MoAbs reactive with human CD22 were generated by fusion of NS-1 myeloma cells with splenocytes from a Balb/c mouse previously immunized three times with 30×10⁶ CD22-transfected 300.19 cells. Supernatants from hybridoma-containing wells were screened by flow cytometry for moAbs reactive with CD22-transfected 300.19 cells. Three reactive hybridomas were chosen for further characterization and were subcloned by limiting dilution (clones hCD22.7, hCD22.316 and hCD22.401). The three antibodies were IgG1κ isotype, as determined using a mouse moAb isotyping kit (Boehringer Mannheim, Mannheim, Germany). Antibody specificity was further validated using a CD22-transfected COS cell line (pUNO1-hCD22; InvivoGen, Toulouse, France), the cell lines Raji, Daudi, PRMI 82229, U266 and Jurkat, and peripheral blood (PB) mononuclear cells (PBMCs). CD22 epitope binding to the hCD22.7 clone was determined by cross-blocking24 with an antibody against the N-terminal domain of CD22 (S-HCL-1).25 The CD22-specific scFv derived from the hCD22.7 clone was obtained using the mouse IgG Library Primer Set (Progen, Heidelberg, Germany).

Immunogenicity potential was predicted using the in silico tool NetMHC V.4.0.26 27 Analyzed scFvs were broken in 9-mer peptides, length for which human leukocyte antigen (HLA) molecules have a strong preference. Peptides were considered to bind major histocompatibility complex class I (MHC-I) when a half-maximum inhibitory concentration (IC₅₀) is <500 nM, being considered strong binders when IC₅₀ <50 nM. Only HLA supertype representatives were analyzed. The predicted epitope-binding affinity was calculated as 1/IC₅₀ 100.28

Reactivity of the hCD22.7 moAb was subjected to PEPperMAP Conformational Epitope Mapping against CD22 translated into cyclic constrained peptides (PEPperPRINT GmbH, Heidelberg, Germany). The sequence of CD22 (UniProtKB P20273) was elongated with neutral GSGSGSG linkers at the C-terminus and N-terminus to avoid truncated peptides. The elongated antigen sequence was translated into 7, 10 and 13 amino acid peptides with a peptide–peptide overlap of 6, 9 and 12 amino acids. After peptide synthesis, all peptides were cyclized via a thioether linkage between a C-terminal cysteine and an appropriately modified N-terminus. The resulting conformational CD22 peptide microarrays contained 2556 different peptides printed in duplicate (5112 peptide spots), and were framed by additional human influenza hemagglutinin (HA, YPYDVPDYAG, 146 spots) control peptides. Hibridoma supernatant containing hCD22.7 moAb was diluted in incubation buffer (phosphate-buffered saline (PBS), pH 7.4, with 0.005% Tween 20) and 10% MB-070 blocking buffer (Rockland Immunochemicals, Limerick, Pennsylvania, USA) and incubated with peptide microarrays for 16 hours at 4°C and swirling at 140 rpm. Unbound antibodies were removed by washing the arrays with washing buffer (PBS, pH 7.4, with 0.005% Tween 20). Prestaining of the peptide arrays was done with 0.2 µg/mL goat anti-mouse IgG (H+L) antibody conjugated to DyLight680 (Thermo Fisher Scientific). HA control peptides framing the peptide arrays were finally stained with 0.5 µg/mL monoclonal anti-HA (12CA5)-DyLight800 (Thermo Fisher Scientific) for 45 min at room temperature (RT). Quantification of spot intensities and peptide annotation was performed with the LI-COR Odyssey Imaging System (scanning offset 0.65 mm, resolution 21 µm, and scanning intensities of 7/7 (red=700 nm/green=800 nm) and PepSlide Analyzer.

The structural model of the CD22-hCD22.7-scFv complex was obtained using directed docking as follows. The structure for CD22 was taken for the PDB databank29 (identification code 5vkj30) and the structure of the scFv anti-CD22 was derived by homology modeling using M4T31 and templates 4h0h,32 1dqd33 and 4okv.34 The initial docking conformations were inferred using PatchDock35 by selecting the complementarity-determining regions of anti-CD22-scFv and mapped epitope of CD22 as docking interfaces. Resulting docking conformations were further refined and minimized using Rosetta36 and the top 200 poses were visually inspected. A more detailed description of the structural analysis is shown in online supplementary information.

