Development and characterization of anti-CXCR4 chimeric antigen receptor T cells

Highlights

• •CXCR4 is widely and highly expressed on patients with AML or ALL.
• •Anti-CXCR4 CAR-T cells were generated based on a CXCL12 extracellular domain.
• •Anti-CXCR4 CAR-T cells demonstrate potent activity against CXCR4 expressing AML and ALL cell lines in-vitro.

Abstract

Acute Myeloid Leukemia (AML) and Acute Lymphoblastic Leukemia (ALL) are aggressive hematologic malignancies characterized by dysregulation of normal hematopoiesis and acquisition of stem-like, self-renewing properties leading to oncogenesis. Current treatments primarily rely on toxic chemotherapies with or without hematopoietic stem cell transplantation (HSCT). These are associated with significant short- and long-term side effects and sub-optimal outcomes across treatment of both AML and ALL. Improved HSCT methods are needed that can both eliminate leukemic cells and improve outcomes. The C-X-C chemokine receptor type 4 (CXCR4) plays a key role in both normal hematopoiesis and leukemogenesis by attracting and retaining cells in the bone marrow niche. It has been targeted using antibody or drug-based approaches leading to clinical trials and therapeutics across indications. In the context of chimeric antigen receptor (CAR) T cells, it has been expressed as a co-receptor to improve bone marrow homing and amplify tumor eradication. Here, we confirm high CXCR4 expression across AML and ALL using publicly available tumor transcriptomic datasets. Subsequently, we report the development of anti-CXCR4 CAR-T cells that demonstrate potent activity against a panel of leukemic cell lines in vitro without activity against other T cells. We observe decreased CXCR4 protein expression in CAR-positive populations indicating a potential pathway for T cell survival. Our findings support the potential of anti-CXCR4 CAR-T cells as a broadly applicable strategy for eliminating both AML and ALL, with possible extension to hematopoietic stem and progenitor cells, unlocking the potential of this strategy as a dual HSCT-conditioning and anti-leukemia agent.

Introduction

Children diagnosed with Acute Myeloid Leukemia (AML) face five-year survival rates of <70%, with relapse occurring in nearly 25% of cases [1]. In children diagnosed with Acute Lymphoblastic Leukemia (ALL), five-year survival rates are approximately 90% with relapse rates of 10–15%. In adults, outcomes are worse across both diseases with five-year survival rates of 30–40% and relapse rates reaching up to 66% [[1], [2], [3], [4], [5]]. Current treatment approaches for leukemia primarily include aggressive chemotherapies, as well as allogeneic hematopoietic stem cell transplantation (HSCT). Although allogeneic HSCT can provide durable remission through replacement of the hematopoietic system and additional graft-versus-leukemia effect, it requires non-specific myeloablation resulting in substantial toxicity across the body and often does not prevent relapse [1,[6], [7], [8], [9]]. For successful allogeneic HSCT to occur in the leukemia setting, robust leukemic cell clearance is necessary, and additionally, native HSPCs must be removed from the bone marrow niche to enable donor HSPC engraftment [6].

Chimeric Antigen Receptor (CAR) T cells are T cells genetically engineered to express synthetic receptors composed of antigen-specific extracellular domains fused to intracellular signaling domains, enabling targeted effector responses [10]. CAR-T cells have shown great efficacy in aggressive precursor B-type ALL, with remission being established in otherwise refractory patients in ∼80% of cases [[11], [12], [13]]. Despite the successes of CAR-T cells in ALL, similar approaches have yet to be extended to AML. One challenge with this treatment approach for AML is that many receptors of interest on AML cells are shared with healthy HSPCs and their targeting can thus lead to bone marrow failure. However, sparing HSPCs can also lead to sparing leukemic stem cells (LSCs), cells capable of perpetuating disease and leading to relapse [14,15]. Hence, treatments capable of targeting both leukemic cells and HSPCs with subsequent allogeneic HSCT to restore the hematopoietic system may be the ideal treatment approach [16,17].

