FGF18 mediates fibroblast-leukemia cross talk to promote acute myeloid leukemia progression

Key Points

• •Fibroblast-derived FGF18 binds to FGFR3 on AML cells to drive leukemic progression and suppress CD8⁺ T-cell effector function.
• •FGF18 neutralization disrupts stromal-leukemic cross talk and enhances anti–PD-1 immunotherapy efficacy in AML.

Visual Abstract

Abstract

Nonhematopoietic stromal cells are essential regulators of hematopoiesis; however, their contribution to leukemogenesis and immune dysfunction remains poorly defined. In this study, we identified fibroblast-derived fibroblast growth factor 18 (FGF18) as a novel stromal cytokine that reprograms leukemia-immune interactions. Single-cell RNA sequencing of the bone marrow niche in acute myeloid leukemia (AML) revealed the upregulation of Fgf18 in stromal fibroblasts. Administration of recombinant FGF18 accelerated AML progression, whereas fibroblast-specific Fgf18 depletion markedly delayed disease development and improved the survival of mice. We performed a pooled CRISPR-Cas9 screen in AML cells and identified FGF receptor 3 (FGFR3) signaling as a critical mediator of leukemic fitness in the FGF18-rich microenvironment. Genetic loss of Fgfr3 in AML cells recapitulated the effects of FGF18 deficiency and limited leukemic expansion in vivo. Mechanistically, FGF18 binds to its receptor, FGFR3, on AML cells, activating the protein kinase B (AKT)-mechanistic target of rapamycin (mTOR) (AKT-mTOR) signaling pathway and inducing interleukin-6 (IL-6) production. IL-6 acts autocrinely to reinforce leukemic signaling and paracrinely to activate fibroblast JAK-STAT3 signaling, thereby amplifying stromal fibroblast FGF18 expression and forming a feed-forward loop that suppresses CD8⁺ T-cell effector function and weakens antileukemic immunity. Clinically, elevated FGF18 expression correlates with poor prognosis in patients with AML. To therapeutically target this malignant cross talk, we generated an FGF18-neutralizing antibody that disrupted the stromal-leukemia feedback loop, restored CD8⁺ T cell effector function, and synergized with anti–PD-1 therapy to elicit durable antileukemic immunity in vivo. Collectively, these findings identify an FGF18-dependent stromal-leukemia cross talk axis that drives AML progression and immune dysfunction, highlighting FGF18 neutralization as a potential therapeutic strategy.

It is now well recognized that there is a dynamic interaction between acute myeloid leukemia (AML) cells and the bone marrow (BM) niche, but how to exploit this is unclear. Zhang and colleagues identified fibroblast growth factor 18 (FGF18) secreted by BM fibroblasts as critical for the support of AML by promoting proliferation and stimulating interleukin-6 production, which in turn drives T-cell exhaustion and strengthens stromal support of the AML. In murine models, interference with FGF18 slows AML progression and enhances efficacy of immune checkpoint interventions, suggesting that targeting FGF18 warrants further assessment.

Introduction

Acute myeloid leukemia (AML) is the most common acute leukemia in adults and remains highly aggressive, with a 5-year survival rate of <40%. Despite therapeutic advances, treatment resistance and disease relapse remain major challenges associated with AML.^{1, 2, 3, 4} Beyond leukemia-intrinsic mutations, the poor prognosis of AML is increasingly attributed to profound remodeling of the bone marrow (BM) microenvironment, which provides metabolic, immune, and survival signals that enable leukemic cells to resist therapy.^{5, 6, 7}

The BM microenvironment is a network of hematopoietic, immune, and stromal cells that collectively regulate hematopoietic and immune homeostasis.^8,9 Within this niche, stromal cell subsets, including fibroblasts, osteoblasts, adipocytes, endothelial cells (ECs), and mesenchymal stromal cells (MSCs), are pivotal in maintaining hematopoietic stem cell function.^{10, 11, 12, 13} In AML, these cells are reprogrammed to support leukemic expansion; however, the molecular basis of stromal-leukemic cross talk remains incompletely understood. Increasing evidence indicates that stromal-derived cytokines reshape the BM milieu to affect leukemic survival.^{14, 15, 16, 17, 18} We recently reported that tuberous sclerosis complex 1–deficient osteoclasts release interleukin-34 (IL-34) to drive myeloid differentiation, exemplifying how dysregulated stromal signaling can perturb normal hematopoiesis.¹⁹ These findings prompted us to identify additional stromal mediators involved in AML niche remodeling, particularly fibroblast growth factor (FGF) family members, as candidates for intercellular communication signals.^{20, 21, 22} Although FGF18 has been studied in bone biology and tissue regeneration, its role in malignancy, especially in hematologic cancers, remains unclear.^{23, 24, 25}

Microenvironmental remodeling in AML is a bidirectional process. Leukemic cells adapt to their surroundings and actively reshape them by releasing paracrine factors that modulate stromal and immune components.²⁶ Inflammatory cytokines such as IL-6, IL-1β, and tumor necrosis factor-α are well recognized for their contribution to microenvironmental alterations and the disruption of hematopoietic homeostasis.^{27, 28, 29, 30} IL-6, in particular, promotes leukemic growth and fosters T-cell exhaustion, thereby enabling immune escape. Consequently, immune checkpoint inhibitors have also been explored, yet their clinical efficacy in AML remains limited.^{31, 32, 33, 34}

In this study, we identified FGF18 as a stromal-derived cytokine that mediates fibroblast-leukemic cross talk. FGF18 levels in BM supernatants are markedly elevated in patients with AML and correlate with poor prognosis. To explore the interactions between the stroma and AML, which are essential for leukemogenesis, we focused on receptors critical for AML progression in vivo and found that FGF receptor 3 (FGFR3) is associated with unfavorable outcomes. Mechanistically, FGF18 binds to FGFR3 on leukemic blasts, activating the phosphoinositide 3-kinase (PI3K)-AKT-mTOR pathway and promoting leukemic proliferation and survival. In turn, FGF18-stimulated AML cells secrete IL-6, which acts in an autocrine manner to reinforce leukemic signaling and in a paracrine manner to reactivate stromal fibroblasts through JAK-STAT3, thereby amplifying FGF18 production and forming a feed-forward loop. By developing an FGF18-neutralizing monoclonal antibody, we demonstrated that therapeutic blockade of FGF18 restores CD8⁺ T-cell effector function and, in combination with anti–PD-1 (aPD-1) therapy, elicits effective antileukemic immunity. Collectively, these results identify the FGF18-FGFR3 axis as a promising therapeutic target for AML.

