Bone marrow lymphocyte dynamics during chemotherapy in pediatric acute myeloid leukemia

Abstract

T-cell-directed immunotherapy, which aims to boost or induce T-cell-mediated anti-tumor immunity, has shown remarkable success in various cancers, including B-cell precursor acute lymphoblastic leukemia (BCP-ALL), making it a compelling avenue for investigation in acute myeloid leukemia (AML).^{1, 2} Bispecific T-cell-engagers (TCEs) are a promising form of T-cell-directed immunotherapy that redirect CD3⁺ T-cells to tumor cells, thereby inducing T-cell activation and subsequent tumor cell lysis.³ However, bispecific TCEs, mainly targeting CD33 or CD123, have shown limited efficacy and/or high toxicity in relapsed/refractory AML.^4-7 A proposed strategy to enhance TCE therapy in AML is their administration during periods of measurable residual disease, for example, in between chemotherapy courses, as demonstrated in BCP-ALL.^{2, 8-10} Chemotherapy may, however, significantly alter the immune landscape¹¹: anthracyclines, for example, can promote anti-tumor immunity via immunogenic cell death,¹² but chemotherapy may also deplete lymphocytes and induce T-cell dysfunction.^{13, 14} Since pre-treatment T-cell infiltration and dysfunction in the tumor microenvironment are key predictors of bispecific TCE efficacy,^15-18 understanding how chemotherapy alters the immune landscape in the leukemic bone marrow (BM) is crucial for assessing the potential of TCEs in between chemotherapy courses in AML. Given differences in disease biology, immune system maturity, and treatment regimens between pediatric and adult AML,^{19, 20} pediatric-specific studies are necessary. Here, we examined the impact of chemotherapy-based regimens on the BM lymphocyte compartment in newly diagnosed pediatric AML (pAML).

We first characterized the treatment-naïve pAML BM lymphocyte compartment using diagnostic bulk RNA-sequencing (RNA-seq) data (Figure 1A). To reliably infer the lymphocyte composition from bulk RNA-seq data, we acquired a publicly-available single cell (sc) RNA-seq dataset²¹ to generate a healthy BM cell type signature matrix for use with CIBERSORTx.²² To validate its performance, we retrieved BM scRNA-seq data from 27 pAML cases at diagnosis, remission, and/or relapse,²³ and generated pseudo-bulk profiles (n = 62). Applying CIBERSORTx with the healthy BM reference to these pseudo-bulk profiles and comparing the deconvoluted estimates with the original scRNA-seq annotations (Figure S1A), we observed strong correlations for T-, B-, and NK-cells (T-cells: r = 0.72, P < 0.001; B-cells: r = 0.87, P < 0.001; NK-cells: r = 0.68, P < 0.001; Figure 1B). Similarly, CD4⁺ naïve, CD8⁺ effector, and CD8⁺ memory T-cells showed good concordance, while CD4⁺ memory and CD8⁺ naïve T-cells did not (Figure S1B), supporting the method's accuracy for most but not all lymphocyte subsets. Applying this approach to our primary study cohort (51 newly diagnosed pAML cases and seven age-matched controls; Figure 1A; Table S1), we found significantly lower fractions of T- and B-cells in the pAML BM compared to controls (P < 0.001 and P = 0.012, respectively; Figure 1C). Specifically, CD4⁺ naïve, CD8⁺ effector, and CD8⁺ memory T-cells were all less abundant (Figure S1C). NK-cell fractions did not differ (Figure 1C). These findings, as anticipated, indicate a diminished lymphocyte compartment in the BM in newly diagnosed pAML.