The CD22-specific scFv was cloned into a pCCL lentiviral-based backbone containing a human CD8 hinge and transmembrane (TM) domain, human 4-1BB and CD3ζ endodomains (second-generation CAR), and a T2A-GFP cassette.37 An identical lentiviral vector with the CD8 TM-4-1BB-CD3ζ domains linked to a His-Tag was used as a mock intracellular (mock-IC) control. CAR-expressing viral particles pseudotyped with VSV-G were generated by cotransfection of HEK 293 T cells with the pCCL vector and the packaging plasmids VSV-G and psPAX2 using polyethylenimine (PEI, Polysciences, Warrington, Pennsylvania, USA). Supernatants were collected at 48 and 72 hours after transfection and concentrated by ultracentrifugation.

PBMCs were isolated from buffy coats from healthy volunteers by Ficoll-Hypaque gradient centrifugation (GE Healthcare, Chicago, Illinois, USA). Buffy coats were obtained from the Barcelona Blood and Tissue Bank on Institutional Review Board approval (HCB/2018/0030). T cells were activated by plate coating with anti-CD3 (OKT3) and anti-CD28 (CD28.2) antibodies (BD Biosciences, Franklin Lakes, New Jersey, USA) for 2 days and were transduced with the CAR-expressing lentivirus at a multiplicity of infection of 10 in the presence of interleukin 7 (IL-7) and IL-15 (10 ng/mL; Miltenyi Biotec, Bergisch Gladbach, Germany).37 T cells were expanded in RPMI-1640 medium (Gibco/Invitrogen, Waltham, MA) containing 10% heat-inactivated fetal bovine serum (FBS, Sigma, St. Louis, Missouri, USA), penicillin-streptomycin (Gibco/Invitrogen), and IL-7 and IL-15 (10 ng/mL), for up to 10 days. Surface expression of CD22-CAR was traced by fluorescence-activated cell sorting (FACS).

SEM, NALM6, MV4-11 and REH cells lines were purchased from the DSMZ cell line bank (Braunschweig, Germany). CD19-knock out (KO) and CD22-KO SEM cells were generated by CRISPR-mediated genome editing. Briefly, 200,000 cells were electroporated (Neon transfector, ThermoFisher Scientific, Waltham, Massachusetts, USA) with a Cas9 protein/tracrRNA/crRNA complex (IDT, Coralville, Iowa, USA). Two guides were designed for each gene: CD19-exon 2.1, CAGGCCTGGGAATCCACATG and CD19-exon 14.1, AGAACATGGATAATCCCGAT; and CD22-exon 3.2, TCAATGACAGTGGTCAGCTG and CD22-exon 9, CAGGTGTAGTGGGAGACGGG. After electroporation, cells were allowed to recover and then CD19^– or CD22^– cells were FACS sorted (>99% purity).

Target cells (cell lines or primary B-ALL cells; 100,000 target cells/well of a 96-well plate) were incubated with CD22-CAR or mock-IC T cells at different effector:target (E:T) ratios for the indicated time periods. Cell lines were cultured in RPMI-1640, 10% FBS and penicillin–streptomycin. Primary cells were cultured in StemSpan^TM SFEM media (StemCell Technologies, Vancouver, Canada), 20% FBS, penicillin–streptomycin, insulin–transferrin–selenium (Gibco/Invitrogen), hSCF (100 ng/mL), hFLT3L (100 ng/mL), hIL3 (10 ng/mL) and hIL7 (10 ng/mL, all from Miltenyi Biotec, Bergisch Gladbach, Germany). CAR T cell-mediated cytotoxicity was determined by analyzing the residual live (7-aminoactinomycin D^–; 7-AAD^–) target cells at each time point and E:T ratio. For absolute cell counting, BD TruCount^TM absolute count tubes (BD Biosciences) were used. Quantification of the proinflammatory cytokines IL2, TNFα and IFNγ was measured by ELISA using BD OptEIA^TM Human ELISA kits (BD Biosciences), in supernatants harvested at 24 hours (cell lines) and 48 hours (primary cells) post T cell-exposure, using an 1:1 E:T ratio. Online supplementary table 1 shows clinic-biological features of the B-ALL samples used.

Cell surface expression of the CD22-CAR was confirmed by green fluorescence protein (GFP) expression, binding to an AffiniPure F(ab')₂ fragment goat anti-mouse IgG (H+L)-APC (Jackson ImmunoResearch, Westgrove, Pennsylvania, USA), and an anti-His-APC (J095G46; BioLegend, San Diego, California, USA), after prior incubation with human recombinant CD22-His (rCD22-His, ThermoFisher Scientific). Activation of transduced T cells was confirmed 48 hours after plating of PBMCs by surface staining with CD3-PE (UCHT1; BD Biosciences), CD25-APC (M-A251; BD Biosciences) and CD69-VioBlue (REA824; Miltenyi Biotec). T cells and target cells were identified in in vitro assays using the following human antibodies: CD3-PE (UCHT1), CD22-APC (HIB22), CD19-BV421 (HIB19) and CD10-PECy7 (HI10a) or CD13-PECy7 (WM15, for SEM cells) or CD33-APC (WM53, for MV4-11 cells, BD Biosciences). Dead cells were discarded by 7AAD staining. Cells collected from mouse PB, bone marrow (BM) and spleen were stained with HLA-ABC-PE (G46-2.6), CD45-BV510 (HI30), CD3-PerCP (SK7), CD22-APC, CD19-BV421 and CD10-PECy7 (BD Biosciences). Cells were incubated for 30 min and then lysed and fixed with the BD FACS^TM Lysing solution (BD Biosciences). Fluorescence Minus One controls were used to set the gates (online supplementary figure 1). A FACSCanto^TM-II flow cytometer and a FACSDiva^TM software was used for the analysis (BD Biosciences).