Methods

BM stromal cells were isolated from AML and control mice to capture the nonhematopoietic BM compartment. Single-cell RNA sequencing (scRNA-seq) and Smart-seq2 (switching mechanism at 5′ end of the RNA transcript 2) profiling were performed to define stromal heterogeneity. Fibroblasts were sorted from Pdgfra^CreERT2;Rosa26^mT/mG reporter mice, ECs were isolated from Cdh5^Cre;Rosa26^mT/mG mice, MSCs were sorted from Prrx1^Cre;Rosa26^mT/mG reporter mice, and osteoblasts were obtained from Ocn^Cre;Rosa26^mT/mG mice. Fibroblast-specific Fgf18 conditional knockout (KO) mice (Pdgfra^CreERT2;Fgf18^flox/flox) and patient-derived xenograft (PDX) models were used to evaluate stromal FGF18 function in vivo. A genome-wide CRISPR-Cas9 screen was performed in AML cells to identify the specific FGF18 receptor responsible for promoting leukemic progression. Coculture systems were used to examine FGF18-driven leukemic growth and T-cell modulation, and the interaction between FGF18 and FGFR3 was verified by coimmunoprecipitation. The therapeutic efficacy of FGF18-neutralizing antibodies, alone or in combination with aPD-1, was tested. Human BM and serum samples were examined for FGF18 expression and clinical correlations. Detailed methods are provided in the supplemental Materials and Methods, available on the Blood website.

Results

Remodeling of BM stroma and identification of Fgf18⁺ fibroblasts in AML

Leukemic-stromal cross talk supports leukemia survival and drug resistance.^{35, 36, 37} Our previous study identified osteoclasts as key stromal regulators of leukemia development.¹⁹ However, the contributions of other stromal subsets remain poorly characterized. To map the nonhematopoietic BM stromal landscape during AML progression, BM stromal cells (Lin^-CD45^-CD71^-) were isolated from MLL-AF9 leukemic and control mice and profiled using scRNA-seq (Figure 1A-B). After hematopoietic exclusion and quality filtering, 26 429 cells were retained (control, n = 11 651; AML, n = 14 778). Unbiased clustering resolved 21 lineage clusters encompassing ECs, fibroblasts, pericytes, MSCs, and osteo-associated populations (Figure 1C-E; supplemental Figure 1A-C), revealing stromal remodeling in AML. Extracellular matrix (ECM)–receptor interaction and focal adhesion pathways were downregulated in fibroblasts, suggesting impaired fibroblast function in AML (supplemental Figure 1D).

Figure 1.

Single-cell transcriptomic analysis reveals remodeling of the BM stroma and Fgf18 upregulation in stromal fibroblasts in AML. (A) Schematic illustration of the experimental workflow for the isolation and analysis of BM stromal cells from control and AML mice. (B) Flow cytometry plots showing the gating strategy and percentage of cells in the BM stromal cell population. Sequential gates were applied to exclude debris and doublets, followed by identification of live, single cells. Representative plots display (from left to right) total cells, single cells, live cells, lineage cells (Alexa Fluor 700), and CD45 markers (FITC) and the CD71 gate (PE). (C) UMAP visualization of scRNA-seq data from BM stromal cell populations. (D) Proportional bar charts comparing the relative abundance of stromal cell subpopulations in control and AML BM samples. (E) Dot plot of representative marker gene expression for each stromal cell type. (F) Top: Relative luminescence of MLL-AF9 and C1498 cells cultured in the absence (−) or presence (+) of primary murine BM stromal cell-derived CM for 24 and 48 hours (n = 3). Bottom: Relative luminescence of C1498 cells treated with the indicated agents (sorafenib, venetoclax, doxorubicin, fludarabine, decitabine, and cytarabine) with or without CM (n = 3). Data are shown as relative luminescence units, with the −CM condition used as a reference control. (G) Heat map showing differentially expressed genes in BM stromal cells from control and AML mice. (H) Venn diagram integrating scRNA-seq, Smart-seq2, and secretome (UniProt) data sets, showing their intersection and highlighting the 4 commonly upregulated secreted factors (Bdnf, Reg4, Fgf18, and Lif) in leukemic stroma. (I) Quantitative reverse transcription polymerase chain reaction (qRT-PCR) analysis of Bdnf, Reg4, Fgf18, and Lif mRNA levels in BM stromal cells from control and AML mice. For each gene, expression is shown as a fold change relative to the corresponding control group (set to 1). (J) UMAP feature plots visualizing the expression of Fgf18 and Postn across BM stromal subpopulations in AML and non-AML control samples. (K) Bar graph showing the relative expression of Fgf18 in each major stromal cell type in control and AML mice. (L) Dot plot showing the percentage of cells and average scaled expression of Fgf18 in fibroblasts from the control and AML samples. Data are presented as mean ± standard error of the mean (SEM). Data in panels F, I were analyzed using a 2-tailed Student t test. FITC, fluorescein isothiocyanate; FSC-A, forward scatter area; PE, phycoerythrin; SSC-A, side scatter area.

To determine whether these microenvironmental changes contribute to leukemogenesis, we focused on stromal-derived factors that mediate leukemia-niche interactions. This was motivated by the finding that conditioned medium (CM) from murine BM stromal cells promoted AML cell line proliferation and reduced sensitivity to multiple chemotherapeutic agents (Figure 1F). To identify the specific stromal factors responsible for these proleukemic effects, Smart-seq2 transcriptomic analysis was performed on BM stromal cells isolated from MLL-AF9 AML and nonleukemic control mice (Figure 1G; supplemental Figure 2A). Integration of the single-cell transcriptomic, bulk transcriptomic, and secretome data sets identified four key secreted factors at their intersection that were robustly upregulated in the leukemic stroma, including Bdnf, Reg4, Fgf18, and Lif (Figure 1H). Among these candidates, FGF18 showed the most pronounced increase in MLL-AF9 AML mice compared with nonleukemic controls (Figure 1I; supplemental Figure 2B). Uniform manifold approximation and projection (UMAP) mapping and gene quantification showed that Fgf18 was specifically enriched in AML fibroblasts, with no upregulation in other stromal populations (Figure 1J-L; supplemental Figure 2C-E). Together, these data identify remodeled fibroblasts as the key source of FGF18 and implicate Fgf18 in leukemia-stroma cross talk in AML.

