AdCAR T cells mediate specific lysis of AML cell lines

One major obstacle to CAR T-cell-based immunotherapy in AML is the heterogenous expression profile of target antigens. We analyzed 32 pAML samples at initial-diagnosis for their expression of the target antigens CD33, CD123, and CLL-1. The majority expressed these antigens at high levels, however, we did observe samples in which a minority of cells expressed one of the three antigens (Fig. 1A, Supplementary Fig. S1A and Table 1). Hence, to make CAR T-cell therapy applicable in all AML subtypes and at the same time counteract antigen escape variants, a CAR T platform allowing to address several target antigens either in parallel or sequentially is desirable. We therefore established an AdCAR T-cell platform allowing flexible targeting of various antigens by uncoupling antigen recognition and T-cell activation (Fig. 1B).

Fig. 1: AdCAR T cells mediate specific lysis of AML cell lines.

A Percentage of CD33-, CD123- and CLL-1-positive pAML cells assessed by surface marker staining with biotinylated AMs and subsequent secondary staining (n = 32). B Schematic illustration of the AdCAR T-cell platform recognizing an AML cell via AMs directed against the target antigens CD33, CD123, or CLL-1. C AdCAR characteristics after transduction and 14 days of expansion in IL-7/IL-15. The CD4/CD8 ratio of the AdCAR T-cell product was determined by flow-cytometry (n = 20). The transduction efficiency of AdCAR T cells was measured by biotin-PE staining. A representative contour plot depicts the percentage of AdCAR+ fraction in black (untransduced cells in blue). D Target antigen expression on AML cell lines determined by surface marker staining with biotinylated AMs and subsequent secondary staining (n = 3). MFI ratios were calculated based on corresponding controls without AMs. E AdCAR T-cell-mediated cytotoxicity after 48 h (n = 4–12) against the AML cell lines MV4-11, HL-60, and OCI-AML-3 (E:T = 1:1) in co-cultures containing αCD33-AMFab, αCD123-AMFab, or αCLL-1-AMFab at concentrations ranging from 1 pg/ml to 1000 ng/ml. Target-irrelevant αCD19-AMFab was used as a control AM. Specific lysis was calculated relative to the mock T-cell condition. F Secretion of IFN-γ, TNF, and IL-2, determined by cytometric bead array (CBA) analysis, from corresponding cytotoxicity assays at an AM concentration of 10 ng/ml (n = 3). All graphs present the mean ± SEM. Statistical analysis: paired t-test; ns p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Table 1 Patient characteristics.

Lentiviral transduction of human T cells with the AdCAR construct was consistent and highly efficient (% transduction efficiency: 55 ± SEM), whereas CD4+ T cells were significantly more susceptible to lentiviral transduction than CD8+ T cells (Fig. 1C and Supplementary Fig. S1B).

HD-AdCAR T cells were co-cultured with different AML cell lines (MV4-11, HL-60, OCI-AML- 3) expressing CD33, CD123, and CLL-1 at various levels (Fig. 1D) in the presence of target-antigen-specific Fab- or Ab-based AMs. Specific cytotoxicity was observed against all three AML cell lines and was dependent on target-antigen specificity, antigen density, AM concentration, and E:T ratio (Fig. 1E and Supplementary Fig. S1C–F).

Addition of the control αCD19-AMFab did not result in nonspecific lysis (Fig. 1E), and we found no target-antigen-independent cytotoxicity for the AMs (Supplementary Fig. S1E). AdCAR T-cell activation was measured by the secretion of the effector cytokines IFN-γ, TNF, and IL-2, and was observed only in the presence of target-antigen-specific AMs and corresponding target cells (Fig. 1F).

