The Role of the Bone Marrow Microenvironment in the Pathogenesis of Acute Myeloid Leukemia

1. Introduction

Darwin observed that “a grain in the balance will determine which individual shall live or die, which species will increase or become extinct” (On the Origin of Species, 1861, chapter XIV). Evolution proceeds through small, random variations, of which only a few prove advantageous: the surviving species is the one that adapts most favorably to its environment.

In the 1950s and 1960s, Till and McCulloch proposed a hierarchical, pyramidal organization of hematopoiesis, with a self-renewing, asymmetrically dividing hematopoietic stem cell (HSC) at its apex (1). Asymmetric mitosis allows the cell both to differentiate toward a specific function and to self-regenerate at the top of the pyramid, maintaining the stem pool in quiescence and ensuring tissue inexhaustibility. The same logic applies to neoplastic tissues: relapse after morphological remission implies persistence of a quiescent, chemoresistant cancer stem cell with indefinite self-renewal, which drives carcinogenesis (2).

Stemness is not a cell-surface identity but a function, defined by the capacity to regenerate the tissue of origin in serial xenotransplantation into immunodeficient hosts (3). A stem cell — normal or leukemic — is conventionally defined by its capacity to sustain tissue in at least three serial transplants; hence the terms leukemic stem cell (LSC) or leukemia-initiating cell (L-IC). The observation that LSCs share the CD34⁺/CD38⁻ immunophenotype of HSCs led to the hypothesis that acute myeloid leukemia (AML) arises from oncogenic transformation of a normal HSC (4).

Immunophenotypic studies of LSCs began to disclose a role for the bone marrow microenvironment (BMM) in AML pathogenesis. In NOD/SCID/IL2Rγc-deficient models, lacking B, T, and NK cells (5), even committed progenitors expressing CD38 or CD45RA can regenerate AML in serial transplants, fulfilling LSC criteria (6). In a sufficiently permissive microenvironment, stemness is therefore not restricted to primitive cells.

These data reframe malignancy: aggressiveness depends not only on genetic and epigenetic traits but on the tumor microenvironment that shapes proliferative advantage. A neoplasm is an ecosystem in which malignant cells interact with non-neoplastic components that modulate tumor fitness and response to therapy (7), honoring Darwin’s intuition at a microscopic scale.

2. Clues From Clinical Experience

Allogeneic hematopoietic stem-cell transplantation (HSCT) is a curative treatment for eligible patients with high-risk AML. Sustained remission depends on the graft-versus-leukemia (GVL) effect exerted by donor immunity against residual leukemia stem cells (LSCs) (8). Relapse — the principal obstacle to cure — still occurs in 30–40% of transplanted patients (9). Most relapses are of host origin; donor cell leukemia (DCL), in which leukemia arises de novo from engrafted donor hematopoiesis, is rare but mechanistically informative.

Registry data estimate DCL at 0.08% (European Society for Blood and Marrow Transplantation, EBMT) and 0.16% (Japan Society for Hematopoietic Cell Transplantation, JSHCT) across indications (10,11). Among AML recipients, the reported incidence approaches 2%, with individual series describing rates up to 4–6%, and with a clear preponderance in adults (12). DCL develops after a median of 28 months post-HSCT, although the range is wide (1 month to 24 years). The first described case (1971) involved a 16-year-old female with acute lymphoblastic leukemia (ALL) who, after transplantation from her brother, developed donor-derived ALL in XY hematopoietic cells (12).

Diagnostic sensitivity has improved with chimerism analysis by short-tandem-repeat (STR) and variable-number-tandem-repeat (VNTR) profiling. The pathogenesis is multifactorial, reflecting donor, recipient, and procedure-related contributions in a multiple-hit model.

Prior to 2010, cytogenetic analysis of DCL identified chromosome 7 abnormalities in 23% of cases versus ≤5% of de novo AML (13) — a distribution reminiscent of therapy-related myeloid neoplasms and consistent with persistent mutagenic effects of recipient-directed chemotherapy and conditioning on the marrow stroma. Next-generation sequencing has refined this picture. Among the transcription factors recurrently mutated in DCL — CEBPA, GATA2, RUNX1, and JAK2 — three are typically of germline origin and one is predominantly somatic (13). Donor clonal hematopoiesis, particularly DNMT3A-mutated clonal hematopoiesis of indeterminate potential (CHIP), has been documented in recipients who later developed DCL (14,15); in related donors aged ≥50 years, CHIP prevalence reaches 16% (16). Additional recurrently mutated genes cluster into splicing regulators (DDX41), epigenetic modifiers (ASXL1, DNMT3A, EZH2, IDH1/2), and DNA-repair genes (CHEK1, XPD, XRCC3) (13).

