Evolving Exhaustion of T Cells during the Course of the Disease in AML Can be Abrogated By CD33 BiTE ^® Construct Mediated Cytotoxicity

Novel immunotherapeutic strategies like BiTE ^® (bispecific T cell engager) constructs aim to eradicate neoplastic cells by TCR-independent T-cell activation, and therefore rely on the function of autologous T cells. Currently, their efficacy is also evaluated in heavily pre-treated patients with relapsed/refractory acute myeloid leukemia (AML). Previous data demonstrated dysfunction in CD8 ⁺ T cells of AML patients (Knaus et al 2018). Thus, we aimed to characterize the progressive modulation of T-cell activity over the course of AML progression to improve the optimal application of T-cell based immunotherapeutic approaches.

Bone marrow mononuclear cells (BMMCs) from AML patients at time of initial diagnosis (ID), complete remission (CR), relapse (RL), as well as of age-matched healthy donors (HD) were analyzed for T-cell subset distribution and expression of exhaustion markers by flow cytometry. Additionally, T-cell function was assessed after stimulation with 1) CD3/CD28 beads; 2) AMG 330, a CD33/CD3 specific BiTE ^® construct, after incubation with OCI-AML3 target cells; or 3) AMG 330 in an autologous ex vivo long-term culture system after incubation with primary AML cells (pAML). After 6 days, T cell proliferation, expression of effector molecules and cytokines, and AMG 330-mediated T-cell cytotoxicity were assessed by flow cytometry. Lastly, we performed longitudinal bulk RNA-sequencing on 5000 sorted T cells from 7 matched ID-RL primary AML samples.

Immunophenotypic analysis of BM T-cell subsets revealed a shift from T _NAIVE toward central/effector memory subsets during AML progression. We observed lower percentages of T _NAIVE in RL (n=3) compared to CR (n=3) CD8 ⁺ T cells(11.8 vs. 45.2%, p=0.07; RL vs. CR). Conversely, RL patients showed increased percentages of CD8 ⁺ memory T cells (T _CM: 23.4 vs. 6.7%; T _EM: 29.4 vs. 20.2%; T _EMRA: 35.3 vs. 27.8%; RL vs. CR). Further characterization of exhaustion markers exhibited a significantly higher percentage of both CD4 ⁺ and CD8 ⁺ T cells expressing 2B4 (CD244) in ID (n=19) and RL (n=13) compared to HD (n=10, both p < 0.001). A higher percentage of PD-1 ⁺ CD8 ⁺ and TIM-3 ⁺ CD4 ⁺ T cells was detected in both ID and RL relative to HD (all p < 0.05). However, a significantly increased percentage of CD8 ⁺ T cells expressing TIM-3 and CD160 was detected in ID relative to HD (p < 0.05). Intriguingly, RL CD4 ⁺ T cells demonstrated a significantly higher level of LAG3 compared to ID (p < 0.01). In line with phenotypic exhaustion features, ID (n=4) and RL (n=5) CD8 ⁺ T cells showed reduced proliferation compared to HD (n=4) CD8 ⁺ T cells after CD3/CD28 bead stimulation (both p < 0.01). Correspondingly, we observed a marked reduction in the expression of Granzyme B (GZMB) by CD8 ⁺ T cells (both p < 0.05). Interestingly, when compared to ID, RL CD4 ⁺ T cells showed decreased TNF-α secretion (p < 0.05). In contrast to these findings, AMG 330-mediated T cell cytotoxicity against OCI-AML3 target cells was superior with RL T cells compared to ID T cells (p < 0.001). The percentage of GZMB ⁺ CD8 ⁺ T cells strikingly enhanced in RL relative to ID (p < 0.01). In an autologous setting with pAML samples, T cells from RL patients (n=6) showed higher AMG 330-mediated cytotoxicity compared to ID (n=9) T cells (67.7 vs. 35.2; RL vs. ID).

In our longitudinal RNA-sequencing, differentially expressed genes analysis detected 61 up- and 30 downregulated genes (log2 FC > 1 or < -1; p < 0.01) in RL T cells compared to their matched ID counterparts. Among the significantly upregulated genes in RL, we identified genes associated with memory T cell function (TP53INP2, DUSP4) and exhaustion (NR4A1, TOX2). Moreover, Gene set enrichment analysis showed significant enrichment of gene signatures associated to memory and exhausted T cells (normalized enrichment score (NES)=1.2 and 1.3; p-value= 0.026 and 0.008, respectively), depletion of oxidative phosphorylation (NES=-2.05; p _adj < 0.0001) and protein secretion (NES=-1.49; p _adj < 0.05) gene signatures in RL vs. ID T cells.

Taken together, our data show that patient T cells acquire an activated/exhausted phenotype upon AML progression. However, this is not reflected in the T-cell effector functions upon AMG 330 stimulation, in contrast to bead stimulation. These observations may highlight the significant role of the AML target cells in shaping a T-cell response. To this end, we will further analyze the longitudinal communication between T cells and their corresponding AML blasts.