To investigate BM lymphocyte dynamics during chemotherapy, we performed bulk RNA-seq on 42 BM samples from 21 pAML cases, collected at end of induction 1 (EOI1) and EOI2 (similar time intervals, P = 0.62, Figure S1D). All patients were treated according to the NOPHO-DBH AML-2012 protocol (Figure 1A,F and Table S1). During induction 1, 19/21 patients received mitoxantrone, etoposide, and cytarabine (MEC). At EOI1, eighteen patients had good responses (<5% blasts by flow cytometry), while three (AML5, AML45, AML47) were poor responders (Figure 1D,F). Among good responders, lymphocyte fractions increased (>125% of baseline) in ten patients, remained stable (75%–125%) in four, and decreased (<75%) in four (Figure 1E). Although an increase in lymphocyte fraction was expected due to the substantial blast clearance in good responders (median 62.5%–0.1%), lymphocyte changes did not correlate with blast reduction (r = −0.32, P = 0.20; n = 18; Figure S2A), suggesting differential effects of MEC on the BM lymphocyte compartment. No specific cytogenetic alterations were associated with a particular direction of lymphocyte change, which was expected due to the relatively small number of cases. Notably, all three poor responders showed marked lymphocyte increases at EOI1 (median 392%, range: 327%–492%), despite high residual AML burden (median 39%, range: 23%–70%; Figure 1E), suggesting that significant lymphocyte infiltration and/or expansion can occur even in the context of persistent leukemic infiltration. Lymphocyte subset analysis revealed that T-cells predominated at diagnosis (mean 75 ± 14%) and further increased by EOI1 (mean 86 ± 6.5%, P = 0.016; Figures 1F and S2B). Within the T-cell compartment, CD4⁺ naïve T-cells represented the most abundant subset at diagnosis (mean 56 ± 27%), followed by CD8⁺ memory (29 ± 15%) and CD8⁺ effector T-cells (6.3 ± 7%, Figure S2C,D). CD4⁺ naïve and CD8⁺ memory T-cell proportions remained largely stable following induction 1 (64 ± 11%, P > 0.99 and 21 ± 11%, P = 0.37, respectively), whereas CD8⁺ effector T-cells increased (11 ± 5.4%, P = 0.01; Figure S2C,D). B-cell fractions decreased from 21 ± 14% at diagnosis to 7.5 ± 6.1% at EOI1 (P = 0.006), while NK-cell proportions increased in just over half of patients (11 > 125%, six 75%–125%, and four <75%; 3.4 ± 6.6% vs. 6.1 ± 5.6%; P = 0.37; Figures 1F and S2B). To extend this analysis beyond relative proportions—of relevance due to the marked reduction in leukemic blasts from diagnosis to EOI1—we used sample-wise scaled abundance scores (CIBERSORTx absolute mode), which adjust inferred cell-type fractions by the overall transcriptomic content of each sample. This analysis revealed that the scaled abundance of T-cells also increased following induction 1, including CD4⁺ naïve and CD8⁺ effector subsets, whereas CD8+ memory T-cells remained stable (Figure S2E,F). Scaled B-cell abundance scores declined in about two-thirds of cases, while NK-cells showed a trend towards an increase (P = 0.076; Figure S2E,F). To verify our deconvolution-based results using an orthogonal method, we performed flow cytometry on a subset of pAML patients (n = 5, diagnosis-EOI1-EOI2) and four healthy pediatric donors (Figure S3A, Tables S1 and S4). Although a direct comparison of matched values was not feasible due to differences in sample processing between the bulk RNA-seq and flow cytometry datasets (Supporting Information Methods), we observed a clear increase in BM T-cell abundance relative to all BM mononuclear cells (BMMCs) from diagnosis to EOI1 in these pAML patients, in line with our bulk RNA-seq data from the full cohort (Figure S3B). Moreover, the dynamics of CD4⁺ T-, CD8⁺ T-, and B-cells closely mirrored those inferred from bulk RNA-seq (Figure S3B; gating strategy in Figure S3C; NK-cell detection was not possible due to marker overlap with leukemic blasts), supporting the robustness of our deconvolution-based analyses. Altogether, following induction 1, most patients showed increased or stable lymphocyte proportions alongside marked blast reduction. This was accompanied by a shift in the lymphoid compartment towards a higher T-cell fraction—in particular CD8⁺ effector T-cells—whereas B-cell proportions declined. Importantly, abundance scores adjusted for overall transcriptomic content confirmed these trends.

During induction 2, chemotherapy regimens were more heterogeneous: 13 patients received ADE (cytarabine, daunorubicin, and etoposide), five FLA(D) (fludarabine and cytarabine ± daunorubicin), and three other regimens (Figure 1F and Table S1). Seventeen patients maintained remission, whereas one (AML40) showed disease progression (from 0.3% to 10%; Figure 1D). Of the three initial poor responders, AML5 achieved remission, whereas AML45 and AML47 had persistent disease (>5%; Figure 1D). Despite regimen variability, lymphocyte fractions declined significantly at EOI2 (P = 0.026; Figure 1E). No differences between ADE and FLA(D)-treated patients were observed, though small group sizes precluded statistical testing (Figure S3D). The abundance of T-cells out of total lymphocytes remained stable in most cases, while B-cell fractions frequently increased (11 >125%, four 75%–125%, six <75%) and NK-cell levels declined in nearly two-thirds of patients (13/21, P = 0.09; Figures 1F and S2B). Taken together, despite diverse treatment regimens, more than half of patients experienced a decline in total lymphocyte levels following induction 2, contrasting with the earlier induction phase.