That donors do not typically develop AML, except in extremely rare cases, suggests that the recipient microenvironment plays a significant role in supporting leukemogenesis. Preclinical data link alterations in bone marrow stromal cells and other niche components to the development and maintenance of leukemia and to chemotherapy resistance (17–21). Aldoss et al. reported two sequential myeloid DCLs in a single recipient after two HSCTs from different donors — a matched-related transplant followed three years later by a first DCL, then, after a new induction, a second transplant from a young unrelated donor with subsequent development of a second DCL from the second donor. The authors concluded that the recipient likely had a leukemogenic microenvironment that predisposed donor cells to transform (22,23).

DCL after cord-blood transplant (CBT) occurs earlier (median 14.5 vs 36 months) and is more frequently associated with chromosome 7 abnormalities. Fusion genes such as TEL-AML1 or AML1-ETO may be present in cord cells and predispose to expansion of a pre-leukemic clone; these are class 2 genetic alterations that impair differentiation but do not alone induce overt AML, and they often occur years before disease onset. Because cord blood cells are less likely to have acquired predisposing genetic mutations, DCLs after CBT provide a compelling argument for a role of the microenvironment in initiating leukemic transformation.

3. Possible Direct Oncogenic Role of the Bone Marrow Microenvironment

Mesenchymal stromal cells (MSCs) sustain marrow homeostasis by regulating HSC self-renewal, differentiation, and immune tolerance. Primary MSCs from patients with myelodysplastic syndromes (MDS) and AML differ structurally, genetically, and functionally from healthy controls; dysregulation of KIT-ligand, angiopoietin-1, and Jagged1 signaling is associated with impaired osteogenic commitment and reduced growth capacity (24–27). TGF-β signaling has been implicated as an important contributor to the aberrant MSC phenotype in both MDS and AML (28). While the role of the BMM in disease maintenance and chemoresistance is increasingly recognized, its contribution to disease initiation is only now emerging.

Raaijmakers et al. deleted Dicer1 — a ribonuclease III endonuclease indispensable for microRNA biogenesis — in murine MSCs at the osteolineage differentiation stage (19). Osteolineage-specific loss disrupted osteoblastic differentiation without affecting the osteoclastic component, induced severe cytopenias with marked myelodysplastic features, and accelerated progression to acute leukemia. Transplantation of hematopoietic cells from Dicer1-mutant mice into wild-type recipients restored normal hematopoiesis, and the emergent leukemic clone retained wild-type Dicer1 — demonstrating that the initiating event lay within the altered marrow microenvironment, not the hematopoietic compartment. Dicer1 inactivation in osteoprogenitors was also linked to downregulation of the Shwachman-Bodian-Diamond (Sbds) gene, which is involved in ribosomal biogenesis and whose germline loss produces Shwachman-Bodian-Diamond syndrome — characterized by skeletal alterations, inefficient hematopoiesis with bone marrow failure, and increased propensity to progress to MDS and AML. Reduced expression of Dicer1, Drosha, and Sbds has since been confirmed in MSCs from MDS and AML patients, together with loss of miR-155, miR-181a, and miR-222 — microRNAs that regulate hematopoiesis (29–31).

Activating β-catenin mutations in murine osteoblasts reproduce an analogous niche-initiated leukemia (32). β-catenin activation drives osteoblastic Jagged1 expression cooperatively with FoxO1, engaging Notch signaling in adjacent hematopoietic progenitors and producing AML with recurrent karyotypic alterations and autonomous progression (33). Notch activation in this setting is associated with chemoresistance.