Here, we aim to build upon previous work to develop a novel array of CAR-T cells that target both hematopoietic and leukemic cell populations as dual transplant enabling and immunotherapeutic agents to improve patient care and outcomes [[18], [19], [20], [21], [22]]. To accomplish this, we assessed targets shared between HSPCs and leukemic cells that could enable simultaneous clearance of both cell populations. These efforts led us to interrogate C-X-C chemokine receptor type 4 (CXCR4) and its native ligand C-X-C motif chemokine 12 (CXCL12) which have recently been shown to play critical roles in AML tumor initiation and development through the activation of multiple signaling pathways as well as participate in the regulation of cancer stem cells [17,23]. CXCR4 is generally expressed amongst lymphoid cells, however it is also highly expressed on HSPCs, AML blasts, and LSCs [[24], [25]]. Furthermore, it has been shown that CXCR4 expression is required for the engraftment and development of AML in mouse models [26].

Targeting CXCR4 has shown promise in treatment of leukemia with a variety of approaches that are at various stages of clinical development. Radioisotopes targeting CXCR4 reduced leukemia burden in mouse models with concurrent depletion of HSPCs in the bone marrow microenvironment [26]. Upon clinical translation of this approach, three patients were reportedly successfully treated with this therapy which resulted in leukemia clearance and enabled allogeneic HSCT with hematopoietic engraftment and few adverse effects [26]. A recent similar approach using radio-targeting and Pentixather, a synthetic peptide that targets CXCR4, showed response and HSPC engraftment in 6 out of 7 patients with relapsed and refractory AML [27]. Low adverse events were also observed with antibody-targeting based approaches, some of which reached Phase Ib/II clinical trials [28,29]. In combination, these previous studies demonstrate the feasibility of targeting CXCR4 in the context of leukemia. We aim to improve upon these prior treatments as a CAR-T cell approach has not been previously explored and offers potential improvements through direct, sustained cytotoxic activity and potential for direct homing to the tumor microenvironment.

However, one potential concern with an anti-CXCR4 based CAR-T cell approach which may have hindered prior efforts is the possibility of T cell fratricide due to the presence of CXCR4 on native T cell populations [24,30]. Previous CAR-T cells that have been developed targeting receptors that are also expressed on T cell populations have been found to exhibit fratricide, such as those targeting CD7 and CD45. In these settings compensatory approaches including gene editing, epitope shielding, and others strategies have been used to mitigate this issue [31,32]. For this reason, despite its expression on native T cell populations, CXCR4 was still seen as a viable CAR-T cell target for AML and ALL with the ability to employ some of these strategies were fratricide to be observed.

In this study, we first aimed to confirm that CXCR4 was highly expressed on HSPCs, LSC, and other leukemic cells at the RNA level in published datasets. To subsequently efficiently target CXCR4, we developed single chain variable fragment (scFv)-based and CXCL12 ligand-based CAR-T cells which differed in their susceptibility to fratricide. Overall, we show the development and potential of novel ligand-based (LiBa) anti-CXCR4 CAR-T cells as an effective strategy for targeting AML and ALL that can likely be extended for the targeting of HSPCs.

Methods

Bioinformatic analyses

Normalized bulk RNA-seq data was acquired from publicly available sources (GSE17855, GSE68720, EGAS00001004701, GSE94669, Bloodspot, Gene Expression Atlas) [[33], [34], [35], [36],25,37,38]. Gene expression data was processed in R (version 4.3.3) and visualized using ggplot2 [39].

Single-cell RNA-seq data was additionally acquired from publicly available sources (GSE154109) [40]. Unique Molecular Identifier (UMI) counts generated with 10x Genomics barcodes were downloaded and process with Seurat (version 5.2.1) [41]. Genes were kept if they were expressed in greater than 3 cells and cells were kept if they contained >200 detected genes. Furthermore, cells were retained if they had greater than 300 RNA features but fewer than 6000 and if they had <15% mitochondrial DNA. Gene expression values were normalized using Seurat’s LogNormalize() method. After merging all samples into a single object, highly variable genes were identified using FindVariableFeatures() (3000 features). Data was then scaled using ScaleData() and dimensionality reduction was performed with principal component analysis using RunPCA() (30 principal components). Batch correction across patient samples was performed using Harmony (version 1.2.3) [42] (RunHarmony()) using orig.ident as the batch variable. UMAP embeddings were generated using RunUMAP() (dimensions 1:20), and graph-based clustering was performed using FindNeighbors() and FindClusters() at resolution=0.5 over the Harmony-corrected embeddings.