AdCAR-mediated cytotoxicity against pAML cells: impact of receptor-mediated internalization on AdCAR T-cell efficacy

Next, we assessed whether AdCAR T cells can effectively target and lyse pAML cells. We focused on CD33 as the target antigen and observed high and specific lysis of pAML cells (% specific lysis: 75 ± SEM at 100 ng/ml αCD33-AMFab) after 3 days of co-culture with AdCAR T cells using our previously described pAML culture system [21]. Target elimination was AM dose and E:T ratio dependent (Fig. 2A). Notably, AdCAR T cells generated from HD or pAML T cells were equally effective in vitro (Supplementary Fig. S2C), suggesting the vector system is suitable for clinical applications.

Fig. 2: AdCAR-mediated cytotoxicity against pAML cells: impact of receptor-mediated internalization on AdCAR T-cell efficacy.

A AdCAR T-cell-mediated cytotoxicity after 72 h (n = 3–13) against pAML cells co-cultured on irradiated MS5 feeder cells (E:T = 1:5 and 1:10) in the presence of 100, 10 or 1 ng/ml αCD33-AMFab. αCD19-AMFab at 100 ng/ml served as a negative control. Specific lysis was calculated relative to the AdCAR T-cell condition without AM. B Growth of pAML samples in long-term co-cultures with AdCAR T cells (E:T = 1:10) for 12 days with initial (one-time) addition of 100, 10 or 1 ng/ml αCD33-AMFab. pAML cell counts over time are plotted as normalized target cell count relative to starting conditions on day 0 (n = 4–9). C Levels of receptor-bound αCD33-AMFab were monitored daily for 5 days on MV4-11 cells stained with 500 ng/ml AM. AM was added on day 0 for 15 min at 4 °C, unbound AM was either removed from the supernatant or not (n = 3–4). D Internalization assay. Left: Representative confocal image of MV4-11 cells stained for 6 h at 37 °C with 500 ng/ml αCD33-AMFab coupled to pHrodo Red Avidin (gray). Right: Nuclei were stained with Hoechst 33342 (red) and merged with the pHrodo Red Avidin (green) channel. AM internalization can be seen as puncta located in the cytoplasm. Control conditions at 4 °C did not yield a measurable pHrodo Red Avidin signal (data not shown). Three independent experiments were performed. E Representative example of flow-cytometry-based indirect internalization assay of αCD33-AMFab at 4 °C and 37 °C for 6 h on MV4-11 and OCI-AML-3 cells. AML cells were labeled for 15 min at 4 °C with 500 ng/ml AM. The unbound AM was removed and the percentage of surface-bound AM (red histograms) was assessed at each indicated time point by secondary staining with anti-Biotin-PE antibody. F Quantitative representation of the internalization assay described in E (n = 6). The kinetics of internalization of Fab- and Ab-based AMs were compared on AML cell lines. G The influence of AM internalization on AdCAR T-cell-mediated cytotoxicity (n = 6). AML cell lines were pre-incubated for either 0 or 12 h with 500–1 ng/ml αCD33-AMFab before addition of AdCAR T cells (E:T = 1:1). Cytotoxicity was assessed by flow-cytometry after an additional 48 h. Data are plotted as mean ± SEM. Statistical analysis: A Ordinary one-way ANOVA with Dunnett’s comparison; F Mixed-effects analysis with Geisser–Greenhouse correction. ns p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

However, over an extended period of 12 days, we observed an outgrowth of pAML cells with faster kinetics at lower AM concentrations (Fig. 2B). In line with this, T-cell proliferation peaked at around day 6 (Supplementary Fig. S2A) and the percentage of PD1/TIM3/LAG-3+ T cells declined over time, indicating a loss of T-cell activation (Supplementary Fig. S2B).

Because the AMs were added only at the start of the co-cultures, we speculated that the available AMs were consumed by target antigen receptor-mediated endocytosis. We assessed the surface retention of αCD33-AMFab bound to the CD33 receptors of MV4-11 AML cells, by labeling the cells with AMs in the absence of AdCAR T cells. We observed almost complete loss of receptor-bound AMs within 72 h, with even faster kinetics after prior removal of unbound AM, suggesting that free AM was binding to recycled or newly synthesized antigen receptors (Fig. 2C).