Epigenetic dysregulation also contributes to the altered functional state of the BMM. In primary stromal cells from MDS patients, aberrant cytosine hypermethylation silences FRZB, a WNT antagonist, thereby activating the WNT/β-catenin pathway in surrounding HSCs (34). WNT activity affects HSC homeostasis in a dose-dependent manner, and increased WNT/β-catenin signaling has been linked to leukemic transformation (35,36). Whether this stromal methylation defect precedes emergence of the myelodysplastic clone remains unresolved.

Cytotoxic therapy imprints a durable pro-inflammatory state on the MSC compartment. Alkylating agents induce cellular senescence and a senescence-associated secretory phenotype (SASP) with elevated pro-inflammatory cytokines and reactive oxygen species (37). Multi-omic profiling of MSCs from patients with therapy-related myeloid neoplasms (t-MN) confirms induction of CDKN1A and β-galactosidase, FOS repression, elevated cytokine output, and impaired proliferation — a senescent-secretory signature that is stable regardless of the latency between cytotoxic exposure and t-MN diagnosis (38). Analogous changes are seen in MSCs from patients with Fanconi anemia (FA), an inherited disease characterized by congenital abnormalities, bone marrow failure, and increased susceptibility to myeloid malignancies: baseline downregulation of FLT3-ligand, TGF-β1, and CXCL12, with cytoplasmic mislocalization of Dicer1 even before etoposide exposure. In healthy-donor MSCs, etoposide redistributes Dicer1 to the cytoplasm and induces ER-stress genes (GADD34, ATF4, NUPR1, HSPB), phenocopying the FA baseline (39). Taken together, these data indicate that microenvironmental alterations may contribute to the pathogenesis of myeloid neoplasms, including AML, and suggest that they may also serve as therapeutic targets.

4. the Medullary Vascular Niche

Evidence indicates that self-renewing HSCs reside in the bone-marrow perivascular niche rather than exclusively in long-term dormant reserves (Figure 1). This specialized, protective compartment is composed of MSCs (40,41) and endothelial cells of sinusoidal microvessels (42,43). From this location, HSCs monitor blood-borne factors and activate the recruitment of additional HSCs from quiescent endosteal niches to stimulate hematopoiesis and osteogenesis, counteracting hematological stress (44). During embryogenesis, primitive hematopoiesis arises from hemogenic endothelium of the aorta-gonad-mesonephros region and yolk sac via endothelial-to-hematopoietic transition (45–47); the close spatial relationship between endothelial and hematopoietic progenitors persists into adulthood, with the vascular niche traditionally considered comparatively oxygen-rich relative to the endosteum (48) — a view that direct intravital oxygen measurements have since challenged and partially inverted (see §5).

Sinusoids are fenestrated, low-pressure channels (49,50) that maintain niche homeostasis through secretion of IL-1, IL-6, IL-8, GM-CSF, TNF, G-CSF, chemokines, and matrix metalloproteinases (51). The endosteal niche contains transition-zone vessels (52), roughly 15% of HSCs (53), and approximately 10% of marrow volume. The central vascular niche occupies the remaining ~90% of marrow volume and is estimated to harbor ~85% of HSCs (43). The endosteal compartment is relatively drug-resistant and supports post-chemotherapy regeneration (54), whereas the vascular compartment is sensitive to irradiation and myeloablation (55). Two-thirds of perivascular HSCs associate with sinusoids and LepR⁺ stromal cells; the remaining third associate with arterioles and NG2⁺/Nestin^high pericytes (56–58).

Specific alterations in the BM vascular niche can act synergistically with genetic or epigenetic lesions in hematopoietic cells to facilitate mutant-cell survival and expansion, contribute to progression and protection from chemotherapy, and promote relapse. Using intravital two-photon microscopy, Passaro et al. showed that vascular leakiness and increased hypoxia in AML are dependent on endothelial nitric oxide (NO); induction chemotherapy that fails to restore normal vasculature may be associated with poor prognosis, while inhibition of NO production promotes normal hematopoiesis and enhances treatment response in preclinical models (59). CXCL12-abundant reticular (CAR) cells are heterogeneously distributed in the perivascular endothelium: arteriolar CARs (NG2⁺/LepR⁻) enforce HSC quiescence, whereas sinusoidal CARs (NG2⁻/LepR⁺) support retention and expansion (57,60,61). The consistent association of HSCs with CXCL12-secreting reticular cells in both vascular and endosteal locations blurs the historical distinction between the two niches (59,62).