To annotate cell populations, cluster-level cell annotation was performed using ScType [43] based on their specific “Immune System” reference gene set. The RNA assay was scaled using ScaleData() prior to scoring. Cell types were then collapsed into broader categories according to their labeled lineage. Expression of CXCR4 and other marker genes within these cell types was visualized using ggplot2.

Cell lines

Human AML cell lines Kasumi-6 and Kasumi-1 and human B-ALL cell line Nalm-6 were purchased from American Type Culture Collection (ATCC, Manassas, VA). Other AML cell lines MOLM-13 and NOMO-1 were purchased through DSMZ (Germany). Packaging cell line 293GP was acquired from ATCC. Cell lines were stably transduced with GFP-luciferase [22].

Generation of CAR constructs

Protein sequences for the alpha and beta splice variants of CXCL12 were previously published and obtained from UniProt (accession P48061) [44]. Protein sequences for the heavy and light chain of the anti-CXCR4 antibody MDX-1338 were obtained from the published patent [45]. Protein sequences were reverse-translated and codon optimized for expression in human cells using the Integrated DNA Technologies (IDT) codon optimization tool. Extracellular domains for the CAR-T cells were designed and custom synthesized by IDT as follows: the alpha splice variant of CXCL12, the beta splice variant of CXCL12, the alpha splice variant followed by the beta splice variant connected by a (G₄S)₃ linker, the light chain of MDX-1338 followed by the heavy chain of MDX-1338 connected by a (G₄S)₃ linker, and the heavy chain of MDX-1338 followed by the light chain of MDX-1338 connected by a (G₄S)₃ linker. Extracellular domains were subcloned into an MSGV1-based retroviral backbone plasmid encoding an N-terminal CD8α peptide, followed by one of the aforementioned extracellular domains, a FLAG tag in the case of the CXCL12-based constructs [46], a CD28 hinge region, a CD28 transmembrane region, a CD28 intracellular domain, and the intracellular domain CD3ζ, as previously described [47]. For constructs containing a truncated costimulatory and signaling domain, DNA oligo-primers were designed to amplify the areas surrounding the CD28 costimulatory and CD3ζ domains such that these two domains were deleted from amplified DNA.

Generation of retroviral supernatant

Retroviral particles containing CAR DNA were generated by transfecting 5 × 10⁶ 293-GP cells with CAR plasmid and RD114 plasmid DNA (2:1 ratio) using Lipofectamine 2000. Supernatant was harvested at 48 and 72 h post transfection.

CAR-T cell generation

Peripheral blood mononuclear cells were acquired from Stanford Blood Center. Healthy donors provided peripheral blood and T cells were isolated using Miltenyi anti-CD3 magnetic beads. Cells were cryopreserved using Cryostor CS10. Cells were thawed and activated with CD3/CD28 Dynabeads (Gibco) at a 3:1 bead:cell ratio in T cell media (RPMI1640 supplemented with 10% FBS, 10 mmol/L HEPES, 2 mmol/L GlutaMAX, 100 U/mL penicillin, 100 μg/mL streptomycin [Gibco]) and 100 IU/mL IL-2 (Peprotech, Cranbury, NJ). Activated T cells were retrovirally transduced with CAR on days 2 and 3 on Retronectin (Takara)-coated plates and identical media plus 300 IU/mL IL-2 was added. Anti-CD3/CD28 beads were removed on day 5. Media and IL-2 (100 IU/mL) were added on day 5 and day 8. T-cells were used for assays on day 10 or 11 post transduction.

CAR-T cell cryopreservation and preparation

For cryopreservation, 5–10×10⁶ CAR-T cells were suspended in 1 mL of Cryostor CS10 and slowly frozen to -80 °C in a specialized freezing container (i.e. Mr. Frosty). For thawing frozen CAR-T cells, a single tube of CAR-T cells was submerged in warm water (37 °C) until only a small frozen pellet remained. Cells were then washed in 13 mL of T cell media (RPMI1640 supplemented with 10% FBS, 10 mmol/L HEPES, 2 mmol/L GlutaMAX, 100 U/mL penicillin, 100 μg/mL streptomycin, and 100 IU/mL IL-2) and resuspended in 10 mL of T cell media with 200 IU/mL IL-2. Media was changed after 24 h and cells were overall allowed to recover for 48 h before experiments were performed.