Fibroblast-specific secretion of FGF18 contributes to AML progression

The specificity of Fgf18 expression in fibroblasts was further validated using multiple genetically modified mouse models, including fibroblast-specific Pdgfra^CreERT2;Rosa26^mT/mG, osteoblast-specific Ocn^Cre;Rosa26^mT/mG, endothelial-specific Cdh5^CreERT2;Rosa26^mT/mG, and MSC-specific Prrx1^Cre;Rosa26^mT/mG mice (Figure 2A). Green fluorescent protein-positive (GFP⁺) BM stromal cell populations were sorted using flow cytometry, and the results revealed significantly higher Fgf18 expression in PDGFRA-GFP⁺ fibroblasts from MLL-AF9 AML mice than in other cell types, including osteocalcin (Ocn)-GFP⁺ osteoblasts, cadherin 5 (Cdh5)-GFP⁺ ECs, Prrx1-GFP⁺ MSCs, and CD11b⁺Ly6C⁺ monocytes (Figure 2B-D). Immunofluorescence staining of hindlimb bone sections showed colocalization of PDGFRA and FGF18, with markedly higher FGF18 expression in AML mice than in non-AML controls (Figure 2E).

Figure 2.

Fibroblast-specific upregulation of FGF18 in the leukemic BM microenvironment is associated with poor prognosis in AML. (A) Schematic overview of experimental workflow. Genetically modified reporter mice (Pdgfra^CreERT2;Rosa26^mT/mG for fibroblasts, Ocn^Cre;Rosa26^mT/mG for osteoblasts, Cdh5^CreERT2;Rosa26^mT/mG for ECs, and Prrx1^Cre;Rosa26^mT/mG for MSCs) were used to trace and isolate BM stromal subsets for downstream analysis. (B) Bar graph showing Fgf18 expression levels in freshly isolated BM stromal and myeloid cell populations from control (nonleukemic) and AML mice (n = 3). Expression levels are presented relative to the corresponding non-AML population for each cell type. (C) Immunoblot analysis showing FGF18 protein expression in sorted stromal and myeloid cell populations from non-AML (N) and AML mice (A). (D) Representative flow cytometry histograms and quantification of FGF18 expression in BM stromal cells from non-AML and AML mice (n = 7). (E) Immunofluorescence staining of BM sections showing colocalization of PDGFRA depicted in red (fibroblast marker) and FGF18 in green, in control and AML mice. DAPI (4′,6-diamidino-2-phenylindole) stains the nuclei blue, and merged and bright-field images are shown. Scale bars, 50 μm. Quantification analysis was performed using the intensity of FGF18 immunostaining on the bone surface (n = 5). (F) Quantification of FGF18 protein concentration in BM interstitial fluid from non-AML and AML mice by enzyme-linked immunosorbent assay (ELISA; n = 6). (G) Relative FGF18 mRNA in BM stromal cells isolated by enzymatic digestion of BM biopsy specimens from patients who did not have AML and patients with AML (n = 5). (H) Immunohistochemical staining images and quantification of FGF18 in BM sections from patients who did not have AML and patients with AML (n = 5). Scale bars, 100 μm. (I) FGF18 levels in BM aspirate supernatants from newly diagnosed patients with AML (n = 51), relapsed patients (n = 15), remission patients (n = 46), and healthy donors (n = 8). Data are presented as mean ± SEM. Data in panel I were analyzed using 1-way analysis of variance (ANOVA) with Šídák correction for multiple comparisons, and data in panels B,D-H were analyzed using a 2-tailed Student t test. FACS, fluorescence-activated cell sorter.

The clinical relevance of FGF18 in AML was further supported by an approximately sixfold elevation in FGF18 levels in leukemic BM interstitial fluid compared with FGF18 levels in controls (Figure 2F). Consistently, FGF18 mRNA levels were significantly higher in BM aspirates from patients with AML than in those from non-AML individuals (Figure 2G; supplemental Table 1), and immunohistochemistry showed strong FGF18 staining in AML BM sections but low staining in controls (Figure 2H). In the patient cohort, FGF18 levels in BM supernatants were higher in newly diagnosed or relapsed AML than in samples from patients in complete remission and healthy donors (Figure 2I; supplemental Table 2). Together, these data suggest that stromal fibroblast-derived FGF18 contributes to AML pathogenesis and is associated with poor clinical outcomes.

To determine whether elevated FGF18 reflects AML-specific induction or expansion of a hematopoietic population, we analyzed FGF18 expression across purified hematopoietic subsets. FGF18 levels were comparable between AML and non-AML controls in sorted monocytes, neutrophils, T cells, and hematopoietic stem and progenitor cells (supplemental Figure 3A-C). In contrast, Fgf18 was enriched in nonhematopoietic stromal cells in the MLL-AF9 model (supplemental Figure 3D-E). Consistently, BM cells or CD11b⁺ myeloid cells from AML, but not control mice, induced robust FGF18 upregulation in stromal cells upon coculture (supplemental Figure 3F-G). These data indicate that elevated FGF18 in AML arises from leukemia-driven stromal remodeling.

FGF18 promotes AML progression

To determine whether FGF18 drives leukemic progression, we examined its gain- and loss-of-function effects in AML models. Recombinant FGF18 promoted the proliferation of AML cell lines and patient-derived blasts in a dose- and time-dependent manner and significantly increased their colony-forming capacity (supplemental Figure 4A-B). In vivo, FGF18 administration in MLL-AF9-GFP⁺, Molm13-luc-GFP⁺, and PDX AML models (Figure 3A; supplemental Figure 5A; supplemental Table 3) increased peripheral blood (PB) white blood cell (WBC) counts, enhanced leukemic tissue infiltration, induced splenomegaly, and markedly shortened survival (Figure 3B-F; supplemental Figure 5B-G). These findings highlight the critical role of FGF18 in AML progression and poor outcomes.

Figure 3.