Increased microvascular density has been documented in ALL (63), chronic myeloid leukemia (CML) (64–66), AML (67), and multiple myeloma (MM) (68) and carries prognostic weight (69). Tumor neovasculature is abnormally permeable and tortuous (70,71). Endothelial cells (ECs) and leukemic cells engage in bidirectional signaling: ECs provide trophic and paracrine support, while blasts release proangiogenic factors (FGF-2, VEGF-A, VEGF-C, angiopoietins) and cytokines such as GM-CSF (72,73). AML cells can become quiescent and resist chemotherapy by integrating into the vascular endothelium, forming vascular tissue-associated AML (V-AML) (74–76). Combining anti-angiogenic therapy with consolidation chemotherapy has been proposed as a practical approach for patients with acute leukemia (77).

Adrenomedullin (ADM) is a 52-amino-acid peptide mainly synthesized by endothelial cells, vascular smooth muscle cells, and pericytes (78–80). It is expressed in many human tissues and exerts predominantly hypotensive, immunomodulatory, and vasoactive effects, including those observed during septic shock (81–83). ADM signals through calcitonin-receptor-like receptor (CALCRL) in complex with receptor activity-modifying proteins (RAMPs); among these, the CALCRL–RAMP2 heterodimer mediates canonical ADM signaling. ADM is also expressed by leukemia stem cells and defines an inflammatory signature in AML (84). Moreover, ADM is overexpressed in solid tumors under hypoxic conditions via HIF-1 regulation (85), exhibiting pro-angiogenic activity and stimulating tumor progression and chemoresistance (86). Our group showed that AML cell lines constitutively express ADM and its receptors to sustain proliferation and block differentiation, suppressing Cullin-family gene expression and activating PI3K/AKT and MAPK/ERK (87). Larrue et al. subsequently demonstrated that the ADM–CALCRL axis preserves relapse-initiating cells (RICs) that persist after chemotherapy; high CALCRL or ADM expression in RICs correlates with adverse outcome, and CALCRL knockdown reduces LSC frequency and restores cytarabine sensitivity in patient-derived xenografts (88).

Events that promote blast–endothelium adhesion are of particular interest given the capacity of AML cells to integrate into the vascular wall. ADM downregulates CD38 on blasts — thereby weakening the CD38/hyaluronic-acid axis — and simultaneously upregulates CD31, enhancing trans-endothelial migration and circulatory dissemination (87,89). In murine models, Cogle et al. documented progressive localization of AML blasts around hepatic portal vessels and their incorporation into endothelial syncytia (V-AML) that acquire quiescence and chemoresistance (76). Analogous observations in chronic myeloproliferative neoplasms, where marrow endothelial cells carry disease-defining mutations, suggest similar endothelial participation in other hematologic malignancies.

5. the Osteoblastic Niche

In AML, LSCs preferentially home to the osteoblastic compartment — the endosteal surface lined by osteoblasts — rather than to the vascular niche, as confirmed by homing and serial engraftment experiments (21,90) (Figure 1). Elevated endosteal ATP supports LSC retention through P2X7 signaling (91). LSCs in this compartment exhibit high rates of symmetric division, strong colony-forming potential, elevated glycolytic flux, and superior leukemogenic capacity in vivo (90). Expression of active Rac1 in murine AML reduces disease latency by enhancing niche homing, quiescence, and apoptosis protection (92).

Oxygen tension in the leukemic endosteum falls to ~0.1% (90,93), substantially below that of the vascular niche; in healthy mice the relationship is inverted, with deeper peri-sinusoidal regions of the vascular niche slightly more hypoxic (~1.3%) than the endosteum (94). Hypoxia can promote HSC quiescence (95) and, via hypoxia-associated factor (HAF), HIF-1α, and HIF-2α, upregulates OCT-3/4, NANOG, and SOX2 — transcription factors whose ectopic expression can reprogram somatic cells into induced pluripotency (96–99). In cancer stem cells, MYC, NANOG, and SOX2 stabilize HIF-2α, which represses p53 and preserves self-renewal (100); loss of p53 in turn impairs MDM2-mediated HIF-1α degradation (101). Although a clear consensus has not been reached, increased expression of OCT4 and NANOG has been reported in AML relative to normal marrow (102).