To further validate our findings, αCD33-AMFab was coupled to a pH-sensitive dye. Confocal microscopy confirmed that internalization of AMs occurred rapidly at 37 °C (Fig. 2D). Furthermore, a deficiency of CD33 receptors, as well as blocking vesicle transport with monensin, completely disrupted internalization by MV4-11 cells, hinting convincingly at receptor-mediated endocytosis of the AMs (Supplementary Fig. S2D).

To quantitatively describe this phenomenon, we characterized the internalization kinetics of different AM formats (Fab vs Ab based). We observed a rapid decline of all analyzed AMs within the first 2 h (Fig. 2E, F and Table 2), with kinetics dependent on the AML cell line, AM format, and the target antigen. Interestingly, Fab-based AMs had shorter half-lives than the Ab-based formats, and, after 6 h, only 19%, 8%, and 8% of the initial surface-bound concentrations of αC33-AMFab, αCD123 AMFab (both on MV4-11), and αCLL-1-AMFab (on HL-60), respectively, were detected. To validate that AM internalization also occurs in a clinically relevant setup, we confirmed the results with pAML cells (Supplementary Fig. S2E). Of note, based on the detection of surface-bound AM we cannot exclude AM-dissociation and potentially other elimination pathways, having contributed to the decrease of AM.

Table 2 Adapter molecule internalization kinetics.

To gain a deeper understanding into whether and how AM internalization influences AdCAR T-cell cytotoxicity, we performed co-culture cytotoxicity assays with three AML cell lines expressing high or low target antigen levels. AML cells were labeled with αCD33-AMFab and AdCAR T cells were added 12 h later. Consistent with the internalization study, cytotoxicity was also dependent on AM level, as lysis of target cells declined upon prolonged pre-incubation of AML cells with AMs (Fig. 2G). Importantly, this effect was most pronounced if target antigen densities were low (OCI-AML-3 cell line), leading to almost complete loss of AdCAR T-cell-mediated cytotoxicity, even at initial αCD33-AMFab concentrations of 500 ng/ml (decrease of specific lysis: 73% to 15% ± SEM). We did not observe long-term downmodulation of CD33 expression levels by AM internalization (Supplementary Fig. S2F).

In summary, we showed that AdCAR T cells efficiently lyse pAML cells. However, we describe AM internalization as a common phenomenon for a variety of Fab- and Ab-based AM formats that target known internalizing antigens. AM internalization reduced their respective serum levels, thereby contributing to a form of “antigen sink” that impairs AdCAR T-cell cytotoxicity, especially at low levels of target antigen.

AdCAR T cells allow for sequential targeting of pAML cells

Based on the short half-lives of the AMs, we examined whether repetitive AM dosing prolongs AdCAR T-cell functionality and leads to better elimination of pAML cells compared to single administration. Owing to the heterogeneity of target antigen expression in AML, we took advantage of the versatility of the AdCAR technology, which enables a sequential application of AMs against different target antigens. AMs against CD33, CD123, or CLL-1 were used once at a concentration of 10 ng/ml or replenished every third day until day 12 of co-cultures of pAML cells and AdCAR T cells (E:T = 1:10). Additional experiments included a dose increase to 100 ng/ml and/or a switch to an AM of different target specificity on day 6 of co-culture (10 or 100 ng/ml), followed by AM replenishment on day 9.

All AMs reduced leukemia growth compared to αCD19-AMFab controls, underlining the specificity and potency of the AdCAR platform. However, a single addition of 10 ng/ml of αCD33-AMFab or αCLL-1-AMFab was insufficient to stop leukemia growth (Fig. 3A, B and Supplementary Fig. S3C, D).

Fig. 3: AdCAR T cells allow for sequential targeting of pAML cells.