FGF18 promotes AML progression through stromal fibroblast-derived FGF18 signaling. (A) Schematic representation of the experimental design. NCG mice were either left untransplanted (non-AML) or IV transplanted with Molm-13-luc cells or PDX AML cells. Six days after transplantation, mice were treated i.p. every other day with vehicle or rFGF18 at doses of 0.01, 0.1, or 1 μg. Treatment started on day 7, and mice were analyzed on day 21. (B) WBC counts in PB on day 21 in non-AML and AML mice treated with rFGF18 (n = 6). (C) Bioluminescence imaging of leukemic burden in Molm-13-luc-transplanted mice over time (days 7, 14, 21, and 28). (D) Quantification of BM infiltration by human AML cells (BM hCD45⁺%) on day 21 with rFGF18 treatment (n = 5). (E) Kaplan-Meier survival curves for Molm-13-luc models treated with vehicle or rFGF18 (n = 8). (F) Kaplan-Meier survival curves for 4 independent PDX models (PDX-1 to PDX-4) treated with vehicle or rFGF18 (n = 8). (G) Immunofluorescence analysis of bone tissue from Pdgfra^CreERT2;Fgf18^flox/flox and control mice. Scale bars, 50 μm. Quantification analysis was performed based on the intensity of FGF18 immunostaining in the BM stromal compartment (n = 5). (H) Schematic illustration of the experimental design of the study. MLL-AF9-GFP⁺ or C1498-GFP⁺ leukemic cells were injected into Pdgfra^CreERT2;Fgf18^flox/flox mice and their control littermates. Mice were treated with vehicle or rFGF18 every other day, starting on day 7 after injection. (I) Flow cytometry analysis of GFP⁺ leukemic cells in the BM of mice injected with MLL-AF9-GFP⁺ cells (n = 6). (J) Spleen weights of mice injected with MLL-AF9-GFP⁺ cells (n = 6). (K-L) Kaplan-Meier survival curves of mice injected with MLL-AF9-GFP⁺ (K) or C1498-GFP⁺ (L) leukemic cells (n = 10). Data are presented as mean ± SEM. Statistical significance was determined using a 2-tailed unpaired Student t test (panel G), 1-way ANOVA with Tukey post hoc test for multiple comparisons (panels B,D, I-J), or the log-rank test for survival analysis (panels E-F,K-L). i.p., intraperitoneally; NCG, NOD/ShiLtJGpt-Prkdc^em26Cd52Il2rg^em26Cd22/Gpt.

To test whether fibroblast-derived FGF18 is required for AML progression, we generated an inducible, conditional Fgf18-KO model (Pdgfra^CreERT2;Fgf18^flox/flox; supplemental Figure 6A-B). Efficient deletion of Fgf18 in fibroblasts was confirmed (Figure 3G; supplemental Figure 6C). Primary fibroblast-like stromal cells from Pdgfra^CreERT2;Fgf18^flox/flox or control mice were cocultured with MLL-AF9 or C1498 AML cells. FGF18-deficient fibroblasts, or their CM, significantly reduced AML cell proliferation compared with control fibroblasts, whereas AML monoculture showed no difference (supplemental Figure 6D-E). To further evaluate the in vivo effects of fibroblast-derived FGF18 loss, MLL-AF9-GFP⁺ and C1498-GFP⁺ leukemia cells were transplanted into Pdgfra^CreERT2;Fgf18^flox/flox and control mice, followed by treatment with either vehicle or recombinant FGF18 (Figure 3H). Conditional deletion of Fgf18 in Pdgfra⁺ fibroblast-like stromal cells significantly reduced the number of GFP⁺ leukemic cells in the BM (Figure 3I; supplemental Figure 6F), accompanied by reduced spleen weight and tissue infiltration (Figure 3J; supplemental Figure 6G-H). Consistent with a proleukemic role, FGF18 worsened the disease in control mice, whereas its effects were attenuated in Pdgfra^CreERT2;Fgf18^flox/flox mice, resulting in a significantly prolonged survival (Figure 3K-L).

FGF18 binds to FGFR3 and activates the PI3K-AKT-mTOR signaling pathway

Transcriptomic profiling of Molm-13 cells treated with recombinant FGF18 showed enrichment of leukocyte activation (supplemental Figure 7A) and critical signaling pathways, including the mTOR and PI3K-AKT pathways (Figure 4A). Western blot analysis further confirmed that FGF18 activated the PI3K-AKT-mTOR signaling pathway, as evidenced by the increased phosphorylation of PI3K, AKT, and mTOR (Figure 4B; supplemental Figure 7B), which was reversed by the AKT inhibitor MK-2206 or the mTOR inhibitor rapamycin (Figure 4C).

Figure 4.

FGF18 promotes AML progression through FGFR3-mediated activation of the PI3K-AKT-mTOR pathway. (A) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis showing the significantly enriched pathways associated with FGF18 signaling. (B) Western blot analysis of PI3K-AKT-mTOR signaling pathway activation in Molm-13 cells treated with increasing concentrations of FGF18 (0, 25, 50, 100, and 200 ng/mL) for 24 hours. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as the loading control. (C) Representative western blot showing phosphorylated and total PI3K, AKT, and mTOR in Molm-13 cells treated with recombinant FGF18 in the absence or presence of pathway inhibitors. Lanes including FGF18−, FGF18+ (50 ng/mL, 24 hours), FGF18 + MK-2206 (5μM, 24 hours), and FGF18 + rapamycin (1μM, 24 hours). GAPDH was used as the loading control. (D) Schematic illustration. C1498-Cas9-GFP cells were transduced with a library of sgRNAs using a pooled screening format. These transduced cells were subsequently injected into mice with the transgenic genotype Pdgfra^CreERT2;Fgf18^flox/flox. After FGF18 administration, GFP⁺ cells were isolated from the mouse model for further sequencing and data analysis. (E) Ranked distribution of β-scores derived from independent genome-wide CRISPR screens conducted in the presence (left) or absence (right) of FGF18. A total of 21 373 high-quality genes are ranked by their respective degrees of enrichment or depletion. Colored circles highlight representative functional hits exhibiting significant positive or negative selection within each screening context. (F) Reactome, KEGG, and Gene Ontology (GO) enrichment analyses of significant CRISPR screen hits. (G) Representative genes from enriched pathway modules among CRISPR screen hits, grouped by the functional pathways. (H) Coimmunoprecipitation analysis showing the interaction between FGF18 and FGFR3 in Molm-13 cells. (I) Immunofluorescence imaging of Molm-13 cells showing colocalization of FGF18 (red) and FGFR3 (green). DAPI (blue) was used to stain the nuclei. Scale bar, 20 μm. (J) Western blot analysis of FGFR3 expression in C1498 and MLL-AF9 cells cultured in control medium (−CM) or primary murine BM stromal cell-CM (+CM). (K) Flow cytometry analysis of GFP⁺ Molm-13 cells in the BM of mice transplanted with negative control (NC) or FGFR3-KO Molm-13 cells, followed by treatment with phosphate-buffered saline (PBS) or rFGF18 (n = 6). (L) Kaplan-Meier survival curves of AML mice transplanted with NC or FGFR3-KO Molm-13 cells and treated with PBS or FGF18 (n = 8). (M) Relative mRNA expression of FGFR3 in peripheral blood mononuclear cells (PBMCs) from patients with AML (n = 27) compared with that in PBMCs from healthy donors (n = 12). Expression was normalized to GAPDH; values are relative to the healthy donor reference, set to 1. (N) Percentage of AML blasts from patients expressing FGFR3, determined using flow cytometry (isotype control vs FGFR3-specific antibody; n = 12). Data in panel K were analyzed by 1-way ANOVA with Šídák correction for multiple comparisons, data in panels M-N were analyzed by 2-tailed Student t test, and the log-rank test was used for survival analysis in panel L. HD, healthy donor.

