Rational redesign of antigen binding domain improves in vivo efficacy of the CD22-CAR

Abstract

Chimeric antigen receptor (CAR) T cells targeting CD19 are highly effective against B-lineage malignancies. However, about half of patients either fail to achieve complete remission (in the case of lymphoma) or relapse (in the case of acute lymphoblastic leukemia). CD22 represents an alternative highly B-lineage-restricted target. Although CD22-targeted CAR T cells are clinically active, targeting this antigen has proven difficult relative to CD19, attributable, in part, to lower expression levels. Commonly, patients relapse or progress with reduced CD22 expression compared to pre-treatment, contrasting with the loss of CD19 expression typically observed after CD19-CAR therapy. Prior work demonstrated that an antigen-independent “tonic” signaling CD22-CAR has enhanced clinical efficacy relative to other CD22-CARs tested. However, tonic signaling has been shown to be detrimental to long-term CAR T cell function. Here, we demonstrate a balance between binding affinity and antigen-independent signaling (determined by length of the linker between fragment variable regions) in determining CAR function against CD22. We show that CAR function against both CD22^Lo and WT leukemia can be augmented by boosting binding affinity without shortening the linker to induce tonic signaling, establishing rational combinatorial modification of antigen binding domain as an important approach for modulating the function of cellular therapeutics.

Graphical Abstract

CD22-CAR T cells are highly effective but necessitate a short scFv linker for clinical activity. A short linker mediates antigen-independent signaling, which may drive T cell dysfunction and toxicities. Increasing CAR binding affinity eliminates the need for tonic signaling and augments CAR function, demonstrating potential to improve future patient outcomes.

INTRODUCTION

T cells bearing chimeric antigen receptors targeting CD19 (CD19-CAR) have been highly successful in treating relapsed and/or treatment-refractory B cell malignancies. CD19-CAR therapy induces remissions in the majority of patients with B cell acute lymphoblastic leukemia (B-ALL)¹. However, approximately 50% of patients relapse, commonly due to CD19 loss or downregulation via multiple mechanisms². Although initial response rates to CD19-CAR therapy can be higher than 70% in the case of B cell lymphomas, only 50% of patients achieve a complete remission (CR)². Therefore, targeting additional B-lineage antigens as salvage following prior CAR T cell infusions or in multi-targeting modalities has the potential to further improve patient outcomes. While multiple CD19-CAR products have now been FDA-approved, approval for CARs targeting additional antigens have been limited to B cell maturation antigen (BCMA) for multiple myeloma¹. Like CD19, CD22 has B lineage-restricted expression and is expressed across a range of B cell malignancies including ALL and lymphoma. Indeed, CD22-targeted CAR T cells (CD22-CAR) have demonstrated promising results in clinical trials for patients with relapsed and/or treatment-refractory B-ALL and lymphoma, including those resistant to CD19-targeted CAR therapy. However, recent follow-up data from the ongoing CD22-CAR trial at the NCI indicates that, among patients who initially achieved remission, relapse-free survival is only 25% at 10 months post-CAR without consolidative hematopoietic stem cell transplant (HSCT), and relapse with ALL blasts expressing low levels of CD22 (CD22^Lo) is common. Furthermore, patients who achieved minimal residual disease (MRD)-negative CR had a significantly higher pre-treatment leukemic CD22 expression than patients who did not ^3–5. Together, these findings indicate that downmodulation of CD22 expression is a mechanism of leukemic resistance to CD22-CAR therapy. Multiple studies with CARs targeting different hematologic target antigens, including CD19, CD20 and CD22 and solid tumor antigens like ALK and GPC2 demonstrate that antigen downregulation occurs with CARs targeting a broad array of antigens ^4,6–9. These findings highlight the importance of developing strategies to improve both the sensitivity and durability of CAR efficacy in response to challenging target antigens such as CD22^3–5. While direct modulation of target antigen density has proven to be a feasible strategy to boost efficacy of the CD22-CAR in preclinical models⁴, modifying structural components of the CAR such as antigen binding domain may be a more universally feasible strategy to boost CAR responses.

A component of CAR T cell biology highly linked to antigen binding domain is the intensity of constitutive antigen-independent signaling in the T cell, termed “tonic signaling”. Prior work demonstrated that an affinity-matured anti-GD2 binding domain drove strong tonic signaling, producing a functional, phenotypic and transcriptomic profile consistent with T cell exhaustion, which was improved by incorporating a 4-1BB costimulatory domain rather than CD28¹⁰. In contrast, tonic signaling was found to be essential for clinical responses in the context of the CD22-CAR. The NCI trial was the first to test the CD22-CAR and saw a complete remission rate exceeding 70%³. Trials at other institutions used nearly identical CARs with the same antigen binding domain from the m971 antibody clone but replaced the short Glycine-Serine linker (G₄S)_x1 between antibody heavy and light chains with a more standard “long” linker consisting of 3 (G₄S) motifs. Despite the minimal difference in CAR construct, survival of patients in these trials was drastically different, with the long linker CAR inducing complete remissions in only about 50% of patients, depending on the trial ^3,11,12. Comparative laboratory studies found that the short linker significantly alters CAR biology, facilitating antigen-independent clustering of the chimeric receptor on the T cell surface, increasing tonic signaling, and increasing T cell activation as compared to the long linker CAR, seemingly augmenting T cell functionality. However, in other studies, the short linker CD22-CAR showed a similar phenotypic profile to the highly dysfunctional anti-GD2 CAR during in vitro expansion, indicating potential for T cell dysfunction in the long term (albeit with partial “rescue” via 4-1BB costimulation)¹³,¹². While adjustment of binding affinity has also been explored as a method to enhance CAR efficacy, functional effect has been highly dependent on CAR construct and target antigen biology^{14 15–17}.