To assess whether induction therapy was associated with changes in T-cell diversity, we profiled the T-cell receptor (TCR) repertoire using MiXCR²⁴ (successful in 20/21 cases; Figure 2A). Shannon diversity indices increased from diagnosis to EOI1 and EOI2 (P = 0.008 and P = 0.08, respectively), but remained within a relatively narrow range throughout induction therapy in most patients (EOI1: 75%–125% in 15/20 patients, >125% in 4, <75% in 1; EOI2: 75%–125% in 17, >125% in 2, <75% in 1), indicating only modest changes in overall TCR diversity (Figure 2B). Identical CDR3 β-chain sequences were detected at multiple timepoints in 8/20 cases, representing a median of 1.9% of the repertoire (range 0.3%–5.1%; Figure 2C). These data suggest that chemotherapy is associated with a diverse and largely distinct post-treatment T-cell repertoire. Whether this includes tumor-reactive clones remains unclear. Future studies should investigate the tumor-specificity of T-cells persisting or emerging during therapy, as these may enhance TCE efficacy.²⁵ In addition, the modest sensitivity of bulk RNA-seq-based TCR repertoire profiling requires validation using dedicated TCR-sequencing approaches.

Given its relevance for responses to T-cell-directed immunotherapies, we next assessed T-cell functionality.^{15, 16} To this end, we applied established gene signature scores for T-cell cytolytic activity,²⁶ exhaustion,²⁷ and senescence²⁸ to our bulk RNA-seq dataset, corrected for sample-wise scaled T-cell abundance. Cytolytic activity scores rose significantly following induction 1 (increased in sixteen cases [>125%], remained stable in two [75%–125%], and declined in three [<75%]; P = 0.006; Figure S3E). From EOI1 to EOI2, cytolytic activity scores showed a more heterogeneous pattern—rising in six, remaining stable in another six, and decreasing in nine patients—yet the overall increase from diagnosis to EOI2 remained statistically significant (P = 0.041; Figure S3E). In contrast, senescence and exhaustion scores showed substantial interpatient variability without consistent directional change (Figure S3E). To further evaluate the functionality of T-cells at EOI1 and EOI2 in pAML, we next investigated the ability of a CD33/CD3-TCE (AMV564) to induce AML cell lysis via autologous T-cells derived from EOI1 or EOI2 BMMCs. Co-culturing EOI1/EOI2 BMMCs with CD3⁺ T-cell-depleted diagnostic BMMCs containing CD33⁺ AML cells (effector-to-target ratio 1:3) for three days in the presence or absence of AMV564 showed robust CD33⁺ cell lysis (mean specific lysis: 54 ± 37% at EOI1, 57 ± 37% at EOI2; Figure 2D,E). AMV564-induced cytotoxicity was accompanied by robust T-cell activation, evidenced by the upregulation of the T-cell activation markers CD25 and CD137, granzyme B expression, and T-cell proliferation (although not statistically significant in case of CD137; Figure 2F,G). Subset analysis revealed that TCE therapy led to a phenotypic shift of naive to effector memory and central memory T-cells (Figure S3F). These data suggest that cytolytic potential increases, and that TCE treatment is capable of activating autologous T-cells ex vivo and inducing lysis of primary CD33⁺ cells, at EOI1 and EOI2. While these results indicate functional T-cell potential at EOI1 and EOI2, further studies including long-term stimulation assays are required to assess the durability of T-cell function, T-cell function relative to healthy donor T-cells, and in vivo relevance.

Finally, we assessed BM lymphocyte dynamics in patients treated on other pAML protocols. Using data from the COG AAML1031 protocol (BM scRNA-seq data from seven treatment-naïve patients with paired diagnosis-EOI1 samples²³; Table S2) and NOPHO-AML 2004 protocol (immunohistochemistry data for 13 patients with diagnosis-EOI1-EOI2 BM samples²⁹; Table S3; patients in both cohorts had a good response to induction therapy), we observed both similarities and discrepancies in lymphocyte dynamics compared to the primary study cohort. Consistent with our previous findings, six out of seven COG patients (cytarabine, daunorubicin, etoposide, bortezomib/sorafenib) showed increased lymphocyte levels at EOI1 (P = 0.031; Figure S4A–D). Furthermore, the proportion of T-cells out of lymphocytes increased, while B-cells declined, and NK-cell changes were variable (Figure S4B,C). Conversely, NOPHO-AML 2004 patients (course 1: cytarabine, idarubicin, etoposide, 6-thioguanine; course 2: cytarabine, mitoxantrone) exhibited profound T-cell heterogeneity and a reduction in B-cells at EOI1 (P = 0.003; Figure S4E,F; considering T- and B-cells in aggregate was not feasible because of the single-stain IHC). By EOI2, B-cell proportions recovered (P = 0.007; Figure S4F) and nine out of thirteen patients showed increased (>125%) T-cell levels, which was notable given the shorter EOI1–EOI2 interval compared to the NOPHO-DBH AML-2012 protocol (P < 0.001; Figure S4G). These findings suggest common trends in BM lymphocyte dynamics but also highlight protocol-specific variations, possibly linked to the use of specific chemotherapeutic agents. Given the limited number of patients in these external pAML cohorts, confirmation in larger cohorts is warranted.