In endosteal LSCs, HIF-1 drives metabolic reprogramming by upregulating PDK2 and blocking pyruvate entry into the tricarboxylic acid (TCA) cycle (103). Concurrent AMPK activation induces GLUT1 and the pentose-phosphate pathway, raising reduced glutathione and buffering against reactive oxygen species (ROS)-mediated oxidative stress (90,104). In T-cell acute lymphoblastic leukemia (T-ALL), NOTCH mutations engage AMPK with comparable cytoprotective effect (105), supporting AMPK as a therapeutic target for sensitization. Conversely, oxidative phosphorylation (OXPHOS) is required to maintain quiescent human LSCs (106), leaving the quantitative relationship between the endosteum, hypoxia, glucose availability, and LSC fate open.

Blasts engage osteoblasts through fibronectin–hyaluronic-acid/CD44, ICAM1/LFA, WNT/Frizzled, Notch/Jagged, CXCL12/CXCR4, VCAM1/VLA4, and N-cadherin/N-cadherin pairings, sustaining a stemness program in vitro (107–109). Osteoblasts shield AML cells from chemotherapy- and chemokine-induced apoptosis. CXCR4-high AML cells escape CXCL12-mediated apoptosis in co-culture, independently of direct contact or MSC involvement (110); this protection is lost after histone-deacetylase inhibition through downregulation of tissue-nonspecific alkaline phosphatase (111,112). Osteoblasts protect blasts from cytarabine in both two- and three-dimensional systems (113), and AML-derived osteoblasts shield cells from daunorubicin more effectively than healthy-donor osteoblasts (114).

The interaction is reciprocal. Osteoblast numbers are reduced in AML patients and murine models (115–117). Co-culture with AML blasts arrests stromal differentiation at a pre-osteoblastic stage that overexpresses pro-leukemogenic cytokines (IL-6, CCL2, CXCL8) (118). Elevated marrow CCL3 suppresses mature osteoblasts, as previously shown in MM (117,119). Genetic depletion of osteoblasts accelerates leukemic engraftment, while pharmacologic preservation of osteoblasts restores marrow function, reduces tumor burden, and prolongs survival (116).

Mechanistically, blast-derived kynurenine binds serotonin receptor 1B (HTR1B) on osteoblasts, inducing serum amyloid A, which upregulates IDO1 — the rate-limiting enzyme in kynurenine synthesis — and closes a self-sustaining kynurenine–HTR1B–SAA–IDO1 loop that generates a leukemia-permissive niche (120). Secondary transplantation of long-term HSC-phenotype cells from osteoblast-ablated donors paradoxically prolongs survival, because separation from the endosteum disengages LSCs from quiescence, drives proliferation and differentiation, and depletes the LSC pool (121).

Taken together, this evidence indicates that cells expressing the characteristic LSC functional program reside in the endosteal niche through direct and indirect interactions with microenvironmental cells that recruit and anchor them there. Osteoblasts, in turn, are instructed by AML blasts either to preserve LSC properties or to actively generate functional LSCs — cells held in a state of metabolic and proliferative quiescence, resistant to oxidative stress, and shielded from chemotherapy. The osteoblastic niche therefore remains a main suspect for the persistence of measurable residual disease (MRD) during therapy.

6. Chemoresistance Mechanisms Induced by the Bone Marrow Microenvironment

Chemoresistance in AML depends on a set of non-cell-autonomous mechanisms that the preceding sections have begun to outline: MSC- and osteoblast-mediated protection, adhesion- and chemokine-driven survival signaling, immune cell-mediated drug tolerance, and metabolic reprogramming sustained by the niche (Figure 2). This section integrates those mechanisms.