Given the importance of improving CD22-CAR responses and ample evidence linking antigen binding domain design to CAR function and antitumor efficacy, we sought to rationally modify the antigen binding domain to understand whether we could maintain antigen sensitivity while removing the requirement for tonic signaling. To uncouple the impacts of CAR affinity and tonic signaling, we generated CARs with the standard affinity (SA) m971 binder, or an m971-derived affinity matured (HA) binding domain, bearing either short linker (SL) or long linker (LL). We reveal that the standard affinity/long linker CAR (termed SA-LL), shown to be less clinically effective, has equivalent in vitro and in vivo function against WT leukemia to the standard affinity/short linker CAR (SA-SL), but specifically lacks function against CD22^Lo leukemia. Based on these findings, we hypothesized that for CD22-CAR, the binding affinity could be increased to specifically boost response to antigen-low leukemia, and that a long linker could be incorporated to reduce tonic-signaling-mediated dysfunction. We found that CARs incorporating the HA binder with either SL or LL format showed consistent improvement against CD22^Lo leukemia. However, at a limiting CAR+ cell dose (mimicking high tumor burden), the HA-LL CAR improved clearance of WT leukemia relative to the SA-SL, SA-LL or HA-SL CAR, while extending survival at a higher CAR dose, indicating that a LL may allow for resistance to dysfunction relative to tonically signaling SL CARs. These findings underscore the influence of multiple properties of antigen binding domain on anti-leukemic efficacy in the setting of clinically relevant relapse modalities and indicate that an increased affinity binder abrogates the requirement for tonic signaling in the CD22 CAR, allowing for resistance to dysfunction at a low CAR dose, while maintaining the ability to respond to antigen-low leukemia.

RESULTS

Antigen density impacts in vitro cytotoxicity and activation and in vivo leukemia clearance of CD22-CAR T cells

Prior studies by our lab and others have shown that target antigen density can have a profound impact on CARs targeting a variety of antigens, reducing in vitro effector functions and in vivo anti-tumor efficacy^4,6–9. We set out to explore the impacts of antigen density on T cells expressing a clinically validated CD22-CAR (SA-SL) using Nalm6 clones engineered to express varying levels of CD22 (Figure S1A)³. In vivo, the standard CAR almost entirely cleared parental Nalm6 leukemia by day 8; however, while the CAR did show some efficacy in delaying the progression of CD22^Lo leukemia relative to CD22^Neg leukemia, the CD22^Lo leukemia continued to progress (Figures S1B–C). To further interrogate the impacts of CD22 density on CAR T cell functions, we tested this CAR in multiple in vitro assays across CD22 antigen densities. CAR T cell degranulation was directly impacted by antigen density, both in assays measuring bulk T cell function (Granzyme B ELISA) and on a single cell basis (flow cytometry for CD107a expression) (Figures S1D–E). Additionally, classical cell-surface markers of T cell activation including CD69 and CD25 were upregulated to a lesser extent in CARs co-cultured with CD22^Lo leukemia as compared to WT or CD22^Hi variants (Figures S1F–G). Together, these data demonstrate that low target antigen density negatively impacts multiple facets of in vitro and in vivo anti-CD22 CAR T cell activation and anti-tumor functions.

Short linker boosts function of standard affinity CAR T cells in the setting of CD22^Lo leukemia

Recently, several clinical trials showed disparate results between CD22-targeted CAR T cells used to target B-cell malignancies, with one of the major distinctions being only in the length of the linker between the heavy and light chain in the CAR antigen-binding domain^3,11,12. Given the impact of target antigen density on anti-CD22 CAR T cell function, and the reported mechanism of CD22-downregulation after CAR treatment, we interrogate the role of linker length on CAR T cell function in the context of both WT and CD22^Lo leukemia, with the prediction that a short linker would enhance CAR T cell function, particularly in the setting of low target antigen density. To test our hypothesis, we used a CAR differing only in the length of the linker between the heavy and light chains of the m971 scFv, using a longer (G₄S)_x3 flexible linker rather than the (G₄S)_x1 linker found in the original CAR, and similar to the (G₄S)_x4 linker CAR previously reported to have reduced function against high-antigen expressing tumors¹⁴. We adopted the naming scheme “standard affinity/short linker” (SA-SL) for the original CAR, and “standard affinity/long linker” (SA-LL) for the long linker variant. In vitro, we found that while the linker length had minimal impact on the proportions of cells producing IFNg or IL-2 in response to WT Nalm6 leukemia, the SA-SL CAR showed enhanced response against CD22^Lo Nalm6 (Figures 1A–B). While linker length showed no significant impact on in vivo tumor clearance or long-term survival of mice bearing parental Nalm6, SA-SL CAR T cells mediated enhanced leukemia clearance and survival of mice bearing Nalm6-CD22^Lo (Figures 1C–E). Therefore, we demonstrate that despite similar performance against leukemia with high antigen density, the short linker is necessary for enhanced in vitro and in vivo function against CD22^Lo leukemia, with proportions of CAR T cells producing cytokine in vitro correlating with in vivo efficacy against CD22^Lo leukemia.

Figure 1: A short linker boosts function of standard affinity CAR T cells in the setting of CD22^Lo but not WT leukemia.

1A: Flow cytometry plots showing IL-2 by IFNg production after 6 hour coculture of the indicated CD22-CAR T cell with the indicated leukemia. 1B: Quantification of cytokine data in A. 1C: Schematic: Timeline for in vivo experiment. NSG mice were injected with 1e6 indicated Nalm6 leukemia on day −3, followed by 4e6 CD22-CAR T cells on day 0. Bioluminescent imaging was performed before CAR dosing on day −1, and biweekly post-CAR injection. Mice were monitored for survival. 1D: Quantification of bioluminescence data against WT leukemia from C. 1E: Survival of mice bearing WT leukemia. 1F: Quantification of bioluminescence data against CD22^Lo leukemia from C. 1G: Survival of mice bearing CD22^Lo leukemia. All in vitro assays performed with n=3 technical replicates and are representative of two experiments with two independent donors. In vivo experiment was performed with n=5 mice per group and is representative of two experiments with two independent donors. Data represent mean +/− SD. * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001.