A better understanding of lymphocyte dynamics during current treatment regimens in pAML is urgently needed to understand whether the application of TCEs during periods of low tumor burden could be a viable treatment strategy. In this study, we found that induction 1, comprising MEC in nearly all patients, led to preserved or increased relative lymphocyte abundances alongside marked blast reduction in most cases. This was accompanied by a shift towards higher T-cell fractions, potentially creating a favorable window for TCE therapy.^{15, 18} Importantly, lymphocyte abundance was assessed at standardized timepoints immediately preceding the subsequent chemotherapy course—that is, once patients had met hematologic recovery criteria (ANC ≥ 0.5 × 10⁹/L and platelets ≥50 × 10⁹/L). As such, our findings reflect the immune landscape at the end of each treatment cycle, rather than continuous dynamics throughout the inter-treatment interval. Future studies—including longitudinal sampling during inter-treatment intervals—are warranted to more precisely define optimal windows for TCE intervention in between chemotherapy courses. The absence of a correlation between blast reduction and lymphocyte changes suggests that chemotherapy exerts differential effects on the lymphocyte compartment. Further studies are needed to clarify the mechanisms underlying divergent lymphocyte recovery, which may support the adaptation of treatment regimens to optimize conditions for immunotherapeutic interventions. Despite the heterogeneity of agents used in induction 2, more than half of patients showed a decline in lymphocyte levels. Nonetheless, the increase in T- and B-cells observed in most patients from the NOPHO-AML 2004 cohort after induction 2 suggests that lymphocyte recovery at this treatment stage is not uniformly impaired. Our transcriptomic and ex vivo functional data align with preclinical findings in adult AML³⁰ and provide a basis for further investigations in in vivo models and early clinical trials. Such efforts should prioritize novel TCE constructs targeting multiple tumor-associated (e.g., NCT05673057) or tumor-specific antigens.

Joost B. Koedijk: Conceptualization; methodology; investigation; validation; writing—original draft; data curation; writing—review and editing. Farnaz Barneh: Methodology; investigation; validation; formal analysis; writing—review and editing; data curation. Joyce E. Meesters-Ensing: Methodology; data curation; writing—review and editing. Marc van Tuil: Methodology; writing—review and editing. Edwin Sonneveld: Methodology; data curation; writing—review and editing. Sander Lambo: Data curation; methodology; resources; writing—review and editing. Alicia Perzolli: Methodology; writing—review and editing. Elizabeth K. Schweighart: Writing—review and editing; methodology; investigation; data curation. Mauricio N. Ferrao Blanco: Methodology; investigation; data curation; writing—review and editing. Merel van der Meulen: Methodology; data curation; validation; writing—review and editing. Anna Deli: Methodology; investigation; writing—review and editing. Elize Haasjes: Methodology; investigation; writing—review and editing. Kristina Bang Christensen: Writing—review and editing; methodology; investigation. Hester A. de Groot-Kruseman: Methodology; data curation; validation; writing—review and editing. Soheil Meshinchi: Methodology; resources; writing—review and editing; investigation. Henrik Hasle: Investigation; methodology; validation; resources; writing—review and editing; formal analysis. Mirjam E. Belderbos: Methodology; writing—review and editing; formal analysis. Maaike Luesink: Data curation; writing—review and editing. Bianca F. Goemans: Data curation; writing—review and editing. Stefan Nierkens: Supervision; conceptualization; writing—review and editing. Jayne Hehir-Kwa: Data curation; formal analysis; writing—review and editing. C. Michel Zwaan: Conceptualization; supervision. Olaf Heidenreich: Conceptualization; methodology; formal analysis; resources; supervision; funding acquisition; writing—review and editing; project administration.

O. H. receives institutional research support from Syndax and Roche. C. M. Z. receives institutional research support from Pfizer, AbbVie, Takeda, Jazz, Kura Oncology, Gilead, and Daiichi Sankyo; provides consultancy services for Kura Oncology, Bristol Myers Squibb, Novartis, Gilead, Incyte, Beigene, and Syndax; and serves on advisory committees for Novartis, Sanofi, and Incyte. The remaining authors declare no competing financial interests.

This study was approved by the Institutional Review Board of the Princess Máxima Center for Pediatric Oncology (approval codes: PMCLAB2021.207, PMCLAB2021.238, and PMCLAB2022.328; biobank) and the NedMec review board (prospective observational MIMIC study: NL75515.041.21). Written informed consent was obtained from all patients and/or guardians.

This work has been funded in part by a KIKA (329) program grant to OH.