis of AML patients, the 5-year overall survival (OS) among patients younger than 60 years is approximately 35% to 40%. 1 In elderly patients, the 5-year OS rate further decreases to approximately 5% to 15% with median survival of only 5 to 10 months, particularly due to frailty and comorbidities that limit the applicability of intensive chemotherapeutic regimens. 1 In addition, therapy with chemotherapeutic drugs can result in drug resistance further limiting available treatment options. 2 Thus, novel therapies for AML patients are needed.
In fact, with recently approved small molecule inhibitors, for example, the fms-like tyrosine kinase 3 (FLT-3) inhibitor midostaurin 3,4 and gilteritinib, 5 or the IDH2-inhibitor enasidenib, 6  CAR T-cell activity and signaling and mediate higher rates of clinical responses. [9][10][11] Third-generation CARs including CD28 and 4-1BB as costimulatory molecules have demonstrated superior proliferation, survival and antitumor activity as compared to second-generation CARs comprising either CD28 or 4-1BB. [12][13][14][15] Gemtuzumab ozogamicin (GO, Mylotarg), the only ADC approved for the treatment of AML, consists of a humanized immunoglobulin (Ig)G4 antibody directed against CD33 that is conjugated to the cytotoxic drug calicheamicin. 16 After binding of the antibody to CD33 and subsequent internalization, calicheamicin mediates cytotoxicity to CD33-positive myeloid cells primarily through induction of DNA damage and subsequent apoptosis. 17 GO was initially approved in 2000 for the treatment of relapsed AML patients older than 60 years of age, but was withdrawn from the market in 2010 due to a lack of benefit. 18,19 Subsequent trials, however, supported safety and efficacy of GO as first-line, single agent for the treatment of older adults with AML unsuitable for intensive chemotherapy, 20 as well as for newly diagnosed AML patients in combination with induction therapy. 21,22 Based on these trials, GO was approved in September 2017 for the treatment of adults with CD33-positive newly diagnosed AML and adults and children aged 2 years or older with relapsed or refractory CD33-positive AML. 23 In June 2020, the FDA extended the approval of GO to newly diagnosed pediatric patients ≥1 month of age. 24 Further studies applying optimized treatment strategies including fractionated GO doses confirmed a lower risk of relapse and prolonged survival for AML patients treated with GO. 21,[25][26][27][28][29][30] CD33 (Siglec-3), the target molecule of GO, is a member of the sialic acid-binding Ig-like lectin (Siglec) family. It is expressed as a transmembrane glycoprotein on healthy and malignant myeloid cells. 31,32 In AML, which accounts for 70% of acute leukemias in adults, 33 CD33 is expressed on blasts of >90% patients and importantly also on leukemic stem cells. [34][35][36] Accordingly, CD33 is a promising target for immunotherapy of AML, particularly with antibody-based treatment [37][38][39] and CAR T cells. 40,41 Full-length CD33 consists of two extracellular domains, the V-set Ig-like (IgV) and C2-set Ig-like (IgC2), a transmembrane, and an intracellular domain. In addition to the full-length isoform, CD33 is expressed as a shorter IgV-lacking isoform (CD33-D2) generated by alternative splicing in some patients. 42,43 CD33 is not only expressed on leukemic cells but also on virtually all healthy myeloid and progenitor cells. 44,45 The ontarget off-leukemia toxicity is a major side effect observed in the clinical practice and in clinical trials investigating CD33-targeting therapies.
Thus, efforts have focused on the optimization of CD33-targeting therapies. Recent studies demonstrated that CD33-deficient hematopoietic stem and progenitor cells (HSPCs) maintain their full function in terms of engraftment, differentiation ability to all lineages, morphology and response to proinflammatory stimuli while conferring resistance to CD33-directed treatment. 44,45 In our study, we evaluated anti-CD33 immunotherapies in AML, using both CAR T cells and the recently reapproved ADC GO. We generated a novel CD33-specific third-generation CAR (3G.CAR33) and performed in-depth analysis of 3G.CAR33 expressing T cells (3G. CAR33-T) regarding their functionality towards both, AML cell lines and primary patient-derived AML cells. We demonstrate superior efficacy of our 3G.CAR33-T cells compared to second-generation CD33-directed CAR T cells (2G.CAR33-T). We further provide evidence that CD33-deletion in healthy and malignant hematopoietic cells confers resistance to CD33-targeted immunotherapies, 3G.CAR33-T cells and GO. Consistent with previous work, we confirm CD33-deletion in primary HSPCs to be an efficient and feasible approach to reduce off-tumor targeting of normal myeloid progenitor cells, while preserving on-tumor efficacy. For CAR T-cell manufacturing, peripheral blood (PB) mononuclear cells (PBMCs) of three healthy donors (HDs) were collected at the Heidelberg University Hospital, Heidelberg, Germany. Primary AML blasts were isolated from patient PB (n = 9) by Ficoll density gradient centrifugation.

| Generation of 3G.CAR33 and 2G.CAR33s
The coding sequence of the CD33-specific single chain variable fragment (scFv) was amplified via polymerase chain reaction (PCR) based on the sequence of the anti-CD33 GO antibody clone p67.7. The purified PCR product was inserted into the retroviral vector RV-SFG.