Contact with MSCs from healthy donors or AML patients confers a side-population (SP) phenotype on AML blasts through α4-integrin engagement and stroma-dependent activation of ABC transporters; SP cells are quiescent and chemoresistant in vitro and in patient-derived xenografts (122). Co-culture with AML-derived MSCs activates Notch signaling more strongly than MSCs from healthy donors, amplifying proliferation and drug tolerance (123). Furthermore, co-culture of AML blasts with MSCs engages a self-sustaining loop in which AML cells, via the apoptosis repressor with caspase recruitment domain (ARC) protein, activate NF-κB and drive IL-1β secretion. IL-1β induces COX-2 expression in MSCs, which release prostaglandin E2 (PGE2); PGE2 feeds back onto AML cells and activates a β-catenin/ARC axis, increasing their oncogenic and chemoresistance potential (124). ARC had previously been reported as an independent adverse prognostic factor in AML (125). These findings are consistent with the notion that a pro-inflammatory microenvironment may be associated with greater aggressiveness, chemoresistance, and unfavorable clinical outcomes.

In the HS-5 co-culture model, stromal contact shifts AML metabolism toward OXPHOS and ATP synthesis, inhibiting AMPK and activating mTORC1 (126). Genetic AMPK knockout accelerates leukemia growth and chemoresistance, supporting the OXPHOS–AMPK axis as a potential therapeutic target.

The CXCL12/CXCR4 axis is a major mediator of niche-mediated resistance to AML therapies, including FLT3 and IDH inhibitors (127). CXCR4 is expressed on HSCs and is overexpressed on AML cells, where it correlates with poor prognosis; chemotherapy and the hypoxic niche further upregulate it (128,129). High CXCR4 allows leukemic clones to outcompete normal HSCs for niche occupancy and drive their exhaustion (130). Blast–stroma adhesion provides a protective environment from chemotherapy-induced cell death by promoting leukemic cell growth and activating anti-apoptotic signals (131). In particular, the CXCL12/CXCR4 axis promotes leukemic cell survival by activating AKT and MAPK pathways and by overexpressing the anti-apoptotic protein BCL-xL — a consequence of CXCR4-driven downregulation of miR-let-7a, which otherwise targets BCL-xL for degradation (132). Stromal GM-CSF, G-CSF, and other cytokines engage JAK/STAT to switch anti-apoptotic dependence from BCL-2 to BCL-xL, producing refractoriness to venetoclax that is reversed ex vivo and in murine models by ruxolitinib (133). CXCL12 also recruits macrophages and polarizes them toward an immunosuppressive M2 phenotype (134–136). Celik et al. reported that AML marrow is enriched for MSCs secreting CXCL8, which supports leukemic survival via PI3K/AKT (137).

Bone-marrow macrophages are major effectors of chemoresistance. CD169⁺ macrophages are required for cytarabine and doxorubicin resistance in murine AML (138); M-CSF-induced CD163⁺CD206⁺ M2 macrophages protect AML cell lines from daunorubicin (139). M2-like tumor-associated macrophages (TAMs) regulate their own polarization (140) and support leukemic cell survival by rewiring their metabolic profile to rely more on amino acids, fatty acids, and oxidative metabolism than healthy macrophages (141). TAMs also sustain the survival of drug-resistant LSCs, which drive disease relapse (142).

LSC drug tolerance depends on metabolic plasticity shaped by the niche (143). One important mechanism of relapse to venetoclax/azacitidine involves increased fatty-acid oxidation (FAO) in LSCs, which can be overcome experimentally by etomoxir (144,145). FAO provides NADH, FADH₂, and acetyl-CoA for mitochondrial respiration and the TCA cycle (146). Bone-marrow adipocytes release CXCL12 to retain LSCs in their proximity (131) and deliver free fatty acids in response to LSC-derived TNF-α, IL-1α, IL-1β, and CSF2 (147). LSCs upregulate the fatty-acid transporter CD36 and the lipid chaperone FABP4, and nuclear activation of PPARγ drives CD36, FABP4, and BCL2 transcription, sustaining drug resistance (148,149). Shafat et al. confirmed that AML blasts activate adipocyte lipolysis and that FABP4 knockdown suppresses leukemic proliferation in co-culture and prolongs survival in Hoxa9/Meis1-driven AML (150). As noted in §5, LSCs maintain low ROS levels; BCL-2 inhibition selectively eliminates quiescent LSCs by disrupting OXPHOS (106), and FAO inhibition triggers the integrated stress response and apoptosis (151–153).