A, B Long-term (12 days) co-cultures of AdCAR T cells and pAML cells (E:T = 1:10). pAML cell counts over time are plotted as normalized target cell count relative to starting conditions on day 0 (n = 7). αCD19-AMFab was replenished every third day at 10 ng/ml and served as a control. αCD33-AMFab was either applied once (10 ng/ml; dotted red line) or every third day until day 6 (solid red line). On days 6 and 9, the AM dose was either maintained at 10 ng/ml or increased to 100 ng/ml (bold dotted line). Alternatively, AMs were switched to AMs of different target specificity (αCD123-AMFab or αCLL-1-AMFab) on days 6 and 9 (10 or 100 ng/ml). C Representative flow-cytometry data from day 9 of co-culture. T cells and pAML cells were distinguished by CD2 and CD33 staining, respectively. Doublets, as well as dead cells were excluded, as described. D Corresponding AdCAR T-cell-mediated cytotoxicity on day 12 of co-culture. Specific lysis was calculated relative to the AdCAR T-cell condition without AM. Data are presented as mean ± SEM. Statistical analysis: Ordinary one-way ANOVA with Dunnett’s comparison; ns p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Repetitive administration of AMs at 10 ng/ml further delayed (αCD33-AMFab) or halted (αCD123-AMFab or αCLL-1-AMFab) leukemia outgrowth compared to a single addition of the AM.

Convincingly, an increase in dose of αCD33-AMFab or αCLL-1-AMFab from 10 to 100 ng/ml on day 6 of co-culture resulted in almost complete elimination of AML blasts (Fig. 3A, B), highlighting the potential for individually adjusting treatment conditions based on response to therapy or on target antigen levels.

Interestingly, switching from αCD33-AMFab on day 6 to 10 ng/ml of either αCD123-AMFab or αCLL-1-AMFab was effective, demonstrating comparable efficacy in eliminating pAML cells to 100 ng/ml αCD33-AMFab. AdCAR T-cell-mediated cytotoxicity was further enhanced not only by changing the target specificity but also by increasing the respective AM doses to 100 ng/ml (Fig. 3A–D). Notably, targeting the same pAML samples first with αCLL-1-AMFab followed by αCD33-AMFab or αCD123-AMFab did not result in higher lysis compared to continuous αCLL-1-AMFab targeting, as opposed to starting the treatment with αCD33-AMFab (Supplementary Fig. S3C–E).

The effects on AdCAR T-cell cytotoxicity were accompanied by a trend for increased T-cell proliferation after AM switching or dose increases (day 9), as well as increased expression of activation markers, indicative of pronounced and sustained AdCAR T-cell activation (Supplementary Fig. S3A, B, F, G).

These results collectively show that the use of highly modular AdCAR technology potently and specifically eradicated pAML cells in a dose- and time-dependent manner.

Fab molecules efficiently activate AdCAR T cells in vivo

Next, we aimed to translate our findings to a clinically relevant AML in vivo model by testing whether Fab-based AMs were able to efficiently direct AdCAR T cells against a highly aggressive OCI-AML-2 model. Therefore, HD-AdCAR T cells were expanded in vitro for 8 days and transferred to NSG mice bearing OCI-AML-2 leukemias (Fig. 4A). The mice were injected daily with αCD33-AMFab. AdCAR T cells were readily activated in vivo using Fab-based AMs. Convincingly, AdCAR T cells showed equipotency in controlling leukemia growth compared to conventional CD33CAR T cells, as quantified by BLI (Fig. 4B, C).

Fig. 4: Fab molecules efficiently activate AdCAR T cells in vivo.

A Schematic representation of the in vivo experimental timeline: NSG mice were inoculated on day −5 with luciferase-expressing OCI-AML-2 tumor cells followed by injection of AdCAR/CAR T cells on day 0. A second-generation conventional CD33CAR T-cell construct served as control. AdCAR/CAR T-cell functionality was assessed regularly by BLI of OCI-AML-2 cells. B In vivo BLI of OCI-AML-2 cells. C Bioluminescence images (n = 5 mice per group). Data are plotted as mean ± SEM. Statistical analysis: Ordinary one-way ANOVA with Tukey’s comparison; ns p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Treatment-free intervals prolonged AdCAR T-cell function