| Generation of CAR T cells
Details of transfection, retrovirus production and manufacturing of CD19-specific CAR T cells have been described previously and were used for generation of CD33-specific CAR T cells. 46

| Western blot
Western blot was performed as previously described. 51 In brief, 1 Â 10 6 cells were lysed in 200 μL radio-immunoprecipitation assay buffer (RIPA buffer; Thermo Fisher Scientific) with 1Â complete protease inhibitor cocktail (Roche, Basel, Switzerland) at RT for 10 minutes followed by centrifugation at 12 000g at 4 C for 10 minutes. Protein-containing supernatant was collected and loaded on SDS-PAGE gels. After electrophoresis, separated proteins were immediately blotted onto nitrocellulose membranes. Prior to incubation with primary anti-CD33 antibody (clone 67.7; BD Biosciences) at RT for 2 hours, the membranes were blocked for 1 hour at RT or overnight at 4 C with 5% milk powder in phosphate-buffered saline with Tween 20 (PBST). Horseradish peroxidase-conjugated secondary antibodies (Agilent, Santa Clara, California) were used in a dilution of 1:500. Proteins were visualized in an Amersham Imager 600 (GE life sciences, Massachusetts).   (Figure 2A). MV4-11 R is a generated cell line that is resistant to varying tyrosine kinase inhibitors (TKI) such as PKC412 (Midostaurine) and AC220 (Quizartinib), as well as standard chemotherapeutic agents like cytarabine (AraC) and daunorubicin, 52 whereas HL60 R is resistant to venetoclax and azacytidine ( Figure S1).
Next, we assessed the cytolytic capacity of 3G.CAR33-T cells using a standardized chromium-51 ( 51 Cr) release assay. 3G.CAR33-T cells mediated cytotoxicity against all myelogenous cell lines in a dose-dependent manner ( Figure 2D). CAR-mediated lysis of MV4-11 and MV4-11 R cells, as well as HL60 and HL60 R was comparable suggesting that CD33-targeting therapy with 3G.CAR33-T cells can overcome AML drug resistance. In contrast, CD33-negative Daudi cells were not eliminated by 3G.CAR33-T cells, but were lysed by CD19-directed CAR T cells (Figures 2D and S3). These 3G.
CAR19-T cells, however, were unable to kill HL60 cells ( Figure S3), indicating that the CAR T cell cytotoxicity of the 3G.CAR33-T cell was antigendependent. Further, we confirmed cytolytic capacity of 3G.CAR33-T cells against primary AML blasts (n = 6) ( Figure 2E) and primary HSPCs (n = 4) ( Figure 2F), although lysis of HSPCs was less pronounced when compared to AML cell lines and primary CD33-positive AML blasts. Of note, the CD33-negative AML blasts derived from patient 14 ( Figure 2B Figures 2G,H and S4). In line with the results from the shortterm killing assay ( Figure 2D-F), cytokine secretion was lower when 3G.
For CD8-positive CAR T cells, the highest cytokine secretion was observed for 2G.CD28.CAR33-T cells ( Figure 3E). Consistent with the cytotoxic activity, 2G.4-1BB.CAR33-T cells displayed the lowest cytokine release levels ( Figure 3E). Taken altogether, 3G.CAR33-T cells displayed optimized properties in terms of proliferation and shortterm antileukemia activity when compared to second-generation constructs.
To assess the long-term killing capacity of different CAR T cells,

| Editing of CD33 using CRISPR/Cas9 in myeloid cell lines and HSPCs
To prevent on-target off-leukemia toxicity of CD33-targeted therapies, we disrupted the CD33 gene in human myeloid cell lines as well as in HSPCs using CRISPR/Cas9 mediated gene editing. Three gRNAs targeting the extracellular IgV domain of CD33 (gRNA V1-V3) and three gRNAs targeting the IgC2 domain (gRNA C1-C3) were designed ( Figure 4A; Table S2). The human CD33-positive myeloid cell line  Figure 4B). We confirmed the KO of CD33 by western blot ( Figure 4C) and Sanger sequencing ( Figure 4D). All transductions were performed independently in triplicates, generating consistent data and thus highlighting reproducibility of the knockout.    Figure 5A).
As random integration of transgenes via lentiviral transduction is of concern in primary cells, we used a nonviral genome editing approach.
3.5 | Cytotoxicity of 3G.CAR33-T cells toward CD33-edited cells Extrapolation of these results to other hematologic disease entities is of significant interest. Prior studies evaluating CAR T cells directed against CD33 in the context of AML have primarily used second-generation CAR33-T cells. 45,55 In our study, we generated a third-generation CD33-directed anti-AML CAR construct using the RV-SFG.
CD28.4-1BBzeta backbone of the HD-CAR-1 trial, hence introducing a new third-generation CAR33-T construct with a backbone that is already successfully applied in the clinic.
In the context of CD19-directed CAR T cell therapy, thirdgeneration CAR T cells have shown significantly superior engraftment, a 23-fold higher expansion and prolonged in vivo persistence when compared to second-generation CAR constructs. 15 Clinically, superior expansion and longer persistence of third-generation CAR T cells was observed when second-generation (CD28 costimulatory domain) and third-generation (CD28 and 4-1BB) CAR T cells were administered simultaneously to lymphoma patients. 56 Here, we demonstrate that also third-generation CD33-directed CAR T cells are associated with enhanced cytotoxic efficacy and prolonged proliferation capacity when compared to second-generation CD33-directed CAR T cells. In addition, our third-generation CAR33-T cells were effective against CD33-positive cells including AML blasts and were even able to over-  In summary, our study confirms the feasibility of targeted immunotherapy, GO and 3G.CAR33-T cells against AML and provides valuable new insight into CAR T therapy in AML. We provide evidence that third-generation CD33-directed CAR T cells are associated with improved clinical efficacy over second-generation CAR constructs.
Based on our clinical data on third-generation CD19 CAR T cells we expect manageable toxicity in the clinical setting. CRISPR/ Cas9-mediated CD33 KO in CD33-positive tumor cells as well as primary HSPCs was feasible and CD33-edited cells were resistant to GO and CAR33-directed T cells.

DATA AVAILABILITY STATEMENT
The data that support the findings of our study are available from the corresponding author upon reasonable request.