IncuCyte cell killing assay

50,000 tumor cells (Nalm-6 or Kasumi-1) stably expressing GFP were co-cultured with CAR-T cells at 1:1 Effector:Target ratio in 250 μL of complete RPMI (RPMI-1640, 2 mM L-glutamine, 10 mM HEPES, 1 mM sodium pyruvate, 4500 mg/L glucose, 1500 mg/L sodium bicarbonate) with 20% FBS and 1x penicillin-streptomycin supplement [Gibco, Grand Island, NY]. Cells were placed in an IncuCyte machine at 37 °C for up to 72 h in triplicate for each condition. Plates were analyzed every 3 h using the IncuCyte ZOOM Live-Cell Analysis System (Essen Bioscience). Total integrated GFP intensity was reported per cell, and values were normalized to the starting intensity to plot fold change over time.

IncuCyte serial dilution assay

50,000 tumor cells (Nalm-6 or Kasumi-1) stably expressing GFP were co-cultured with CAR-T cells at 1:1, 1:2, 1:4, 1:8, and 1:16 Effector:Target ratio in 250 μL of complete RPMI with 20% FBS and 1x PenStrep. Cells were placed in an IncuCyte machine at 37 °C for up to 72 h in triplicate for each condition. Plates were analyzed every 3 h using the IncuCyte ZOOM Live-Cell Analysis System (Essen Bioscience). Total integrated GFP intensity was reported per cell, and values were normalized to the starting intensity to plot fold change over time.

Flow cytometry

All flow cytometry assays were performed on a FACS Symphony A5 machine. CXCR4 staining was performed using anti-CXCR4 BV421 (Clone: 12G5 BioLegend). CAR positivity was determined with anti-(G₄S)₃ linker PE for extracellularly linked concatemeric trivalent cytokine (ELECTRIC) CAR-T cells (Clone: E702V Cell Signaling Technology) and anti-FLAG-tag PE/Cy7 for ligand-based CAR-T cells (Clone: L5 BioLegend). Data was analyzed using FlowJo or FCS Express 7 software with gating, mean fluorescence intensity, or percent positive shown in figures.

Statistical analysis

Data analysis was performed using GraphPad Prism. All P-values were calculated as described in each corresponding figure caption. P < 0.05 was considered statistically significant, and P values are denoted with asterisks as follows: P > 0.05, not significant (ns); *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. Corrected p-values determined from multiple comparisons tests were used if multiple comparisons were made. Data points on graphs were used to show group mean values with error bars of standard error.

Results

CXCR4 is widely expressed on primary cells from pediatric patients with AML and ALL and on leukemia/lymphoma cell lines

To determine the levels of gene expression of CXCR4 across cells from patients with AML and ALL, we analyzed publicly available RNA-seq datasets generated from large cohorts of pediatric patients (GSE17855, GSE68720, EGAS00001004701) [33,34,36]. Gene expression was normalized across a given dataset and log2(FPKM) was used to determine positive or negative expression (log2(FPKM) > 0 indicating positive expression). For reference, we also included expression of other AML and ALL targets such as KIT, MPL, FLT3, CD123 (IL3RA), CD70, CD33 (PROM1), CD19, CD22, and MS4A1 [22,[48], [49], [50], [51], [52], [53]]. We found that CXCR4 was positively expressed across >99% of patient samples with levels similar to those of FLT3 (Fig. 1A, Supp Figure S1). In addition, we analyzed expression data from Lymphoma cell lines (GSE94669) and again confirmed high levels of CXCR4 [35] (Fig. 1B). Data reported by Bloodspot [37] and the Gene Expression Atlas [24] also showed high levels of CXCR4 expression across HSPCs, leukemic cells, and the majority of leukemic cell lines compared to the aforementioned reference genes (Fig. 1C, Supp Figure S2). Finally, we analyzed single-cell RNA sequencing data from bone marrow aspirates of a cohort of pediatric patients with AML (n = 8), ALL (n = 7), or healthy (n = 4) disease status (GSE154109) [40]. ScType [43] was used to annotate clusters with cell-type labels based on an Immune System specific reference gene set. Cells labeled as “Unknown” did not meet confidence thresholds for assignment based on the marker gene set enrichment and thus represent cells with transcriptional profiles that are difficult to assign to any single cell-type [42]. Overall, we found that CXCR4 was broadly expressed across progenitor, myeloid, lymphoid, and unknown compartments in both AML and ALL patient bone marrow (Fig. 1D). Notably, CXCR4 expression is broader than the previously described reference genes across multiple cell populations (Supp Figure S3).