Differential tonic signaling profiles in CD22-CAR T cells are dictated by linker length

Functional disparities in the short and long linker m971-based CARs have been attributed to antigen-independent clustering which drives tonic signaling of the SA-SL CAR at baseline, characteristics not present in the SA-LL CAR¹². In Figure 1, we showed that in the context of the CD22-CAR a short linker enhances anti-tumor responses against CD22^Lo leukemia. However, the short linker CAR has been shown to drive tonic signaling¹², a feature shared with the tonically signaling GD2-CAR, which drives T cell dysfunction in the long term and has a basal transcriptional program consistent with T cell exhaustion. These characteristics of the GD2-CAR were accompanied by a basal phenotypic state characterized by expression of multiple coinhibitory receptors, a profile shared by the CD22-CAR¹³. We hypothesized that a higher affinity antigen binding domain could maintain anti-tumor efficacy against CD22^Lo leukemia while mitigating the need a short linker, allowing for enhanced potency and resistance to dysfunction in the setting of parental leukemia. We previously used error prone PCR mutagenesis and yeast display followed by serial bio-panning over immobilized CD22 to produce a m971-derived scFv with significantly higher affinity and validated a CAR construct with this binder⁴. However, the high affinity CAR previously tested contained a more standardized long (G₄S)_x3 linker (“HA-LL”) rather than the short (G₄S)_x1 linker found in the original m971 CAR, a disparity potentially confounding the comparison of these two constructs. To explore the interplay between affinity and linker length, we generated the final combination of a high affinity binder with a short linker (“HA-SL”). To validate the enhanced affinity of the HA CARs relative to the SA, we transduced all four CARs into T cells (Figure S2A) and performed a direct flow-based antigen binding assay with fluorophore-conjugated CD22 protein Fc (Figure S2B). Indeed, both high affinity CARs showed increased affinity relative to the standard affinity CAR, as indicated by increased GMFI at low antigen concentrations. Both long linker CARs also showed slightly increased affinity relative to their short linker counterparts (Figure S2B).

To define differential signaling properties between CARs (SA-SL, SA-LL, HA-SL and HA-LL), we performed a phospho-peptide array, assaying common mediators of antigen receptor signaling in T cells. We first compared SA and HA CAR T cells at rest to validate previously observed tonic signaling properties dictated by linker length^3,11,12 and determine whether these properties were altered in CARs bearing the affinity matured binder. As predicted, the SA-SL CAR showed higher levels of basal phosphorylation in many key TCR signaling pathways, including calcium (CREB), AKT (GSK3), mitogen activated-protein kinase (RSK1/2, p38) and key signaling hubs (PLCg). Interestingly, the SA-LL CAR exhibited heightened STAT3 and HSP60 phosphorylation, unrelated to TCR signaling (Figure 2A). The HA-SL CAR had many of the same patterns of basal signaling as the SA-SL CAR (CREB, GSK3, p38, PLCg), as well as distinct changes in other kinases (WNK1), with increased STAT3 and HSP60 activity in the HA-LL CAR, similar to SA-LL (Figure 2B). Together, these data establish that a short linker mediates tonic signaling regardless of CAR affinity, driving baseline activation of key TCR signaling pathways, with some distinctions dictated by binder affinity.

Figure 2: A short linker drives tonic signaling in resting T cells, irrespective of CAR antigen binding affinity.

2A,B: Phospho-peptide array performed in resting CAR T cells with either standard affinity (A) or high affinity (B) binding affinity. Lists in (B) and (C) ranked by delta between SL and LL CAR for each plot, then averaged across all four CARs. Vertical line in each plot is the average for signal across all four CARs for the whole array. Each point represents the average of two technical replicates on the array, with each donor (n=2) plotted as a separate point.

Differential functional profiles and signaling profiles in CD22-CAR T cells upon antigen stimulation are dictated by binding affinity and linker length.

We next profiled CARs after stimulation to determine how linker and affinity modulate cytotoxicity and signaling upon antigen engagement. Across multiple donors and against WT and CD22^Lo leukemia, the HA-LL CAR showed significantly enhanced cytotoxicity profile relative to other CARs (Figure S2B,C). Profiling CAR signaling upon antigen stimulation, we found that the SA-LL CAR generally showed weak activation, consistent with poor function observed in Figure 1, and the clinical efficacy profile^3,11,12. The HA-SL CAR drove very strong signaling downstream of multiple TCR pathways (p70-S6, C-JUN). Interestingly, this CAR showed reduced signal in LCK and RSK1/2, potentially a result of negative regulation mechanisms, a feature partially shared with the SA-LL CAR. For both CAR affinities, short linker CARs broadly showed stronger activation of multiple phosphoproteins (STAT3, RSK1/2/3, CREB) than long linker variants. However, both HA CARs showed stronger activation of CREB than the paired SA linker variants, indicative of affinity-dependence of calcium signaling (Figure S2D).

Overall, these data demonstrate that changes in function and signaling profiles upon antigen stimulation are directed by the combination of binding affinity and linker length, giving a distinct profile for each CAR variant.

HA-LL CAR confers enhanced T cell expansion and in vivo leukemia clearance at low CAR dose

We next tested the performance of CD22-CAR variants in the context of a “stress” CAR dose, mimicking a high tumor burden that may drive a dysfunctional state. At this dose, we found that the SA-SL CAR showed reduced function relative to all other CARs, with minimal tumor control, while the SA-LL and HA-SL CARs displayed early activity followed by leukemic relapse. By day 24, only the HA-LL CAR showed consistent prolonged leukemia clearance in all mice (Figure 3A–C, S3A). Upon profiling of the bone marrow, we observed increased expansion of both HA CARs, dominated by the CD4+ CAR T cell pool which has been tied to long-term remissions in CD19 CAR patients¹⁸ (Figure 3D). Despite this, all CARs showed similar profiles of phenotypic exhaustion (Figure 3G–I, S3B–D). Overall, these data demonstrate that at a stress dose, the HA-LL CAR alone maintains a combination of extended leukemic clearance and strong accumulation of CAR+ cells.