To define the niche-derived signaling pathway, we conducted a CRISPR/Cas9-based genome-wide library screen in AML cells with or without FGF18 stimulation (Figure 4D). This screen identified several FGF18-dependent AML effectors required for leukemic cell survival, including Fgfr3, Fn1, and Dock11, as well as AML fitness genes, such as Cd27 and Dph5, which are associated with tumor promotion and immune evasion. In addition, Ptpn20 and Fas emerged as FGF18-dependent AML inhibitors, whereas Sod1 and Spry3 were classified as AML suppressor effectors that constrain leukemia independently of FGF18 (Figure 4E). Pathway enrichment analysis highlighted cytokine signaling, JAK-STAT signaling, and IL-6 family signaling (Figure 4F-G).

Next, we defined functionally relevant receptors using an in vivo genome-wide CRISPR screen and selected transmembrane receptors whose single-guide RNAs (sgRNAs) were depleted by at least twofold. Among the 18 upregulated receptors associated with FGF18 in human AML cells, we prioritized FGFR3, a known FGF18 receptor (supplemental Figure 7C-F). FGFR3 was the most abundantly expressed FGFR family member in AML cells (supplemental Figure 7G-H). CRISPR/Cas9-mediated deletion of FGFR3, but not FGFR1, FGFR2, or FGFR4, completely abolished FGF18-induced cell proliferation (supplemental Figure 7I). Coimmunoprecipitation and immunofluorescence confirmed direct binding and colocalization of FGF18 with FGFR3 (Figure 4H-I). Consistently, culturing AML cells in primary stromal fibroblast- CM increased FGFR3 expression (Figure 4J). In FGFR3-deficient AML cells, FGF18-induced activation of the PI3K-AKT-mTOR pathway and associated proinvasive effects were completely lost, leading to impaired leukemia progression and prolonged survival in vivo (Figure 4K-L; supplemental Figure 7J). Further analysis of clinical samples revealed significantly higher FGFR3 expression in peripheral blood mononuclear cells from patients with AML than in those from healthy donors (Figure 4M; supplemental Figure 7K; supplemental Table 4), and flow cytometry confirmed FGFR3 expression on the surface of primary CD33⁺ AML blasts (Figure 4N). This pathological upregulation was further characterized by a distinct alternative splicing signature. Notably, alternative splicing analysis revealed that AML cells preferentially expressed the high-affinity FGFR3-C isoform, whereas normal cells predominantly expressed the FGFR3-A isoform, a secreted, nonsignaling variant (supplemental Figure 8). Thus, beyond aberrant expression levels alone, the functional oncogenic potential of FGFR3 in AML may be further amplified by tumor-specific transcript maturation. These findings identify the FGF18-FGFR3 axis as a critical driver of AML cell proliferation through PI3K-AKT-mTOR activation, supporting FGFR3 as a promising therapeutic target.

FGF18 stimulates IL-6–mediated activation of the E2F1-MYC/ID1 axis, enhancing AML cell proliferation

Given the strong association between FGF18 and cytokine-mediated signaling (Figure 4A), we profiled cytokines and chemokines in AML cells and identified IL-6 as the most strongly induced factor (Figure 5A). Treatment of Molm-13 and OCI-AML3 cells with recombinant FGF18 markedly increased IL-6 secretion and protein expression, whereas MK-2206 or rapamycin suppressed this effect, demonstrating that FGF18-induced IL-6 expression depends on the AKT-mTOR signaling pathway (Figure 5B-C). Further functional analysis revealed that FGF18-driven AML cell proliferation was dependent on IL-6. An IL-6-neutralizing antibody (IL-6 Ab) or IL-6 knockdown partially reversed the FGF18-induced proliferative effect in AML cells, whereas exogenous recombinant IL-6 partially rescued the inhibitory effects of AKT and mTOR knockdown on FGF18-induced proliferation (Figure 5D-F; supplemental Figures 9A-E and 10A-E). Consistently, serum IL-6 levels increased after FGF18 treatment in the C1498 and MLL-AF9 models (Figure 5G), and FGF18 and IL-6 showed a positive correlation in BM aspirate supernatants from patients with AML (Figure 5H). These findings demonstrate the critical role of IL-6 in mediating the proproliferative effects of FGF18 in AML cells.

Figure 5.

FGF18 promotes IL-6 production and enhances AML cell proliferation through E2F1 activation. (A) Heat map showing differential gene expression in Molm-13 cells treated with recombinant human (rh) FGF18 or control. (B-C) ELISA quantification of IL-6 levels in the supernatant of Molm-13 and OCI-AML3 cells treated with rhFGF18, with or without MK-2206 or rapamycin (n = 3). Western blot analysis of IL-6 protein levels in the cell lysates is shown below. (D) Relative luminescence of Molm-13 cells treated with rhFGF18 and rhIL-6, with AKT knockdown (n = 3). Data are expressed as fold change relative to the untreated group (baseline = 1). (E) Relative luminescence of Molm-13 cells treated with rhFGF18 and rhIL-6, with mTOR knockdown (n = 3). Data are expressed as fold change relative to the untreated group (baseline = 1). (F) Relative luminescence of Molm-13 cells treated with rhFGF18 and/or rhIL-6 or IL-6 Ab, as indicated (n = 3). Data are expressed as fold change relative to the untreated group (baseline = 1). (G) ELISA measurements of IL-6 concentrations in the serum of C1498 and MLL-AF9 mouse AML models treated with various concentrations of FGF18 (n = 3). (H) Correlation analysis between FGF18 and IL-6 levels in BM aspirate supernatants from patients with AML (n = 22). (I) Western blot analysis showing IL-6 expression in AML cells (NC) and FGFR3-KO AML cells treated with FGF18. GAPDH was used as the loading control. (J) Relative mRNA expression levels of Il-6 in fibroblast-like stromal cells and C1498 cells under different conditions, including control; coculture of fibroblast-like stromal cells with C1498 cells (Co-C1498); C1498 cells treated with FGF18 (FGF18); and cells cultured with fibroblast supernatant or in direct coculture for 6 hours. Data are normalized to control (set to 1) (n = 3). (K) ELISA analysis of IL-6 levels in fibroblast-like stromal cells and C1498 cells under the same conditions as in panel J (n = 3). (L) Heat map of gene expression changes detected by PCR array in Molm-13 cells treated with IL-6 vs control. (M) qPCR analysis of E2F1 mRNA expression in Molm-13 cells treated with rhFGF18, rhIL-6, and/or IL-6 Ab (n = 3). qPCR data were normalized to GAPDH (reference gene) and expressed relative to the untreated control (set to 1). (N) Relative luminescence of Molm-13 cells treated with rhIL-6 with or without E2F1 knockdown (n = 3). Data are presented as mean ± SEM. Statistical significance was determined using 1-way ANOVA with Tukey post hoc test (panels B-G,J-N). Ctrl, control; F-sup., cells cultured with fibroblast supernatant; F-coc., cells cultured in direct coculture.