Metabolic cooperation extends to organelle transfer. Following OXPHOS inhibition, AML cells form nanotubes through which functional mitochondria are transferred from stroma to leukemic cells, improving their metabolic activity and promoting drug resistance. The process is driven by ROS-mediated activation of the PI3K/AKT pathway, which opens a transmembrane gap-junction channel enabling mitochondrial trafficking; the resulting increase in mitochondrial mass and ATP production provides a survival advantage (154). AML chemoresistance therefore reflects, at least in part, metabolic cooperation between the leukemic clone and its BMM, rather than being explained solely by cell-intrinsic mechanisms.

7. Remodeling of the Niche Induced by Cross-Talk with Leukemic Cells

Research has mainly described how the BMM affects leukemic cells, with only limited focus on the reciprocal relationship. Increasing evidence indicates a “partnership” between the two components, in which leukemic cells also modify the marrow niches and thereby improve the overall “fitness” of the neoplasm (Figure 1). One of the most informative mechanisms of communication between leukemic cells and their surroundings is the production of exosomes, which are principal mediators of intercellular signaling.

Exosomes carry diverse molecular cargo — RNA (especially microRNA), proteins, lipids, and DNA — and have potential as disease biomarkers or as targets for novel therapies. AML blasts use exosomes to suppress normal hematopoietic progenitors both indirectly, by altering the supporting marrow niche into a microenvironment favorable to leukemic cells (including modification of CXCL12-dependent signaling, discussed in §4 and §6), and directly, by delivering inhibitory microRNA signals to hematopoietic progenitors (155). Exosomes isolated from leukemic blast cultures or from the plasma of AML mouse models are rich in miR-150 and miR-155, which impair the clonogenic ability of normal HSCs by inhibiting c-Myc transcription (156).

Exosomes derived from AML blasts also downregulate genes that support normal hematopoiesis in marrow stromal cells — including IGF1, CXCL12, KITL, and IL-7 — as well as osteogenic genes such as IGF1, OCN, and COL1A1. At the same time, they upregulate DKK1 in the same cells, inhibiting the Wnt/β-catenin pathway, which is essential to physiological hematopoiesis and osteogenesis. Forced DKK1 expression in osteoblasts leads to osteopenia and destruction of the HSC niche, resulting in hematopoietic defects (115). A further study demonstrated that exosomes from an AML cell line induce IL-8 production by bone marrow MSCs, which in turn promotes chemoresistance to etoposide by increasing expression of the 70-kDa heat-shock protein (HSP70) and lysosomal-associated membrane protein 3 (CD63) (157). Overall, these studies highlight the capacity of AML cells to create a pro-leukemic niche that hinders normal hematopoiesis and activates chemotherapy-resistance mechanisms.

8. Therapeutic Implications

These mechanisms provide a rationale for therapeutic targeting of the BMM (Figure 2). The CXCL12/CXCR4 axis is the most extensively studied target. Plerixafor, a CXCR4 partial agonist, has been evaluated across phase I–II trials in newly diagnosed and relapsed/refractory (R/R) AML with intensive, non-intensive, salvage, and conditioning regimens (158–166). In responders it at least doubles blast mobilization into peripheral blood (159,160). Trial NCT00906945 was terminated for futility at interim analysis, and historical comparisons have shown no consistent survival advantage (165). No randomized data are available.

Motixafortide (BL-8040), LY2510924, and the IgG4 antibody ulocuplumab have produced encouraging activity in R/R AML (167,168). Ulocuplumab — exploiting IgG pharmacokinetics and potential antibody-dependent cellular cytotoxicity — administered one week before mitoxantrone–etoposide–cytarabine (MEC) achieved fivefold blast mobilization, four complete remissions (CR) or CR with incomplete count recovery (CRi) before chemotherapy, and an overall response rate (ORR) of 51% (169). Nonetheless, randomized trials of motixafortide plus high-dose cytarabine (HiDAC) consolidation and of ulocuplumab plus low-dose cytarabine (LDAC) in unfit patients failed to improve response or overall survival (OS) (170,171). Timing and intensity of chemotherapy are therefore likely determinants of synergy.