Figure 3: At stress dose, all CARs mediate enhanced tumor clearance relative to the SA-SL, with the HA-LL CAR showing durable response and accumulation of CAR+ cells.

3A: Schematic: Timeline for in vivo experiment. NSG mice were injected with 1e6 WT Nalm6 leukemia on day −3, followed by 1e6 of indicated CD22-CAR T cells on day 0. Bioluminescent imaging was performed before CAR dosing on day −1, and biweekly post-CAR. 3B,C: Quantification of average bioluminescence data for each group (B) and each mouse (C) in A. 3D-F: Flow cytometry quantification of CAR T cell proportions in bone marrow for total CAR+ cells (D) and CD4+ (E) and CD8+ (F) CAR+ cells. 3G-I: Flow cytometry quantification of PD1+/TIM3+ cells (left) or PD1+/LAG3+ cells (right) for total CAR+ cells (D) and CD4+ (E) and CD8+ (F) CAR+ cells. In vivo experiment was performed with n=5 mice per group and is representative of two experiments with two independent donors. One mouse in the HA-LL group excluded from flow cytometry analyses after death for reasons unrelated to leukemia burden. Data represent mean +/− SD. * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001.

A long linker enhances cytokine production by the high affinity CD22-CAR against CD22^Lo leukemia and abrogates persistent expansion.

In Figures 1 and 3, we demonstrated that the SA-LL CAR showed diminished in vitro functionality and in vivo activity against CD22^Lo leukemia relative to the SA-SL CAR, and lower accumulation in vivo against WT leukemia in a comparison with all 4 CARs, consistent with inferior efficacy relative to SA-SL in multiple clinical trials^3,11,12. Given this diminished functional profile, and the similar tonic signaling profile driven by a short linker in both SA and HA CARs (Figure 2), we chose to benchmark the HA CARs against the clinically active SA-SL CAR going forward.

To further validate in vitro functionality of the affinity-matured CARs, we utilized the same in vitro intracellular cytokine staining (ICCS) assay shown to associate with in vivo CAR T cell efficacy in standard affinity CAR T cells (Figure 1). All three CARs showed function against the CD22 target antigen. Against WT leukemia we observed that the SA-SL CAR had higher proportions of cells producing IL-2, IFNg, or both cytokines as compared to either high affinity CAR (Figures S4A–C). However, against CD22^Lo leukemia, the HA-LL CAR had significantly higher proportions of CAR+ cells producing cytokines as compared to HA-SL or SA-SL CARs (Figures S4D–F). Interestingly, the HA-LL CAR also had much greater preservation of proportions of cells producing cytokines with the drop in antigen density, as compared to the other two CARs. All CARs had very low background cytokine production against CD22^Neg leukemia, confirming the antigen-specific activity of these receptors (Figures S4G–H). To compare in vitro expansion potential of the CARs, we used a coculture assay with an internal fluorescent counting bead control in each well to quantify precise ratios of CAR:Beads and Leukemia:Beads. All three CARs expanded and cleared leukemia equivalently against WT Nalm6 (Figure S4I–J). Interestingly, while all three CARs also cleared CD22^Lo leukemia equivalently, both CARs with short linkers continued to expand after leukemia clearance, while the HA-LL CAR contracted post-antigen clearance, similar to a standard T cell response (Figure S4K–L). As the HA-LL CAR had been previously validated⁴, these data established that the HA-SL CAR was functional in the high affinity format and provided rationale for comparing all three CARs in vivo. Additionally, we demonstrate that in contrast to the standard affinity CARs compared in Figure 1, a long linker mediates superior in vitro cytokine production against CD22^Lo leukemia for the high affinity binder.

HA-LL CAR extends survival against WT leukemia at high CAR+ cell dose

We next wanted to test CD22-CAR performance at a higher, more therapeutic dosing level to determine whether leukemia clearance or long-term survival would be impacted. While early in vivo tumor burden of WT Nalm6 treated with a 4e6 CAR+ cell dose was highly variable (Figures 4A,B), long-term survival was significantly extended in mice treated with HA-LL CAR T cells (Figure 4C). In vivo profiling at an intermediate “limiting” dose of 2e6 CAR+ cells showed similar but less distinct trends in leukemia clearance to the previous “stress” dose experiment, further solidifying the important role of dosing studies in preclinical testing (Figure S5). We demonstrate here that at higher CAR+ cell doses, the HA-LL CAR shows superior long-term leukemia clearance and extended survival in mice bearing high-antigen expressing WT leukemia, consistent with an enhanced ability to mediate leukemia clearance in the setting of high tumor burden at lower “stress” dose.

Figure 4: HA-LL CAR extends survival of mice bearing WT leukemia at high CAR+ cell dose.

4A: Schematic: Timeline for in vivo experiment. NSG mice were injected with 1e6 WT Nalm6 leukemia on day −3, followed by 4e6 of indicated CD22-CAR T cells on day 0. Bioluminescent imaging was performed before CAR dosing on day −1, and biweekly post-CAR. 4B: Quantification of average bioluminescence data for each group in D. 4C: Survival of mice treated with 4e6 of the indicated CAR T cells. In vivo assay performed with n=5 mice per group, 3 experiments with 3 independent donors. 4A and 4B are representative data from one experiment. 4C is pooled data, SA-SL (n=15), HA-SL (n=15), HA-LL (n=15). Data represent mean +/− SD. * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001.