Fig. 1.

CXCR4 expression reanalysis of previously reported RNA-seq data and surface expression quantification of cell lines. (A) Bioinformatic analysis of bulk RNA-seq data from pediatric patients with AML (GSE17855, GSE68720). Violin plot showing log2(FPKM) expression distribution for CXCR4 compared to other AML and ALL targets. (B) Analysis of bulk RNA-seq data from indicated lymphoma cell lines (GSE94669). (C) Gene expression reanalysis from bulk RNA-seq data reported by Bloodspot with cell-type labels additionally acquired from Bloodspot. Bar plot shows normalized log2 expression data. (D) Analysis of single-cell RNA-seq data (GSE154109) from pediatric patients with AML or ALL. Cell-type categories were derived from ScType-based cell-labels. Violin plots show normalized log expression data. (E) CXCR4 protein expression quantified on different AML and ALL cell lines using flow cytometry.

To support these transcriptional analyses, we quantified the cell surface expression of CXCR4 on four AML (Kasumi-1, Kasumi-6, Molm-13, and Nomo-1) and one ALL cell line (Nalm-6) by flow cytometry and found that CXCR4 was expressed across all the cell lines (Fig. 1E).

Generation and characterization of CXCL12 based CAR-T cells

To effectively target the CXCR4 receptor, we explored various CXCR4-specific extracellular domain designs for the CAR-T cells. We first cloned the variable domain of the anti-CXCR4 agent MDX-1338 (Ulocoplumab/BMS-936,564) [30,40] on a CD28.CD3ζ backbone with a CD28 hinge, and transmembrane domains. We generated two versions where the light (V_L) and heavy (V_H) chains were combined via (G₄S)₃ linker in different orientations relative to the cell membrane (Fig. 2A). Interestingly, in pilot experiments when we transduced human primary T cells with retrovirus expressing these scFv based CXCR4 CAR constructs, T-cell viability quickly declined possibly due to fratricide which prevented subsequent characterization.

Fig. 2.

(A) Structure of anti-CXCR4 antibody-based CAR-T cells. From N- to C-terminus, CARs were designed with the extracellular domain(s) connected via (G4S)3 linker, CD28 hinge and transmembrane region, a CD28 costimulatory domain, and CD3ζ activation domain. Extracellular domains included a combination of heavy (V_H) and light (V_L) chains from an anti-CXCR4 antibody (Created in BioRender. Seir, G. (2025)). (B) Structure of anti-CXCR4 LiBa CAR-T cells. From N- to C-terminus, CARs were designed with the extracellular domain(s) connected via (G4S)3 linker if applicable, followed by a FLAG-tag for CAR detection, CD28 hinge and transmembrane region, a CD28 costimulatory domain, and CD3ζ activation domain. Extracellular domains included either the Alpha splice variant of CXCL12, Beta splice variant of CXCL12, or both (Created in BioRender. Seir, G. (2025)) (C) CAR-T cell fold-expansion post transduction on day 0. (D) CAR expression on the surface of LiBa CAR-T cells quantified via flow cytometry (FLAG-tag expression). (E) CXCR4 expression comparison among Mock (M) transduced, Alpha (A), Beta (B), or Alpha-Beta (AB) CAR-T cells, and Nalm-6 ALL cell line quantified via flow cytometry. P < 0.0001 for Nalm-6 expression levels versus Mock, Alpha, Beta, and Alpha-Beta (n = 3, one-way ANOVA, Dunnett’s Multiple Comparisons Test) and P < 0.0001 for Mock versus Alpha, Beta, and Alpha-Beta (n = 3, one-way ANOVA, Dunnett’s Multiple Comparisons Test).

