Discovery and preclinical development of a therapeutically active nanobody-based chimeric antigen receptor targeting human CD22

Abstract

Chimeric antigen receptor (CAR) T cell therapies targeting B cell-restricted antigens CD19, CD20, or CD22 can produce potent clinical responses for some B cell malignancies, but relapse remains common. Camelid single-domain antibodies (sdAbs or nanobodies) are smaller, simpler, and easier to recombine than single-chain variable fragments (scFvs) used in most CARs, but fewer sdAb-CARs have been reported. Thus, we sought to identify a therapeutically active sdAb-CAR targeting human CD22. Immunization of an adult Llama glama with CD22 protein, sdAb-cDNA library construction, and phage panning yielded >20 sdAbs with diverse epitope and binding properties. Expressing CD22-sdAb-CAR in Jurkat cells drove varying CD22-specific reactivity not correlated with antibody affinity. Changing CD28- to CD8-transmembrane design increased CAR persistence and expression in vitro. CD22-sdAb-CAR candidates showed similar CD22-dependent CAR-T expansion in vitro, although only membrane-proximal epitope targeting CD22-sdAb-CARs activated direct cytolytic killing and extended survival in a lymphoma xenograft model. Based on enhanced survival in blinded xenograft studies, a lead CD22sdCAR-T was selected, achieving comparable complete responses to a benchmark short linker m971-scFv CAR-T in high-dose experiments. Finally, immunohistochemistry and flow cytometry confirm tissue and cellular-level specificity of the lead CD22-sdAb. This presents a complete report on preclinical development of a novel CD22sdCAR therapeutic.

Keywords: CAR optimization; CAR-T; CD22; MT: Regular Issue; cell therapy; chimeric antigen receptors; hematological malignancy; leukemia and lymphoma; nanobody; preclinical development; single-domain antibody.

Graphical abstract

Figure 1

Identification of novel CD22-targeting nanobodies with varying affinity and epitope specificity (A) Structure of CD22 protein, including 7 large Ig-like ECDs. Antibody structures are also shown for either a camelid heavy-chain-only antibody (top) or conventional heavy-/light-chain antibody (bottom). (B) CD22-specific serum response in a llama was measured via ELISA at preimmunization and after CD22-ECD immunization and boosting. (C) Representative SPR sensorgrams showing specific sdAb binding to immobilized CD22-ECD. (D) Distribution of 20 anti-CD22 sdAb equilibrium K_Ds. (E) Representative SPR co-injection sensorgrams from epitope binning experiments showing a pair of sdAbs binding distinct CD22 epitopes (top) and a pair of sdAbs binding an overlapping CD22 epitope (bottom). (F) YSD of specific CD22 domains was performed to map binding of sdAbs. (G) A summary of the CD22 domain-specific binding for sdAbs.

Figure 2

High-throughput CAR activity screening for sdAbs (A) CD22-specific sdAbs were transferred directly from phagemid-VHH vectors into pSLCAR screening backbone via PCR and tested for CAR-Jurkat assay. Jurkat cells were electroporated with various CD22-sdCAR, CD22-scFv-CAR, or control scFv-CAR plasmids and immediately mixed at 1:1 E:T with various red fluorescent protein-marked target cells. (B) Ramos or Ramos-CD22KO cells. (C and D) Raji or Raji-CD22KO cells (C), or (D) K562-WT, K562-CD22-overexpressing, or irrelevant SKOV3 cells. After overnight coculture, cells were stained with anti-human CD69 antibody and assessed via flow cytometry. (E) The mean CD69 expression of all CD22sdCARs versus on-rate (k_on), off-rate (k_off), or affinity (K_D) derived from SPR measurements is shown. (F) CAR-Jurkat cell target-cell doublet formation was examined using coculture with CD22⁺ Ramos cells or CD22KO cells for 30 min, followed by flow cytometry. Results show the mean of 3 experiments performed in duplicate ± SEM.

Figure 3

CD8TM CAR design increases CAR persistence (A) Diagram of CAR design used for initial screening experiments (CD28TM) and for lead selection studies (CD8TM) in which the hinge and transmembrane domains were changed. (B) CD22-specific CD28TM and CD8TM CARs with various antigen-binding domains were examined via CAR-Jurkat assay with CD22⁺ Ramos cells (red lines) or CD22KO cells (blue lines). (C) Primary CAR-T cells were generated from healthy donor PBMCs and transduced with CD22-CAR lentivirus before being examined for the expression of green fluorescence, CD25, and CD69 expression at various time points after transduction. (D) CAR-T cells were stained with an in-house broadly reactive anti-VHH reagent to assess CAR surface expression. (E–G) Total fold expansion of CD28TM or CD8TM CD22 CAR-T cell products at harvest is shown (E). CAR-T cells were then placed in low-density culture with IL-7/IL-15 supplementation and examined for (F) CAR-T cell signal via green fluorescence without target cells or (G) the total T cell number via confluence measurement. (H and I) Similarly, CAR-T cells were cocultured with CD22⁺ Raji cells, Ramos cells, or Ramos-CD22KO cells and examined for (H) red fluorescent target cell growth and (I) green fluorescent CAR-T growth (bottom graphs). Each graph presents the mean of 2 duplicate wells from a single experiment.