HA-LL CAR confers enhanced in vivo leukemia clearance and extends survival against CD22^Lo leukemia

Based on our results in Figure 1, SA-LL CAR T cells showed reduced function relative to SA-SL in the setting of CD22^Lo leukemia, a known clinical relapse mechanism in the setting of the CD22 target antigen, which could contribute to the poor clinical outcomes seen with long linker CARs³. To test our hypothesis that boosting CAR affinity could eliminate the requirement for a short linker in driving in vivo CAR T cell function in response to leukemia with low levels of target antigen, we tested short and long linker high affinity CARs (HA-SL and HA-LL) against the SA-SL CAR in our xenograft model against CD22^Lo leukemia (Figure 5A). Across multiple T cell donors, we found that both high affinity CARs consistently outperformed the SA-SL CAR, mediating enhanced tumor clearance by bioluminescence imaging and flow cytometry of bone marrow at day 17-19 post CAR treatment. Interestingly, in contrast to the standard affinity CAR, this was not dependent on the linker length, and both high affinity CARs performed equivalently (Figure 5A–C). Despite consistently higher leukemia burden, the SA-SL CAR showed the highest proportions of CAR+ cells in the marrow, indicating that the remaining CAR T cell population likely showed lower functional potential on a cell-by-cell basis (Figure 5D–E). While both high affinity CARs performed similarly in early leukemia clearance, only the HA-LL CAR significantly extended survival against CD22^Lo leukemia relative to the SA-SL CAR (Figure 5F). These data demonstrate that affinity maturation of antigen binding domain, combined with a long linker, enhances CAR T cell function against CD22^Lo leukemia. These results contrast with standard affinity CARs, which necessitate a short linker for function against CD22^Lo leukemia (Figure 1).

Figure 5: HA-LL CAR confers enhanced in vivo leukemia clearance and extends survival against CD22^Lo leukemia.

5A: Schematic: Timeline for in vivo experiment. NSG mice were injected with 1e6 CD22^Lo Nalm6 leukemia on day −3, followed by 4e6 of indicated CD22-CAR T cells on day 0. Bioluminescent imaging was performed before CAR dosing on day −1, and biweekly post-CAR. 5B: Quantification of average bioluminescence data for each group in A. 5C: Quantification of individual bioluminescence data for each group in A. For 4D to 4E, bone marrow was analyzed by flow cytometry at day 18 post-CAR for indicated cell population. 5D: % CAR+ of live marrow. 5E: % leukemia of live marrow. 5F: Survival of mice treated with 4e6 of indicated CAR T cells. Bioluminescence data is representative of four experiments with 4 independent donors. Survival is pooled from 3 experiments performed identically with independent donors. Some sets of experiments compared only some of the CARs and all data was pooled for analysis with the following total mice per group: Mock (n=10), SA-SL (n=15), HA-SL (n=10), HA-LL (n=15). Statistics in 5F compare HA-LL and SA-SL groups. Data represent mean +/− SD. * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001.

Single cell transcriptomic profiles in CD22-CAR T cells show characteristics dependent on affinity and linker length

To further examine the biology of CD22-CARs during in vitro and in vivo responses to leukemia (Figures 3–5), and tonic signaling profiles observed at rest (Figure 2), we performed single cell transcriptomic profiling, capturing several thousand CD22-CAR T cells at rest or upon antigen restimulation in vitro (Figure S6A). As predicted, we observed discrete clusters associated with cells at rest (“Rest”, clusters 2-8) versus cells that were stimulated (“Stim”, clusters 0 and 1), with distinct cluster-associated proliferative signatures and gene expression (Figure 6A, B, S6B). All CARs showed a similar cluster distribution at rest, with slight differences in the HA-LL CAR, likely dictated by the long linker (Figure 6B,C, S6B,C). However, upon stimulation, both HA CARs showed a drastic enrichment of cells within stim-associated clusters (0, 1), while the SA-SL CAR had a greater proportion of cells remaining in the rest-associated clusters (2, 4, 5), indicative of a dynamic transcriptional profile and broadly increased activation potential driven by the HA CARs upon antigen encounter (Figures 6A–C, S6B,C). Differential gene expression analysis revealed enrichment of a highly activated effector-like transcriptional profile (ZBTB32, GZMB, IL2RA, NKG7, LGALS3) in the HA CARs relative to SA-SL, which expressed an alternative activation profile (HLA-encoded class II molecules, IL23R) with some memory characteristics (TCF7). The HA-LL CAR also drove expression of KLF2, a key transcription factor known to enhance effector differentiation while suppressing terminal exhaustion^19,20. Similar gene expression profiles in the two short linker CARs at rest diverged upon stimulation, with the HA CARs becoming broadly more similar (Figure 6D–F, S6D–E). A greater magnitude of transcriptional changes in CD4+ CAR T cells suggested optimization of CAR architecture for each T cell subset as a future strategy to enhance CD22-CAR efficacy (Figure S6D–E). Overall, we observe dynamic transcriptional profiles driven by differences in CAR antigen binding domain at rest and upon stimulation that corroborate observed differences in functional potential.

Figure 6: Single cell transcriptomics reveals dynamic changes between CARs at rest and upon antigen stimulation.

6A: UMAP plot of clustering of all CD22-CAR T cells from rest and stimulation conditions (left), with defining features and gene expression signatures for each cluster (right). 6B: Contour plot showing density of indicated CD22-CAR T cell groups in different clusters at rest (colored) and upon stimulation (black) for replicate 1. 6C: Frequency (%) of CD22-CAR T cells in each cluster at rest (left) and upon stimulation (“Stim,” right) for each replicate. 6D-E: Significantly differentially expressed genes between CD8+ CD22-CAR T cells at rest (D) and upon stimulation (E) for each replicate. 6F: Violin plots showing a subset of indicated genes, plotted by pooled replicates. Differentially expressed genes are upregulated in at least one CAR condition relative to the others, at rest (left) or upon stimulation (right), pooled replicates. Differentially expressed genes shown are FC>1.5, adjusted p value <0.1.