We next engineered a CXCR4 targeting CAR using its natural ligand, CXCL12. We designed the extracellular targeting domain based on the structured, globular domain of the Alpha (α) and/or Beta (β) splice variant of CXCL12 (UniProt:P48061) [44,45]. We designed three vectors expressing either the Alpha or Beta variant on the extracellular domain, or both variants combined via (G₄S)₃ linker in Alpha-Beta orientation relative to the cell membrane (Fig. 2B). Second generation constructs were generated in the same manner as scFv CAR constructs with the addition of a FLAG-tag to facilitate detection by flow cytometry. Retroviral vectors encoding individual CAR constructs were generated and primary human T cells were transduced and expanded. Excitingly, although the affinity of the natural ligands is similar to that of the anti-CXCR4 binders [28,54], the ligand-based (LiBa) CAR-T cells expanded as much as Mock CAR-T cells, which differs from other fratricide sensitive CAR-T cells previously explored in the literature [55,32] (Fig. 2C). Flow cytometric analysis of CAR expression using anti-FLAG tag antibody showed transduction efficiencies of 70–98% for the different LiBa CARs demonstrating that this lack of significant fratricide was not due to poor CAR expression (Fig. 2D).

Intrigued by the different outcomes between CXCR4 CAR-T cells engineered using an scFv or LiBa extracellular domains, and based on previous reports indicating that T cells express CXCR4 on their surface [56], we measured CXCR4 expression in LiBa CAR-T cells, and observed a strong downregulation of CXCR4, especially in those expressing the Alpha or Beta version (Supp Figure 4A). Importantly, we observed that T cells express significantly lower levels of CXCR4 than Nalm-6, which were further reduced in LiBa CAR T cells (Fig. 2E). Since we had not observed evidence of fratricide, this reduction in CXCR4 expression on LiBa CAR-T cells could be due to a masking effect, whereby binding of the CAR would interfere with detection by flow cytometry antibody, or internalization of CXCR4 upon binding with its ligand [57] (Supp Figure 4B). Of note, because we use a FLAG-tag for the detection of the LiBa CAR, any masking effect would not be noticeable in CAR expression.

To further investigate the relationship between CXCR4 and LiBa CARs expression, we engineered a truncated (T-LiBa) construct where both the CD28 costimulatory and CD3ζ domains were removed from the LiBa CAR (Fig. 3A). Analysis of CAR expression showed a small trend towards lower expression of the T-LiBa CARs (Fig. 3B). Furthermore, T-LiBa CARs also resulted in similarly low levels of CXCR4 expression compared to full length LiBa CARs, with a slight increase with Alpha-Beta constructs (Fig. 3C). These observations, together with the absence of fratricide, suggests that the LiBa CARs bind to CXCR4, driving internalization of CXCR4, without activating the CAR cytotoxic activity. This also suggests that the LiBa CARs expressing the Alpha or Beta isoforms separately, may have higher affinity for CXCR4 than the Alpha-Beta LiBa CAR construct, as evidenced by the higher amount of CXCR4 expressed on the surface of Alpha-Beta CARs (Fig. 3C). Alternatively, the higher levels of CXCR4 on the Alpha-Beta CAR could be due to the lower level of CAR expression, compared to the single Alpha or Beta LiBa CARs.

Fig. 3.

(A) Example structure of truncated LiBa (T-LiBA) CAR-T cells compared to the original LiBa construct. Truncated constructs had the same extracellular domain as the original construct but lack CD28 signaling and CD3ζ activation domains (Created in BioRender. Seir, G. (2025)). (B) CAR expression on the surface of T-LiBa CAR-T cells compared to Mock and LiBa CARs quantified via flow cytometry (FLAG-tag expression). (C) Comparison of CXCR4 expression on the surface of T-LiBa and LiBa CAR-T cells compared to Mock quantified via flow cytometry. (D) CXCR4 expression on the surface of LiBa, T-LiBa and Electric CAR-T cells stratified by CAR positivity.