Figure 4

In vivo lead selection study for CD22sdCAR CAR-T cells were generated from healthy donor apheresis product as described in the Materials and methods section and cryopreserved on day 14. (A) CAR-T functionality was assessed via chromium release cytolytic assay using CD22⁺ Raji cells or CD22⁻ MCF7 target cells. (B) CAR-T cells were stimulated with CD22⁺ Ramos cells and allowed to return to rest after 18 days in culture, before repeated cytolytic cell killing assay. Results show the mean of 3 experiments performed in duplicate ± SEM. (C) Nod-SCID-IL-2Rγ-null (NSG) mice were then injected with 5 × 10⁴ Ramos-FLUC cells and randomly assigned to cages. At day 3, cage groups were randomly assigned to treatment groups with 2.5 × 10⁶ CD22 CAR-T cells or equivalent total dose of unmodified (mock) T cells (n = 10 mice per group). Mice were monitored for by distress and euthanized at predetermined humane endpoints. Graphs show the number of surviving mice at various time points. p values show the comparison of survival for treatment groups to untransduced mock T cells via the log-rank test. Mice were also assessed via IVIS imaging for bioluminescent signals from Ramos-FLUC tumors at various time points as shown in (D) images and (E) graphs displaying the biolumnescent signal detected per mouse over the timecourse. (F and G) Blood samples were also obtained at regular intervals for assessment of (F) circulating hCD45⁺hCD19⁺ Ramos cells or (G) hCD45⁺hCD19⁻NeonGreen⁺ CAR-T cells via flow cytometry.

Figure 5

Confirmation of 1ug36-CD22sdCAR activity in multiple donor samples NSG mice were injected with 5 × 10⁴ Ramos-FLUC cells and randomly assigned to cages. At day 3, posttumor injection cage groups were randomly assigned to blinded treatment groups with nonfluorescently labeled 2.5 × 10⁶ CD22 CAR-T cells or equivalent total dose of unmodified (mock) T cells generated from 3 different donor lymphocyte samples (n = 10 mice per group). Mice were monitored for by distress and euthanized at predetermined humane endpoints. (A) The number of surviving mice separated by donor sample is shown in survival graphs. (B) Mice were also assessed for tumor engraftment via IVIS imaging for bioluminescent signals from Ramos-FLUC tumors at day 22 after tumor challenge as shown. p values show comparisons using Student’s t test with log-transformed bioluminescence values. (C and D) A similar experiment was repeated to confirm intradonor variability with this model wherein (C) mouse survival and (D) tumor load via bioluminescent imaging at day 18 after tumor challenge. (E) In the experiments above, after unblinding and analysis of survival, mice were rechallenged with an additional dose of 5 × 10⁴ Ramos-FLUC cells at day 85 posttumor challenge. Pooled results for all of the experiments are shown (n = 40 mice per treatment group, 20 mice for untreated group). p values in survival graphs show intergroup comparisons via the log-rank test.

Figure 6

High-dose CD22sdCAR treatment leads to complete tumor responses CAR-T cells were generated from healthy donor PBMCs as described in the Materials and methods section and prepared for injection on day 14 after transduction. NSG mice were injected with 5 × 10⁴ Ramos-FLUC cells and randomly assigned to cages. At day 3, cage groups were treated with 2.5 × 10⁶ (marginal dose) or 12.5 × 10⁶ (high dose) CD22 CAR-T cells or equivalent to the highest total dose of unmodified (mock) T cells (n = 5 mice per group). Mice were assessed for tumor engraftment via IVIS imaging for bioluminescent signal from Ramos-FLUC tumors at various time points as shown in (A) bioluminescent signal data per mouse over the timecourse and in (B) mouse images. (C) Blood samples were also obtained at regular intervals for the assessment of hCD45⁺hCD19⁻NeonGreen⁺ CAR-T cells via flow cytometry. (D) Surviving mice in the high-dose treatment group were rechallenged with an additional dose of 5 × 10⁴ Ramos-FLUC cells intravenously, including new untreated control mice with no cell treatment. Graphs show the bioluminescent signals for 1ug36, m971, or untreated mice. (E) Proportion of surviving mice at various time points throughout the experiment is shown. p values show intertreatment group comparison at similar dose levels via the log-rank test.

Figure 7

Absence of unexpected cell or tissue binding for 1ug36-sdAb (A) Assay setup for flow cytometry staining approach. Graphs show the median fluorescence intensity across the primary antibody titration range for 3 CD22⁺ cell lines (Raji, Ramos, Jeko-1), 2 CD22KO lines (Raji-CD22ko and Ramos-CD22ko), and 7 irrelevant target lines (MCF7, SKOV3, H1581, U87MG, SKRC52, H292, and HDFs) as measure via flow cytometry. Graphs show the results from a single experiment performed in duplicate using (B) 1ug36-Fc primary antibody or (C) a control irrelevant VHH-Fc antibody. (D) Assay setup for IHC staining approach of human frozen tissue array. (E) IHC images for lymphatic tissues in human tissue array. (F) Representative images for staining in nonlymphoid tissues. The full image set can be found in Table S3.

Figure 8

Model of interactions of various CD22sdCAR constructs with summary of observations Data presented in this study indicate that membrane-distal CARs can induce activation and expansion of CAR cells, but do not have strong cytolytic activity or extend survival in an in vivo model. More membrane-proximal 1ug13 and 1ug36 with adequate cell-binding capacity show better cytolysis and can extend survival in a xenograft lymphoma model, similar to a benchmark CD22-targeting m971-scFv-CAR, which has been engineered for tonic signaling. We also tested several highly membrane-proximal CD22sdCARs that show low overall response to CD22-expressing targets.