DISCUSSION

While CD22-CAR T cells have a 70% rate of remission induction in CD19-CAR refractory leukemia and 68% response rate in CD19-CAR refractory lymphoma^21,22, patients commonly relapse with reduced surface expression of the CD22 antigen relative to pre-treatment levels ^2,3. Therefore, while creating a CAR with high sensitivity to target antigen is essential for extending durability of remissions, the clinically effective CD22-CAR has been shown to rely on antigen-independent CAR clustering and tonic signaling for efficacy, a feature which has been shown to negatively impact T cell function in other settings^13,23. This tonic signaling, driven by a short (G₄S)_x1 linker in the scFv, has been essential for strong clinical responses in patients, despite comparable efficacy of m971-based CARs with longer linkers in most preclinical models^{3,11,12 51, 52}. Known tumor relapse modalities post-CAR therapy present a useful opportunity for understanding the biology of these artificial antigen receptors and the T cells bearing them. Here, we present data from experimental systems designed to model the known relapse modalities of antigen downregulation/heterogeneity (leukemia engineered to express lower target antigen density) and T cell dysfunction (administration of a subcurative CAR+ cell dose). We show that while standard affinity CAR T cells with long (G₄S)_x3 linkers can mediate strong responses against high antigen-expressing WT leukemia in xenograft models, a short linker is required for the in vivo response against CD22^Lo leukemia. We build on this finding using a CD22-CAR construct with an antigen binding domain of significantly higher affinity⁴ and test short and long linker versions of this construct compared to short and long linker versions of the standard affinity CAR. We find that a short linker mediates a tonic signaling profile regardless of binding affinity, and demonstrate that the HA CAR does not require tonic signal driven by the short linker for activity. The long linker HA CAR allows for more durable in vivo functionality in settings of limiting CAR dose and against CD22^Lo leukemia, and enhances leukemic clearance across multiple donors relative to the original SA-SL CAR. Finally, we show that both CARs incorporating the HA binder drive dynamic transcriptional changes upon antigen stimulation relative to the original SA-SL CAR. Thus, we have shown that rational modification of the antigen binding domain in CD22-targeted CAR constructs maintains antigen sensitivity required to successfully target the CD22 antigen, while alleviating the requirement for tonic signaling.

Many questions remain regarding characteristics of antigen binding domain and how they impact CAR T cell biology. While phosphoproteomic studies by Singh et al. indicated that linker length also dictated tonic signaling profile of a CD33-targeted CAR, the same was not true for a CAR targeting CD19, accompanying functional studies were not performed, and binding affinity was not considered for these antigens¹². While our findings are consistent, with a short linker driving tonic signaling in CD22-CAR, our study shows that an affinity matured binder derived from the same antibody clone drives a distinct tonic signaling profile, providing a more nuanced study on combinatorial modification of the antigen binding domain. One important avenue for future studies will be broader characterization of the interplay of affinity and linker length in CARs with other receptor modifications such as alternative costimulatory domains, or with other CD22 antigen binding domains such as the YKCD22 binder described by Pan et al.²⁴. Tonic signaling would be predicted to be detrimental in a solid tumor microenvironment characterized by high antigen burden, repetitive stimulation and microenvironmental factors that further contribute to exhaustion. A tonically signaling GD2-targeted CAR has been tested against solid tumors but activity in patients required costimulatory modifications and addition of small molecule tyrosine kinase inhibitors to culture media to mitigate tonic signaling effects^13,23,25. Finally, we and others have previously shown key differences in the biology of memory versus naïve-derived CAR T cells²⁶, and between CD4 and CD8 CAR T cells²⁷, suggesting a differential role for antigen receptor signal strength between canonical T cell subgroups and differential effects of a tonically signaling receptor, which we observed in our stress dosing and transcriptomic datasets.

Recent findings have shown that the CD22-CAR can cause a severe hyperinflammatory syndrome toxicity similar to classically defined hemophagocytic lymphohistiocytosis (HLH), now referred to as immune effector cell hemophagocytic syndrome (IEC-HS) which occurs at a higher frequency in the SA-SL CAR relative to non-tonically signaling CD19 CAR constructs^5,28. Although the pathophysiology of this toxicity is not fully established, we have previously demonstrated a link to IFNg, IL-18 and IL-1, similar to HLH occurring in other settings ^29,30. Pre-clinical models demonstrate that sustained T cell activation contribute to HLH ³¹. Thus, a plausible hypothesis is that IEC-HS associated with CD22-targeting CARs may be mitigated by using a long-linker CAR more similar to the CD19-CAR, such as the lower-tonic signaling HA-LL CAR described here. Although we did not directly test exact mechanisms contributing to IEC-HS, we observe that the HA-LL CAR rectifies aberrant expansion and delayed contraction, which were implicated in a recent clinical correlative study as drivers of IEC-HS in CD22-CAR patients treated with the same SA-SL CAR that showed these characteristics in our studies²⁸. Intriguingly we also observe reduced cytokine production with the HA-LL CAR against WT leukemia. Further studies are warranted to fully elucidate whether a tonic signaling CAR drives HLH-like toxicity in an immune-intact host.

In conclusion, our work here has tested a non-tonic signaling affinity-matured CD22 CAR based on the same clinically validated m971 antibody clone. Targeting low antigen levels is a requirement for CD22-CARs, and necessary in preclinical evaluations. The SA-SL CAR, which incorporates a short linker to afford antigen sensitivity, demonstrated activity in the clinic. However, recent clinical results (NCT05972720) found that a highly similar CAR is likely to drive unacceptable high-grade toxicities. The HA-LL CAR described here has the potential to directly improve outcomes of patients with B-lineage malignancies by resisting the known relapse mechanisms of antigen downregulation and T cell dysfunction during exposure to high tumor burden with lower tonic signaling, potentially circumventing toxicities associated with the current CD22 CAR.