To determine the cellular source of IL-6, we assessed IL-6 expression in AML cells and stromal fibroblasts. FGF18 elevated IL-6 levels in C1498 cells, and this induction was eliminated by FGFR3 KO, demonstrating that FGFR3 signaling is required for FGF18-induced IL-6 expression (Figure 5I). In contrast, IL-6 levels in stromal fibroblasts remained unchanged after coculture with AML cells. Instead, AML cells with high basal IL-6 expression exhibited further IL-6 induction upon FGF18 treatment, coculture, or exposure to fibroblast supernatants (Figure 5J-K). Although C1498 AML cells exhibited high IL-6 expression in vitro, after in vivo establishment of the model and subsequent isolation of AML cells from mice, IL-6 expression in AML cells was markedly increased (supplemental Figure 10F). These results indicate that IL-6 production primarily originates from AML cells, supporting AML-niche cross talk.

Because IL-6 itself is known to promote AML cell proliferation, we next explored the molecular mechanisms underlying IL-6-mediated effects in AML cells. Using an RT² Profiler PCR array, we profiled transcription factor (TF) expression in Molm-13 cells. Among the 96 TFs analyzed, IL-6 treatment prominently upregulated E2F1 (Figure 5L). JASPAR database analysis further identified E2F1 as a key TF predicted to regulate MYC and ID1 expression. Consistently, IL-6 increased E2F1 expression, an effect enhanced by recombinant human FGF18 and abolished by IL-6 neutralization (Figure 5M; supplemental Figure 10G). E2F1 knockdown blocked IL-6-induced cell proliferation and MYC and ID1 upregulation (Figure 5N; supplemental Figure 10H-I). In addition, IL-6 treatment induced marked nuclear accumulation of E2F1 (supplemental Figure 10J). Chromatin immunoprecipitation assays confirmed the direct binding of E2F1 to the MYC and ID1 promoters, with enhanced recruitment upon IL-6 stimulation in AML cells (supplemental Figure 10K). Collectively, these findings demonstrate a critical role for IL-6 in mediating the proproliferative effects of FGF18 in AML cells.

JAK-STAT3 signaling is essential for FGF18 expression in fibroblast-like stromal cells and is activated by IL-6 in a paracrine manner

To explore how IL-6 regulates FGF18, we analyzed the gene expression profiles of PDGFRA-GFP⁺ fibroblast-like stromal cells sorted from the BM stroma of phosphate-buffered saline or IL-6-treated C1498 mouse models of AML. Kyoto Encyclopedia of Genes and Genomes enrichment analysis revealed the significant activation of the JAK-STAT3 signaling pathway (Figure 6A). Consistently, IL-6 treatment significantly upregulated Fgf18 expression, along with FGF18 secretion, in PDGFRA-GFP⁺ fibroblast-like stromal cells compared with controls, whereas this increase was attenuated by an IL-6 Ab or a JAK1 inhibitor (Figure 6B-C; supplemental Figure 10L). In primary fibroblasts from AML mice, coculture with C1498 cells significantly enhanced JAK1 and STAT3 phosphorylation, whereas JAK inhibitor treatment abolished this activation and suppressed FGF18 expression (Figure 6D). To further elucidate the role of IL-6 in JAK-STAT3 activation, fibroblasts were either cocultured with C1498 cells or treated with AML cell-derived supernatants, with or without IL-6 Ab. In both cases, p-JAK1 and p-STAT3 were significantly upregulated in PDGFRA-GFP⁺ fibroblast-like stromal cells, and these effects were blocked by IL-6 Ab (Figure 6E-F). These data establish an IL-6–driven paracrine loop that mediates JAK-STAT3–dependent FGF18 induction in fibroblasts.

Figure 6.

IL-6 secretion by AML cells activates the JAK-STAT3 signaling pathway in fibroblasts, promotes FGF18 production, and suppresses CD8⁺ T-cell effector function. (A) Primary PDGFRA-GFP⁺ fibroblast-like stromal cells were freshly sorted from leukemic Pdgfra^CreERT2;Rosa26^mT/mG mice through flow cytometry and treated in vitro with IL-6 or PBS, followed by KEGG pathway enrichment analysis. (B) qPCR analysis of Fgf18 mRNA levels in fibroblasts treated with IL-6 (50 ng/mL, 6 hours), IL-6 Ab (1 μg/mL, 6 hours), or the JAK inhibitor itacitinib (JAKi, 1μM, 6 hours) (n = 3). qPCR data were normalized to Gapdh and are presented relative to the untreated control, which was set to 1. (C) ELISA measurement of FGF18 production in fibroblast-like stromal cells treated with IL-6 (50 ng/mL, 6 hours), IL-6 Ab (1 μg/mL, 6 hours), or itacitinib (JAKi, 1μM, 6 hours) (n = 3). (D) Western blot analysis of JAK1, p-JAK1, STAT3, p-STAT3, and FGF18 in fibroblasts (F) and fibroblasts cocultured with C1498 cells, with or without JAK inhibitor treatment (JAKi, +/−) for 6 hours. Each panel represents an independent biological replicate using fibroblasts isolated from a different mouse (3 mice total; separate blots). GAPDH was used as the loading control. (E) Western blot analysis of JAK1, p-JAK1, STAT3, p-STAT3, and FGF18 in PDGFRA-GFP⁺ fibroblast-like stromal cells cultured in CM (supernatant) from MLL-AF9 or C1498 AML cells or cocultured with MLL-AF9 or C1498 AML cells with or without IL-6 Ab for 6 hours, as indicated. (F) Western blot analysis of JAK1, p-JAK1, STAT3, p-STAT3, and FGF18 in fibroblast-like stromal cells treated with IL-6 ± IL-6 Ab for 6 hours. (G) Heat map showing the expression of activation-associated genes (GZMA and SLAMF6) and inhibitory receptor genes (PDCD1, CTLA4, LAG3, and HAVCR2) in response to IL-6 treatment in CD8⁺ T cells, based on the GSE106291 data set. (H) Flow cytometry analysis of cytokine/effector production (GZMB and IFN-γ) in CD8⁺ T cells treated with IL-6 (n = 3). (I) Heat map of activation-associated genes (GZMB and IFNG) and inhibitory receptor genes (PDCD1, CTLA4, and HAVCR2) in CD8⁺ T cells cultured for 6 hours with CM from primary AML cells (AML 1 and AML 2) treated with FGF18 or IL-6 or from F-A, with or without IL-6 Ab. Data from 2 primary AML samples (AML 1 and AML 2) are shown. (J) Percentage of PD-1⁺ CD8⁺ T cells after coculture with treated C1498 or MLL-AF9 cells under the indicated conditions (n = 3). (K) Cytotoxicity of CD8⁺ T cells against C1498 and MLL-AF9 AML cells at different effector-to-target ratios under various treatment conditions (Dulbecco's modified Eagle medium, FGF18, F-A, or IL-6) with or without IL-6 Ab for 6 hours (n = 3). Data are presented as mean ± SEM. Statistical significance was determined using 1-way ANOVA with Tukey post hoc test (panels B-C,H,J-K). coc., cocultured; DMEM, Dulbecco's modified Eagle medium; F-A, fibroblast-AML cocultures; GZMB, granzyme B; IFN-γ, interferon gamma; NC, negative control; sup., supernatant.