Twelve-week-old non-obese diabetic (NOD).Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice (The Jackson Laboratory, Bar Harbor, Maine USA) were bred and housed under pathogen-free conditions. All in vivo procedures were approved by a local ethics committee (HRH-17–0029 P1). In total, 0.5–1.5×10⁶ patient derived xenograft (PDX) B-ALL cells were intra-BM transplanted in sublethally irradiated (2 Gy) NSG mice, followed by intravenous infusion of 4×10⁶ CD22-CAR T cells (average transduction ~51.3%) or mock-IC T cells (average transduction ~54.2%) 2 weeks later. B-ALL engraftment was monitored in PB every other week. BM aspirates were analyzed 6 weeks after transplantation and at sacrifice, along with spleens. Red blood cell counts were determined with a hemocytometer 2800VET V-Sight (Menarini Diagnostics, Badalona, Spain).

All data are expressed as the mean±SEM. Differences between groups were assessed using two-tailed unpaired Student’s t-tests, or paired Student’s t-tests when analyzing data from different donors, unless otherway stated. All analyses were performed with Prism software, V.8.0 (GraphPad software, San Diego, California, USA).

Our novel anti-CD22 scFv (clone hCD22.7) was cloned into a second-generation CAR construct consisting of an anti-CD22 scFv, a CD8 hinge/TM spacer, and intracellular signaling domains from 4-1BB and CD3ζ, coupled in-frame with GFP through a T2A sequence (figure 1A). To identify the domain of CD22 recognized by the hCD22.7 moAb, we performed a cross-blocking assay with the anti-CD22 clone H-SCL-1, known to bind to the most distal Ig extracellular domain 1 of CD22.25 Binding of the labeled H-SCL-1 clone was hindered in cells preincubated with the antibody hCD22.7, indicating an overlapping binding of both moAbs to the most distal Ig domain of CD22 (figure 1B). Additionally, conformational epitope mapping assay, performed to exactly identify the sequence recognized by the hCD22.7 scFv, revealed that hCD22.7 scFv specifically binds to the ESTKDGKVP sequence, located in the Ig-like V-type domain, the most distal domain of CD22, thus confirming results from the cross-blocking assay (online supplementary figure 2). The molecular docking of this binding was computationally modeled (figure 1C and online supplementary file 1). The structural model inferred a CD22-hCD22.7-scFv complex energy of −1407.691 Rosetta score,36 and a binding free energy of −16.5615 kcal/mol, using the InterfaceAnalyzer.38 These binding affinity parameters were comparable, or even slightly higher than those of the clinically approved high-affinity anti-CD22 moAb Epratuzumab (−1005.624 Rosetta score and −11.345 kcal/mol, respectively).30

Healthy human T cells were successfully transduced with both CD22-CAR and mock-IC lentivectors with a transduction efficiency of 20%–50%. The expression of the CD22-CAR in T cells was confirmed by codetection of scFv and GFP, and by using an anti-His moAb on incubation with human rCD22-His (figure 1D). Importantly, activated (CD69⁺CD25⁺) CD22-CAR T cells (figure 1E) continuously expanded >100-fold over a 15-day period, similar to mock-IC T cells (figure 1F), demonstrating that hCD22.7-CAR expression does not hamper T cell expansion.

CD22-CAR T cells from three different healthy donors were then functionally tested in vitro in cytotoxicity assays against the B-ALL cell lines NALM6, SEM and REH, and the AML cell line MV4-11 as a negative control, at increasing E:T ratios (figure 2A). Alive target cells after hCD22.7-CAR T cell exposure were identified as 7AAD^–CD3^–CD19⁺CD10⁺ (or CD13⁺ in the case of the SEM cells and 7AAD^–CD3^–CD33⁺ in the case of MV4-11 cells, figure 2B). Compared with mock-IC T cells, hCD22.7-CAR T cells specifically eliminated CD22⁺ B-ALL cells, even at relatively low E:T ratios (figure 2C). Specificity of the hCD22.7-CAR T cells was further demonstrated using CRISPR/Cas9-mediated CD19-KO and CD22-KO SEM cells (figure 2D). Furthermore, B-ALL cells barely survived exposure to hCD22.7-CAR T cells in a 48 hours absolute number assay at a 1:1 E:T ratio (figure 2E). hCD22.7-CAR T cells produced high levels of the proinflammatory cytokines IL2, TNFα, and IFNγ after coculture with target cells, confirming their cytotoxicity (figure 2F).