CARs are highly modular molecules and have been readily modified, with a primary focus on signaling components, to target an extensive set of antigens. Our evaluation as to how changes in antigen binding affinity and tonic signaling impact CAR T cell biology may be clinically applicable to CARs targeting other tumor types. As CAR T cells are now being used to treat SLE, multiple sclerosis, aging/cellular senescence, and to target virally expressed surface proteins (HIV)³², these findings are potentially applicable toward the design of future CAR variants to mitigate a broad array of diseases.

MATERIALS AND METHODS

Culture of cell lines and preparation of human donor T cells

The parental Nalm6 (GFP/Luciferase-transduced) B-ALL cell line was obtained from Dr. Crystal Mackall, Pediatric Oncology Branch, NCI, NIH, Bethesda, MD, 2008. Nalm6 CD22 site density model cell lines (CD22^Neg, CD22^Lo, CD22^Hi) were previously generated as described ³. Briefly, CD22 was knocked out using CRISPR/Cas9, and a CD22 transgene was subsequently reintroduced using lentiviral transduction. Following transduction with CD22, single-cell cloning by limiting dilution was used to generate Nalm6 clones which stably express varying levels of CD22. CD22 site density was (antigen sites/cell) on Nalm6 clones was previously quantified using CD22-PE, and CD22 MFI was converted to sites/cell using QuantiBrite PE Beads (BD Biosciences, Cat No. 340495) as previously described ³. Nalm6 WT and all CD22 site density clones were cultured in complete RPMI-1640 (cRPMI), RPMI-1640 medium (Gibco, Cat No. 11875119) supplemented with 10% heat-inactivated FBS (Omega Scientific, Cat No. FB-12) and 2mmol/L L-glutamine (Gibco, Cat No. 35050079), at 37C and 8% CO₂. The viral producer Lenti-X cell line was obtained from Takara Bio (Cat No. 632180). Lenti-X cells were cultured in DMEM medium (Gibco, Cat No. 11965118) supplemented with 10% heat-inactivated FBS, 2mmol/L L-glutamine, and 10mmol/L HEPES (Gibco, Cat No. 15630130), and were passaged at least once prior to transfection with lentiviral plasmids. All cell lines were routinely tested for Mycoplasma by the University of Colorado Anschutz core facility.

Generation of human CD22 CAR constructs

For ease of cloning CAR variants, modular monovalent second-generation CAR constructs (4-1BB costimulatory domain with CD8 transmembrane) were previously synthesized (GeneArt, Thermo Fisher) and cloned into a lentiviral plasmid backbone. These constructs each incorporated restriction sites (XhoI and SpeI) flanking the scFv region of the CAR to allow for rapid synthesis of CAR constructs. In the present study, variants of the m971 scFv were synthesized with flanking XhoI and SpeI restriction sites (GeneArt, Thermo Fisher) and subcloned into the CAR backbone using standard restriction enzyme cloning with XhoI and SpeI enzymes (New England BioLabs). CD22 CAR sequences were verified by Sanger sequencing (Eton Biosciences).

Generation of human CD22 CAR T cells

Healthy human donor whole blood obtained from the Children’s Hospital Colorado Blood Donor Center, under an institutional board-approved protocol. Peripheral blood mononuclear cells were isolated using Lymphocyte Separation Medium (Corning) according to manufacturer protocol. T cells were then purified using an EasySep Human T Cell Isolation Kit (StemCell, Cat No. 19051) and cryopreserved in 90% heat-inactivated FBS and 10% DMSO. Lentiviral vectors encoding CD22 CAR constructs were generated by transient transfection of the Lenti-X viral producer cell line with the CD22 CAR construct, as well as lentiviral packaging and envelope plasmids (pMDLg/pRRE, pMD.2G, and pRSV-Rev, all obtained from Addgene) using Lipofectamine-3000 (ThermoFisher) in OptiMEM (Gibco). 6 hours following transfection, medium was replaced with fresh media. At 24 hours and 52 hours following transfection, supernatant containing lentiviral vector was harvested and spun at 3000 RPM for 10 minutes to remove cell debris and frozen at −80C for later use. Healthy human T cells were thawed at 1e6/mL in human T cell expansion media (hTCEM), AIMV medium (Gibco, Cat No. 12055091), supplemented with 5% heat-inactivated FBS, 2mmol/L L-glutamine, 10mmol/L HEPES, and 40 IU/ml recombinant human IL-2 (R&D Systems, Cat No. 202-IL-050), and activated for 48 hours with CD3/CD28 Human T Activator Dynabeads (Gibco, Cat No. 11132D) at a ratio of 1:1 cells to beads. On days 2 and 3, T cells were transduced with lentivirus in the presence of 10 ug/mL protamine sulfate (Sigma, Cat No. P4020-10G) and 40 IU/mL rhIL-2, spinning for 2 hours at 1000xG. On day 4, CD3/CD28 Human T activator microbeads were removed using a magnetic rack and T cells were resuspended at 0.5e6/mL in hTCEM. T cells were expanded for 5-6 more days, with new media and cytokines added every other day. Following expansion, transduction efficiency of the CD22-CAR was evaluated by flow cytometry staining with CD22-Protein Fc (R&D Systems) and CAR T cells were cryopreserved or used immediately for in vitro or in vivo assays.

Nalm6 xenograft in vivo studies

All xenograft studies were performed using NSG mice (NOD-scid-gamma, NOD.Cg-Prkdc^scidIl2rgtm1Wjl/SzJ; Strain #: 005557, Jackson Laboratories). NSG mice received 1e6 GFP/Luciferase-positive leukemia cells (22^Neg, 22^Lo, or WT) via intravenous tail vein injection on Day −3. Mice then received CD22 CAR T cells via intravenous tail vein injection on Day 0 at the indicated dose level. Leukemia burden was measured twice weekly using Xenogen In Vivo Imaging System (IVIS) 200 or Spectrum (Caliper Life Sciences) after intraperitoneal injection with D-Luciferin (Perkin Elmer). Total flux (photons/s) was measured for each mouse using Living Image 4.7 (Caliper Life Sciences). All animal studies were conducted in accordance with an animal protocol approved by the University of Colorado Anschutz Medical Campus Institutional Animal Care and Use Committee (IACUC, Protocol 751).