To distinguish between these two possibilities, we compared CXCR4 expression on CAR+ cells (FLAG+) and CAR- cells (FLAG-) from the same culture and found consistently higher CXCR4 expression on all CAR- populations compared to CAR+ (Fig. 3D). Importantly, T-LiBa CARs show similar distribution of CXCR4 expression between CAR+ and CAR- population than full-length LiBa CAR, reinforcing that downregulation of CXCR4 is not associated with fratricide, since truncated CARs lack cytotoxic domains. Finally, we also performed these experiments with an extracellularly linked concatemeric trivalent cytokine (ELECTRIC) CAR control targeting the KIT, MPL, and FLT-3 receptors [22]. In the ELECTRIC CAR control, both CAR+ and CAR- populations overlapped in abundance of CXCR4, confirming that the CXCR4 downregulation was specific to the LiBa CARs. These results suggest that T cells express levels of CXCR4 sufficiently high to activate scFV-based CARs, resulting in fratricide, but sufficiently low to maintain LiBa CARs cytotoxic activity off despite ligand/receptor interaction.

CXCR4 targeting LiBa CARs target high and low antigen density ALL and AML cell lines

To test whether the LiBa CARs were functional, we cocultured LiBa CAR-T cells with GFP expressing Nalm6, which express very high levels of CXCR4 (Fig. 1D, Fig. 2E), at different effector to target (E:T) ratios. When we measured Nalm-6 viability 72 h later by flow cytometry, we observed that all three LiBa CAR constructs eliminated Nalm6 cells at 1:1 and 1:2 ratios (Fig. 4A). Importantly, the Beta and Alpha-Beta constructs seem to outperform the Alpha CARs, showing partial tumor control even at ratios as high as 1:16 (Fig. 4A).

Fig. 4.

(A) Mock transduced, Alpha, Beta, or Alpha-Beta CAR-T cells were co-cultured with Nalm-6 cells stably expressing GFP at different Effector:Target ratios. GFP expression was quantified via flow cytometry after 72 h, (n = 3, two-way ANOVA, Dunnett’s Multiple Comparisons Test). (B) Mock transduced, ELECTRIC, Alpha, Beta, or Alpha-Beta CAR-T cells were co-cultured with Nalm-6 (top) or Kasumi-1 (bottom) cells stably expressing GFP at a 1:1 Effector:Target ratio. GFP expression was measured in an IncuCyte machine over 63 h and normalized to initial expression. * indicates P < 0.05, ** indicates P < 0.01, *** indicates P < 0.001, **** indicates P < 0.0001, ns indicates P not significant as compared between a construct and Mock (n = 3, two-way ANOVA, Dunnett’s Multiple Comparisons Test).

Finally, we compared the antitumor efficacy of LiBa CARs against Nalm6 vs Kasumi-1, an AML cell line expressing lower levels of CXCR4 (Fig. 4B). For this comparison we co-cultured LiBa CARs with GFP expressing Nalm-6 and Kasumi-1 at 1:1 E:T ratios and tracked GFP signal intensity via IncuCyte over 63 h. For this assay, we included the ELECTRIC CAR control targeting the KIT, MPL, and FLT-3 receptors, which should function as negative controls against Nalm-6, and positive against Kasumi-1 [22]. First, we confirmed complete Nalm-6 clearance by all LiBa CAR constructs, with Alpha-Beta CARs showing a small advantage over Alpha CARs at earlier time points. Excitingly, Alpha-Beta CAR-T cells were also the most efficacious against Kasumi-1 cells, followed closely by Beta and the Alpha falling behind, although all CARs tested were able to clear most tumor cells by the final time point.

Together, our proof-of-concept experiments shows that CXCR4 targeting CARs engineered with different versions of CXCL12 in their extracellular domain are functional against AML and ALL tumor cell lines despite indications that they also bind to CXCR4 on the surface of T cells.

Discussion

Despite major recent advancements in the treatment of patients with acute leukemias, morbidity and mortality remain high. CAR-T cells have the potential to improve the efficacy to toxicity ratio of current therapies; however, they need to be further developed to enable treatment of all leukemias, especially as a bridge to HSCT. A major factor in designing effective CAR-T cells is the selection of an adequate target that is highly expressed and with overlapping expression profiles amongst all desired target cells [58]. CXCR4 has been identified as one such highly expressed receptor on the surface of various hematopoietic cells that plays a major role in hematopoiesis and in leukemic biology. Specifically, it has a major role in cell migration to the bone marrow niche which has been previously exploited in the development of targeted therapies that have reached clinical trials [26,28]. However, co-expression of CXCR4 on T cells has limited development of CAR-T cell immunotherapies capable of targeting these shared receptors. Here we show the first design and development of successful anti-CXCR4 CAR-T cells that express CAR and show activity against AML and ALL cell lines in vitro.