To explore the immunologic implications of IL-6 signaling, we analyzed GSE106291 and found that IL-6 expression was positively correlated with T-cell exhaustion markers (PDCD1, HAVCR2, CTLA4, and LAG3) and inversely correlated with effector genes (GZMA and SLAMF6) (Figure 6G). Moreover, IL-6 suppressed cytotoxic T-lymphocyte (CTL) activity, as shown by reduced interferon gamma and granzyme B production in CD8⁺ T cells and by increasing the proportion of PD-1⁺CD8⁺ T cells (Figure 6H; supplemental Figure 11A-B). CM from FGF18-treated AML cells, fibroblast-AML cocultures, and IL-6–treated AML cells increased expression of the exhaustion markers PDCD1, CTLA4, and HAVCR2 in T cells. This effect was reversed by IL-6 Ab, confirming that IL-6 is a key mediator of T-cell dysfunction (Figure 6I-J; supplemental Figure 11C-D). In cytotoxicity assays, FGF18, fibroblast-AML cocultures, and IL-6 reduced T-cell–mediated AML lysis, whereas IL-6 neutralization restored CTL function (Figure 6K). Thus, IL-6 limits antileukemic immunity in the niche, particularly through its cross talk with fibroblast-AML interactions.

Development and testing of an FGF18 Ab to improve survival in AML mice and overcome resistance to aPD-1 therapy

To evaluate its therapeutic potential, we assessed the FGF18-neutralizing antibody (FGF18 Ab) in AML. Western blot analysis revealed that FGF18 Ab blocked FGF18-induced phosphorylation of PI3K, AKT, and mTOR without altering total protein levels and reduced expression of the downstream effectors c-MYC and ID1 (Figure 7A). Consistently, FGF18 Ab significantly inhibited AML cell growth and reduced colony formation in vitro, with minimal impact on healthy donor cells (Figure 7B-C). In vivo, MLL-AF9 leukemia-bearing mice treated intraperitoneally with FGF18 Ab (100 μg every other day) exhibited markedly lower white blood cell counts and spleen weights (Figure 7D-F), as well as a reduced proportion of GFP⁺ leukemia cells in the BM (Figure 7G). FGF18 Ab treatment substantially prolonged survival (median, 35 days) compared with the vehicle or immunoglobulin G (IgG) control treatment (Figure 7H) and decreased leukemic infiltration in the liver and spleen (Figure 7I). Administration of FGF18 Ab to healthy adult mice was well tolerated, with no evident tissue toxicity or changes in body weight, organ weight, or hematologic parameters (supplemental Figure 12A-C). Given the reported role of FGF18 in bone biology, we evaluated skeletal safety and observed no differences in bone morphology, osteoclast activity, or growth plate architecture by histology (supplemental Figure 12D). Furthermore, micro-computed tomography imaging confirmed no alterations in trabecular bone parameters among the vehicle, FGF18 Ab, and IgG Ab groups (supplemental Figure 12E-F). Together, these data indicate that FGF18 neutralization effectively suppresses AML progression.

Figure 7.

FGF18 Ab suppresses AML progression and improves survival, correlating with the efficacy of immune checkpoint inhibitor therapy in vivo. (A) Western blot analysis of Molm-13 cells treated with the indicated doses of rFGF18 (0, 10, and 50 ng/mL) in the presence or absence of FGF18 Ab for 24 hours. GAPDH was used as the loading control. (B) Growth curve of primary AML cells isolated from AML patients treated with FGF18 Ab or IgG control for over 7 days (n = 3). (C) Colony formation assay showing the number of colonies formed by CD34⁺ hematopoietic stem/progenitor cells from healthy donors and CD34⁺ AML blasts from patients with AML treated with FGF18 Ab or IgG control (n = 5). (D) Schematic representation of in vivo experimental design. MLL-AF9-GFP leukemia cells were transplanted into recipient mice, and treatment with FGF18 Ab (100 μg per mouse, i.p., every other day) or IgG control was initiated 1 week after transplantation. (E) WBC counts in the PB of vehicle-, FGF18 Ab-, and IgG Ab-treated mice (n = 6). (F) Spleen weights of vehicle-, FGF18 Ab-, and IgG Ab-treated mice (n = 6). (G) Flow cytometry analysis of GFP⁺ leukemia cells in the BM of vehicle-, FGF18 Ab-, and IgG Ab-treated mice (n = 6). (H) Kaplan-Meier survival curve of MLL-AF9 AML mice treated with vehicle, FGF18 Ab, or IgG control (n = 8). (I) Representative hematoxylin and eosin staining of liver and spleen tissues from vehicle-, FGF18 Ab-, and IgG Ab-treated mice (original magnification ×20). (J) Flow cytometry analysis of GFP⁺ AML cells in the BM and spleen under the indicated treatments (n = 4). (K) Percentage of CD8⁺ T cells (gated on CD3⁺ cells) in the BM of mice treated with vehicle, FGF18 Ab, aPD-1, or a combination of FGF18 Ab and aPD-1 (n = 4). (L) Kaplan-Meier survival curve of MLL-AF9 mice treated with vehicle, FGF18 Ab, aPD-1, or a combination of FGF18 Ab and aPD-1 (n = 8). Data are presented as mean ± SEM. Statistical significance was determined using 2-way ANOVA (panel B), 2-tailed Student t test (panel C), 1-way ANOVA with Tukey post hoc test (panels E-G,J-K), or the log-rank test for survival analysis (panels H,L). HD, healthy donor; i.p., intraperitoneally; O.D., optical densities.