T-cell exhaustion is an emerging cause of CAR T-cell failure. We previously developed an in vitro system to monitor T-cell dysfunction induced by continuous bispecific antibody (BsAb) exposure in a clinically relevant B-cell lymphoma model [22]. It is unknown if AdCAR T cells react to CONT stimulation in the same way as conventional CAR T cells. Here, we adapted the long-term stimulation system to further evaluate AdCAR T-cell exhaustion in the context of AML. AdCAR T cells were co-cultured for 21 days with OCI-AML-3 in the presence of αCD33-AMFab. AdCAR T cells were isolated on days 0, 7, 10, 14, 17, and 21, and their cytotoxicity against OCI-AML-3 cells was assayed in co-cultures. We observed progressive AdCAR T-cell dysfunction over time, starting between day 10/14 of co-culture (Fig. 5A; 82% specific lysis on day 0 vs 13% on day 21), supporting our hypothesis that prolonged periods of AdCAR T-cell activation lead to loss of effector function.

Fig. 5: Treatment-free intervals prolong AdCAR T-cell function in vitro.

A AdCAR T cells were continuously stimulated for 21 days with 10 ng/ml αCD33-AMFab in the presence of irradiated OCI-AML-3 cells (E:T = 1:4; n = 3–12). OCI-AML-3 cells and AMs were replenished every third day. AdCAR T cells were isolated at the indicated days and cytotoxicity against OCI-AML-3 cells (E:T = 1:1) after 72 h was assessed by flow-cytometry. B Timeline and overview of the continuous and intermittent stimulation of AdCAR T cells co-cultured with OCI-AML-3 cells over 21 days. C Cytotoxicity of AdCAR T cells isolated from co-cultures against OCI-AML-3 cells at the indicated days (n = 3–12; E:T = 1:1; 10 ng/ml αCD33-AMFab). D T-cell proliferation expressed as fold change in CD2+ cells compared to conditions without AM (n = 5–12). E IL-2 secretion determined by CBA analysis of co-culture supernatants on day 17 (n = 6) and granzyme B expression of CD8+ AdCAR T cells isolated on day 17 and transferred to 72 h cytotoxicity assays (n = 5). Data are presented as mean ± SEM. Statistical analysis: paired t-test; ns p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

As previously shown by us and others, intermittent T-cell stimulation using TFIs or molecular and chemical switches can prolong T-/CAR T-cell functionality, mainly through transcriptional and epigenetic remodeling [19, 22, 23].

Hence, we compared continuously to intermittently stimulated AdCAR T cells in our AML-optimized in vitro dysfunction model for 21 days. In both conditions, AdCAR T cells were co-cultured with OCI-AML-3 target cells in the presence of αCD33-AMFab for 7 days. Then, AdCAR T cells were either exposed for another 7 days to AMs and OCI-AML-3 cells or cultured in the absence of AMs. All cultures were then treated for a further 7 days with αCD33-AMFab (Fig. 5B). AdCAR T cells were again isolated on days 0, 7, 10, 14, 17, and 21, and their cytotoxicity was assayed in co-cultures.

We observed significantly improved AdCAR T-cell-mediated cytotoxicity against OCI-AML-3 cells during and after the TFI (specific lysis ± SEM: day 14, CONT vs TFI = 43% vs 64%; day 17, CONT vs post TFI = 36% vs 60%; Fig. 5C). In addition, AdCAR T cells intermittently exposed to AMs demonstrated greater proliferation, IL-2 secretion, granzyme B production, and increased expression of PD1 and LAG-3 (Fig. 5D, E and Supplementary Fig. S4A). The AdCAR receptor was not differentially expressed between the two treatment modes (Supplementary Fig. S4B). Whereas T-cell subset analysis revealed a shift towards an effector memory subtype during CONT AM stimulation (Supplementary Fig. S4C), the resting period had no effect on this, indicating that the functional improvement is not driven by only a subpopulation of AdCAR T cells.

In summary, we showed that continuous AdCAR T-cell stimulation leads to a decrease in effector functions, which can be abrogated by the use of TFIs.