Flow cytometry for cell-surface and intracellular markers

Flow cytometric analysis of cell-surface and intracellular proteins was performed on a BD LSR Fortessa X-20 Flow Cytometer (BD Biosciences). For all CD22 CAR constructs, the CAR was detected by primary staining with a recombinant CD22-Fc chimera protein (R&D Systems) followed by secondary staining with PE-Conjugated Goat anti-Human IgG Fc secondary antibody (Thermo Fisher), or with CD22-Fc pre-conjugated to AF647 (R&D Systems). Cells were stained in FACS Buffer consisting of PBS (Gibco, Cat No. 14190250) with 3mM EDTA (ThermoFisher, Cat No. 15575020) and 1% Bovine Serum Albumin (Sigma-Aldrich, Cat No. A2153-500G) Intracellular cytokine staining was performed using the Cytofix/Cytoperm Kit (BD Biosciences, Cat No. 554714) with manufacturer protocols. Dead cells were excluded using eBioscience Fixable Viability Dye (Thermo Fisher/Invitrogen). Leukemia was identified using GFP.

Phospho-peptide array

6-well plates were coated overnight using CD22 Protein Fc (3ug/mL in PBS) or PBS alone (no antigen). For stimulation, CAR T cells were thawed and allowed to rest in 100IU/mL IL-2 overnight, then cultured for 18 hours with or without plate-bound antigen. Samples were collected and protein lysates prepared according to manufacturer protocols for analysis with the Proteome Profiler Human Phospho-Kinase Array Kit (R&D Systems, Catalog #: ARY003C). Samples were assayed in duplicate, and phospho-peptide quantity was quantified using ImageJ by calculating an inverted pixel density and then subtracting the negative control (PBS) as the background for each sample value. An inverse hyperbolic sine (asinh) transformation was used when plotting fold change activation/baseline data to increase dynamic range in the heatmap without excluding negative datapoints.

Single cell RNA-sequencing

Following CAR transduction and removal from beads, T cells were magnetically selected using unconjugated CD22 Protein Fc to obtain >98% pure CAR+ T cells and frozen down after 5 days of additional expansion in IL-2 as described above. 24-well plates were coated overnight using CD22 Protein Fc (3ug/mL in PBS) or PBS alone (no antigen)³³. For stimulation, T cells were thawed, and allowed to rest in 100IU/mL IL-2 overnight, then cultured for 24 hours with or without plate-bound antigen. At the end of the assay, dead cells were depleted using a Dead Cell Removal Kit (Miltenyi Biotec, Cat #130-090-101) to achieve greater than 90% viability, and cells were resuspended at 1.25e6/mL. Following manufacturer recommendations, emulsion droplets were constructed using 10X Genomics Chromium X controller with the 3’ scRNA-seq On-Chip Multiplexing (OCM) workflow to target 5,000 single cells per condition. Gene expression library preparation was performed using the 10X Genomics Chromium Single Cell OCM 3′ Reagent Kit. Quality control on the generated cDNA was performed using an Agilent 2100 Bioanalyzer using a High Sensitivity DNA Kit (Agilent Technologies). One quarter of the cDNA was used to build the gene expression libraries, and indexed using the 10X, Set TT A index plate. Gene Expression libraries were pooled based off the projected number of cells captured and sequenced using a NextSeq 2000 P4 100c flow cell, aiming for 20,000 reads per sample. Two experimental replicates were performed.

Reads were demultiplexed and mapped to the human genome (GRCh38) using 10X cellranger (v9.0.1) before analysis with Seurat (v5.3.0)³⁴. Cells with fewer than 7% mitochondrial reads and between 1000 and 7000 unique genes were included for subsequent comparisons. Data from cells in both replicates of rest (“Rest”) and stimulation (“Stim”) conditions were log normalized and scaled using all genes before processing with principal component analysis. The first 10 principal components were used to find neighbors, before clustering at a resolution of 0.5, and UMAP dimensional reduction using the 20 nearest neighbors. Subsets of CD4 and CD8 T cells were selected by normalized expression levels of CD4 and CD8A/CD8B. Gene expression comparisons were made between groups of CD4 or CD8 cells expressing different CARs with Seurat function FindMarkers using the MAST test including the replicate as a latent variable and a cutoff of >1.5 fold-change. Z-scores for genes with an adjusted p-value < 0.1 in at least one comparison were generated with SeuratExtend (v1.2.5) function CalcStats³⁵ and plotted as heatmaps with ComplexHeatmap³⁶.

Statistics

ELISA, CD107a, CD69 and CD25 data were compared using ordinary one-way ANOVA with Tukey’s multiple comparisons tests. Relative cell quantities in in vitro leukemia clearance, killing assays and CAR T cell expansion experiments were compared using ordinary two-way ANOVA with Tukey’s multiple comparisons tests. ICCS data were compared using multiple unpaired t-tests with Benjamini, Krieger and Yekutieli procedure to control for false discovery rate, or ordinary one-way ANOVA with Tukey’s multiple comparison tests as appropriate for group numbers. In vivo flow cytometric analysis of % CAR T cells and leukemia burden data were compared using Kruskal-Wallis tests. For in vivo imaging studies, leukemia bioluminescence data were compared using a standard two-way ANOVA with post-hoc Tukey’s multiple comparisons tests. For comparison of survival curves, the Log-rank (Mantel-Cox) test was used, except for the survival data shown in Figure 5F, analyzed using the Cox proportional-hazards model via the coxph() function in R, to account for the fact that some groups only contained 2 of 3 donors. With the exception of the scRNA-seq (detailed above), all other statistical tests were run in Prism 10 software (GraphPad).

DATA AVAILABILITY STATEMENT

All data is available from the corresponding author upon reasonable request. Single cell RNA-sequencing data is uploaded to GEO, under accession number GSE313971.