aPD-1 therapy exhibits limited efficacy in AML; therefore, the combination of FGF18 Ab and aPD-1 was explored to improve antileukemic outcomes. AML-bearing mice were treated with aPD-1 alone, FGF18 Ab alone, or a combination of both. Flow cytometry analysis revealed a significant decline in GFP⁺ leukemic cells, with the combination therapy group showing the greatest decrease compared with either monotherapy (Figure 7J), accompanied by an increased proportion of CD8⁺ T cells (Figure 7K). Kaplan-Meier analysis showed that combination therapy significantly improved survival compared with vehicle or either monotherapy (Figure 7L). These findings support the strategy of cotargeting FGF18 and PD-1 to enhance antileukemic immunity and potentially overcome resistance to immune checkpoint blockade in AML.

Discussion

The BM microenvironment is a key driver of AML progression, with stromal cells shaping a leukemia-supportive niche.^{38, 39, 40, 41} Despite increasing evidence of stromal heterogeneity and dynamic cross talk with leukemic cells, the molecular mechanisms by which stromal cells contribute to AML remain incompletely understood.^{42, 43, 44, 45} In this study, we profiled molecular alterations within BM stromal cells from AML mouse models using scRNA-seq and found that fibroblast function was impaired in AML. By integrating scRNA-seq and Smart-seq2 data sets, we identified FGF18 as a fibroblast-derived factor that is markedly upregulated in the leukemic stroma.

FGF18 is a member of the FGF family and has been extensively studied in nonmalignant contexts, particularly in bone biology, cartilage repair, and tissue regeneration.^{46, 47, 48} However, its role in malignancies, especially in hematologic cancers like AML, remains poorly defined. Our results revealed that elevated FGF18 levels in AML mouse models and in the serum of patients with AML correlate with disease progression and relapse. Consistent with a niche-derived mechanism, FGF18 levels were elevated in leukemic BM interstitial fluid, indicating increased local ligand availability in the BM microenvironment. Fibroblast-specific Fgf18 deletion delayed AML progression, and recombinant FGF18 partially rescued the growth-promoting effects of an FGF18-deficient niche. Moreover, systemic administration of recombinant FGF18 accelerated AML progression, suggesting that exogenous FGF18 may also contribute to disease progression. Although in vivo validation has been demonstrated by our findings, future studies should evaluate FGF18 in more fully humanized AML models to assess therapeutic efficacy and immune interactions.

Using a screening approach to identify AML blast-enriched transmembrane proteins essential for disease progression, we uncovered multiple AML-relevant genes, including Cd274, Il6st, and Ldlr. We focused on FGFR3 because it is a potential cognate receptor for FGF18. Although FGFR3 is well characterized in developmental and epithelial systems, its role in hematological malignancies remains elusive.^{49, 50, 51, 52} Our findings position FGFR3 as a key conduit for stromal signals in AML and as a microenvironment-specific therapeutic target. Other receptors identified in our screen may also play roles in AML progression and warrant further investigation.

Mechanistically, FGF18 engages FGFR3 as its specific receptor on AML cells, activating the PI3K-AKT-mTOR-E2F1-MYC/ID1 signaling cascade to promote leukemic cell proliferation. FGF18 also amplifies inflammatory signaling within the niche. By inducing nuclear translocation and activation of E2F1, FGF18 enhances IL-6 transcription, a cytokine that has long been implicated in AML survival, angiogenesis, and immune evasion.⁵³ Neutralization of IL-6 suppressed FGF18-mediated leukemic proliferation and disrupted stromal-leukemic cross talk, highlighting a positive feedback loop in which IL-6 reactivates JAK-STAT3 signaling in fibroblasts to further increase FGF18 secretion. Notably, IL-6 signaling appears to be biphasically regulated, with moderate levels sustaining pathway activation, whereas excessively high concentrations elicit intrinsic negative feedback that attenuates JAK-STAT3 activity. Moreover, our scRNA-seq analysis showed that fibroblasts in AML display perturbations in ECM-receptor interaction and focal adhesion pathways. Given that aberrant ECM signaling can promote leukemic cell retention and restrict T-cell infiltration, these data suggest a leukemia-permissive, immune-excluded BM niche. The immunologic consequences of this loop are profound. Elevated IL-6 levels drive CD8⁺ T-cell dysfunction, characterized by decreased interferon gamma and granzyme B production and increased PD-1, CTLA-4, and HAVCR2 expression.^54,55 Blocking IL-6 restored cytotoxic T-cell function, demonstrating that stromal cytokine signaling directly impairs antileukemic immunity.^{56, 57, 58} The axis uncovered here provides a mechanistic explanation for the immune checkpoint inhibitor resistance frequently observed in AML and identifies FGF18 as a potential upstream target to restore immune competence.⁵⁹ Therapeutically, antibody-mediated FGF18 blockade disrupted stromal-leukemic feedback, significantly prolonged survival in AML mouse models, and simultaneously suppressed IL-6 production and enhanced CD8⁺ T-cell effector function. Notably, our safety evaluation was limited to a 3-week dosing period. Because FGF18 can activate FGFR1/2/4 in addition to FGFR3, off-target activity may affect the proliferation, differentiation, and metabolism of tissues such as tendons, ligaments, cartilage, and endothelium. Prolonged FGF18 neutralization may trigger compensatory pathways, potentially leading to unforeseen changes in cell behavior or tissue repair, warranting further long-term investigation.

Given that PDGFRA⁺ fibroblast-like stromal cells contribute to osteogenic and adipogenic programs in the BM, an additional implication of our model is that the FGF18–IL-6 circuit may not only act directly on AML blasts but may also secondarily remodel stromal lineage outputs and niche composition (eg, osteolineage/adipogenic balance), with downstream effects on cytokine availability, vascular/ECM features, and immune cell behavior. These long-term lineage and niche remodeling consequences were not directly investigated in the current study and represent an important direction for future research. In addition to FGF18, our data revealed that additional BM stromal-derived factors, including BDNF, LIF, and REG4, were markedly upregulated in the AML stroma. By integrating these findings with our receptor-focused screen for AML progression drivers, we propose the LIF-LIFR axis as another critical tumor microenvironment–AML signaling axis that warrants future investigation. Overall, our findings support the broader concept that targeting microenvironmental signaling hubs can provide durable therapeutic benefits.

Conflict-of-interest disclosure: The authors declare no competing financial interests.