By analyzing publicly available bulk and single-cell RNA-seq data of primary pediatric AML and ALL samples [25,[36], [37], [38],40], we first validated that CXCR4 is highly expressed in pediatric AML, ALL, and hematopoietic cells, then validated protein expression on a variety of AML and ALL cell lines. We developed a panel of second generation anti-CXCR4 CAR-T cells, incorporating either scFv or CXCL12 ligand to promote anti-CXCR4 specificity. We observed major phenotypic differences in scFv and ligand-based (LiBa) anti-CXCR4 CAR-T cells, principally in the lack of viability of scFv-based CAR-T cells. We hypothesize that this lack of viability may be due to T-cell fratricide [30] as a result of the expression of CXCR4 on T-cells [24,30].

Excitingly, LiBa CAR-T cells expanded well in culture, exhibited CAR positivity as quantified by FLAG-tag expression using flow cytometry, and showed anti-tumor activity against Nalm-6 and Kasumi-1 in vitro at E:T ratios as low as 1:16. Differences in transduction efficiencies may have arisen from differences in retroviral integration of CAR transgene or construct changes in FLAG-tag accessibility for antibody staining. When interrogating CXCR4 expression, we found that it is downregulated in CAR-T cells expressing LiBa extracellular domains compared to control Mock or ELECTRIC CAR-T cells as well as the ALL cell line Nalm-6. We hypothesize that these observations may be due to the interaction of extracellularly linked CXCL12 and CXCR4 on the surface of the CAR-T cells (Supp Figure 4B). Self-binding of extracellularly linked CXCL12 may lead to either receptor internalization through a native mechanism when a cell is exposed to excess CXCL12 [59] or CXCR4 masking diminishing fluorescent antibody staining. This blocking interaction may also explain why LiBa CAR-T cells expand in culture as T cell fratricide is prevented as extracellular CXCL12 is blocking free CAR on other T cells from binding.

In conclusion, we have presented a proof-of-concept for the design and potential efficacy of anti-CXCR4 CAR-T cells. This research could be subsequently expanded to enable innovative new treatments for AML and ALL that ultimately result in safer, more effective therapies, and better outcomes for patients.

Data availability

Previously published bulk and single-cell RNA-seq data can be obtained from their original source [24,36,37,40]. Any other original data of this study can be made available from the corresponding authors upon reasonable request.

Appendix / supplementary materials

Additional files included.

Ethics approval

Donor buffy coats were obtained from the Stanford Blood Center under an IRB-exempt protocol.

Funding

This work was funded by Virginia and D.K. Ludwig Fund for Cancer Research (A. C.), American Society of Hematology Minority Hematology Graduate Award (to Q. R. B.), and Paul & Daisy Soros Fellowship for New Americans (to Q. R. B.).

CRediT authorship contribution statement

Gabriel Seir: Writing – review & editing, Writing – original draft, Validation, Formal analysis, Data curation, Conceptualization. Quenton Rashawn Bubb: Writing – review & editing, Validation, Supervision, Formal analysis, Data curation, Conceptualization. Elena Sotillo: Writing – review & editing, Supervision, Data curation. Tanja Gruber: Funding acquisition, Data curation. Rebecca M. Richards: Writing – review & editing, Supervision. Agnieszka Czechowicz: Writing – review & editing, Supervision, Resources, Funding acquisition, Conceptualization.

Declaration of competing interest

G.S., Q.R.B., E.S., R.M.R., and A.C. are inventors on a recently filed patent application related to separate CAR T-related work. Additionally, A.C. discloses financial interests in the following entities working on antibody-based conditioning approaches: Beam Therapeutics, Editas Medicines, GV, Inograft Biotherapeutics, Kyowa Kirin, and Prime Medicines. In addition, she is an inventor on antibody-based conditioning patents licensed to Jasper Therapeutics, Gilead Sciences, Inograft Biotherapeutics and Magenta Therapeutics. E.S. is a coinventor on multiple patents related to CAR-T cells including those from this work. E.S. consults for and holds equity in Lyell Immunopharma, and consults for Lepton Pharmaceuticals and Galaria.