Treatment-free intervals lead to transcriptional reprograming of AdCAR T cells

To better understand how functional superiority is established, we performed bulk RNA sequencing on isolated AdCAR T cells from CONT- and TFI-treated cells at days 0, 14, and 21 from co-cultures from three individual donors. An analysis of differentially expressed genes (DEGs) of day 14 AdCAR T cells identified 115 significantly upregulated and 219 downregulated genes under TFI conditions versus CONT (p < 0.01; Fig. 6A). Unsupervised clustering showed markedly different gene expression patterns under the two treatment modes, indicating transcriptional reprograming. Most importantly, genes related to activation (IL2RA, CD70, LAG3) and cell cycle (CDK1, GMMN, E2F1, CDC45) were downregulated in day 14 TFI-treated AdCAR T cells compared to CONT-stimulated cells, consistent with functional rest (Fig. 6A, B). Interestingly, these genes remained downregulated in day 21 CONT-stimulated AdCAR T cells, indicating a progressive loss of cellular activity due to sustained antigen stimulation. In contrast, other genes related to T-cell activation (CD69, CD44, CD45, Jak1) were upregulated on day 14 TFI-treated relative to CONT-treated AdCAR T cells, pointing towards a better effector function. Pathway comparison of day 14 TFI- and CONT-treated AdCAR T cells was consistent with downregulation of cell cycle (E2F targets, normalized enrichment score, NES = − 3.35; G2M checkpoint, NES = − 3.21; MYC targets V1, NES = − 2.74; mitotic spindle, NES = − 2.25; p < 0.05) and metabolism-associated genes (OXPHOS, NES = − 2.08; glycolysis, NES = − 2.04; p < 0.05), highlighting AdCAR T-cell quiescence during TFIs (Fig. 6C). Compared to a model of chronic LCM virus infection [15], gene set enrichment analysis (GSEA) revealed a shift towards memory-related from effector-related genes in day 14 TFI-treated AdCAR T cells (Fig. 6D; GSE9650, NES = − 2.41, false-discovery rate q = 0.0).

Fig. 6: Treatment-free intervals lead to transcriptional reprograming of AdCAR T cells.

A Volcano plot of DEGs in day 14 TFI-treated versus CONT-treated AdCAR T cells; p < 0.01. Selected genes are highlighted in blue (downregulated) or red (upregulated). B Heatmap with hierarchical clustering of the top 100 DEGs in day 14 TFI-treated versus CONT-treated AdCAR T cells; p < 0.01. Selected genes are indicated. C Hallmark gene set analysis of day 14 TFI-treated versus CONT-treated AdCAR T cells; p < 0.05. D GSEA of day 14 TFI-treated versus CONT-treated AdCAR T cells using MSigDB and the gene set GSE9650_EFFECTOR_VS_MEMORY_CD8_TCELL_UP [15]. E Log2(TPM) values of CDK1 and IL2RA over time for TFI-treated and CONT-treated AdCAR T cells. DEG = differentially expressed gene; NES = normalized enrichment score; GSEA = gene set enrichment analysis.

These data imply that day 14 TFI-treated AdCAR T cells undergo rejuvenation through transcriptional reprograming. Convincingly, although we observed progressive downregulation of cell cycle (CDK1) or activation markers (IL2RA) in CONT-stimulated AdCAR T cells, resting periods led to a re-expression of these markers on day 21 TFI-treated cells (Fig. 6E). Overall, genes and pathways downregulated during day 14 TFI were upregulated again at day 21 (and vice versa; Supplementary Fig. S5A–C). GSEA showed that in contrast to day 14, effector-related versus memory-related genes were enriched at day 21 (Supplementary Fig. S5D). Collectively, these data suggest that day 21 TFI-treated AdCAR T cells are more functional than CONT-treated cells and have a greater potential for being re-activated upon αCD33-AMFab re-exposure. Notably, the re-expression of cell-cycle-related genes in day 21 TFI-treated AdCAR T cells did not reach the level on day 0, indicating that T-cell dysfunction cannot be completely reversed.