It has been reported that remissions post-CD22-CAR T cell therapy are short and are associated with reduced expression of CD22, whereas high expression of CD22 favors the activity of CD22-CAR T cells.22 Moreover, whether CD22^–/dim B-ALL relapses are common after CD22-CAR T cell treatment remains controversial.15 21 Given this information, we next tested the efficacy of our CD22-CAR T cells in vitro, in nine primary B-ALL samples with variable expression levels of CD22. CD22 median fluorescence intensity (MFI) levels above or below the mean value (MFI=6752) were used to categorize primary B-ALLs as CD22^high or CD22^low (figure 3A and online supplementary figure 3a). Cytotoxicity was measured after 24 hours and 48 hours exposure to CD22-CAR/mock-IC T cells (figure 3B and online supplementary figure 3b). Overall, hCD22.7-CAR T cells significantly eliminated primary B-ALL blasts (figure 3C). However, while a sustained cell elimination of CD22^high B-ALL blasts (n=3) by hCD22.7-CAR T cells was consistently observed over a 48 hours period (figure 3C, left panel), hCD22.7-CAR T cell-mediated killing of CD22^low B-ALL blasts (n=6), although significant, was less robust and occurred in a shorter or delayed cytotoxic window (figure 3C, right panel). Nevertheless, a similar trend of robust production of the proinflammatory cytokines IL2, TNFα, and IFNγ was observed for all B-ALL primary samples, regardless the expression levels of CD22 (figure 3D), confirming an efficient CD22 recognition and killing of B-ALL primary cells by our hCD22.7-CAR T cells.

We next assessed the activity of our hCD22.7-CAR in vivo using both CD22^high (ALL#1 and ALL#2) and CD22^low (ALL#10) PDX B-ALL samples (figure 4, online supplementary table 1). NSG mice were intra-BM transplanted with 0.5–1.5×10⁶ primary CD22⁺ B-ALL blasts, followed by infusion of 4×10⁶ CD22-CAR/mock-IC T cells 2 weeks later. Leukemic engraftment and T cell persistence were monitored by FACS biweekly in PB, and analyzed in BM and spleen at sacrifice, when any animal developed signs of overt disease (figure 4A,B). Of note, the kinetics of leukemia reconstitution (aggressiveness) detectable between week 4 and 20 post-transplant was unique for each PDX, and the engraftment of both CD22^high and CD22^low PDXs tended progressively to increase over time in mock-IC T cell-treated mice (figure 4C). Importantly, however, hCD22.7-CAR T cells eradicated (ALL#1 and ALL#10) or largely controlled (ALL#2) the growth of the both CD22^high and CD22^low B-ALL cells (figure 4C,D). T cell persistence adapted to the reconstitution kinetics of each PDX, and was detected in all three PDXs, even 26 weeks after CAR T cell infusion (range, week 6 (n=15) to week 26 (n=5): mock-IC group 2.38%±0.65%–0.34%±0.34%; CAR T cell group 2.79%±1.13%–0.14%±0.09%, Figure 4E and online supplementary figure 4a-b). Disease control was further confirmed at sacrifice in ipsilateral (injected tibia) and contralateral (both femurs and non-injected tibia) BM, PB and spleen through FACS quantification of tumor burden (figure 4F and online supplementary figure 4c), lack of splenomegaly (figure 4G), and signs of hematopoietic displacement (figure 4H). Importantly, persistent leukemic cells at sacrifice maintained expression of CD22, likely ruling out CAR pressure-mediated antigen loss (figure 4I).

Finally, any engineered scFv of either murine or human origin expressed in autologous T cells may be immunogenic if presented on MHC-I, because peptides derived from it may be recognized as ‘non-self’ antigens by the adaptive immune system.39 We have thus compared the immunogenicity capacity of our hCD22.7-scFv side-by-side with that of m97115 and HA2220 clones using the in silico tool NetMHC V.4.0 which predicts the affinity of peptides to bind to the most representative HLA supertypes. As shown in figure 5, we found very similar numbers of total (29, 23 and 26) and strong (8, 9 and 6) binding 9-mer peptides for hCD22.7, m971 and HA22, respectively.