Tipifarnib, a farnesyltransferase inhibitor extensively studied in AML (172–176), illustrates the same principle. Re-analysis of NCT00027872 identified high baseline CXCL12 expression as predictive of response, a finding replicated in R/R peripheral T-cell lymphoma (177–179); tipifarnib reduces CXCL12 transcription in pancreatic and bone-marrow stromal cells in vitro and ex vivo (178,180). Yet phase III trials in newly diagnosed elderly unfit patients and as post-induction maintenance showed no response or survival benefit (181–183). Early-phase combinations with chemotherapy produced prolonged CR duration in matched historical comparisons but have not been validated in randomized studies (184–187).

Dociparstat (CX-01), a 2-O,3-O-desulfated heparin without anticoagulant activity, antagonizes CXCL12, PF4, HMGB1, and P-selectin (188,189). Disruption of the CXCL12 signal is associated not with mobilization of AML blasts but with inhibition of chemotaxis (188,190). In combination with 7+3 induction it delivered CR/CRi rates approaching 90% versus 50% for chemotherapy alone, with longer event-free survival (EFS); the randomized phase III NCT04571645 was terminated for slow accrual (188,191,192). Its pleiotropic effects on PF4 and P-selectin may additionally mitigate chemotherapy-induced cytopenias and blast adhesion (189,193).

Uproleselan (GMI-1271), a synthetic glycomimetic E-selectin antagonist, targets cell adhesion to the endothelium and the pro-survival signaling it activates (194,195). Early-phase trials reported ORRs of 48%, 80%, and 39% with salvage, induction, and bridge-to-transplant regimens in cohorts enriched for adverse cytogenetics and secondary AML (196,197). The phase III trial combining uproleselan with MEC or FLAI did not meet its OS primary endpoint (hazard ratio, HR, 0.89; 95% CI 0.69–1.15) but suggested a significant benefit in refractory disease (31.2 vs 10.1 months) (198), consistent with higher E-selectin ligand expression on refractory blasts (196). CR/CRi rates with motixafortide have similarly been higher in refractory than relapsed disease (159,168), suggesting that chemorefractory patients may preferentially benefit from niche-targeted therapy.

β-catenin dysregulation can be disrupted by sequestering the coactivator CBP with the peptidomimetic CWP232291 (199), by inhibiting the MUC1-C/β-catenin interaction with GO-203 (200–202), or by GSK-3 inhibition. As single agents in R/R AML, all three failed to produce meaningful response despite acceptable safety; only GO-203 combined with decitabine achieved a 43% CR/CRi rate (203–205). The kynurenine–HTR1B–SAA–IDO1 axis has been addressed through IDO1 inhibition: epacadostat development was halted, while indoximod with standard induction delivered 84% CR/CRi and MRD negativity in half of a small, predominantly adverse-risk cohort (206–208). Bioinformatic analyses link IDO1 overexpression to altered B-cell subsets and elevated IL-10, both reversible in vitro by IDO1 inhibition (209).

Pro-inflammatory cytokines — IL-1β, IL-6, IL-8 — released by blasts, MSCs, and pre-osteoblasts sustain the altered niche and drive resistance to daunorubicin and cytarabine, with IL-6 as a recurrent mediator (210,211). In the phase I TOCILAM trial, tocilizumab added to standard induction produced an acceptable response, but one-year survival was below 50%, consistent with the adverse/intermediate European LeukemiaNet (ELN) 2017 risk profile (212).

The Hedgehog pathway regulates HSC quiescence and LSC maintenance (213–215). AML cells exploit both autocrine PTCH1 signaling and paracrine Hh stimulation from osteoblasts, endothelial cells, and stroma. Smoothened inhibitors have shown limited activity; glasdegib as a single agent is active only in R/R AML (216–218). In newly diagnosed patients, randomized trials demonstrated CR/CRi and OS benefit for glasdegib plus LDAC but only a CR/CRi benefit when combined with hypomethylating agents (219–223). The randomized phase III BRIGHT AML 1019 did not confirm an OS advantage (HR 1.05; 95% CI 0.782–1.408; P = 0.749) despite earlier evidence of cytarabine sensitization through UGT1A downregulation (221,224,225); placebo was favored in ELN 2017 intermediate-risk and GLI2-mutated subgroups.

Overall, with the exception of glasdegib plus LDAC, no niche-directed strategy has entered standard practice. Single-agent activity is limited, so combination design — especially with venetoclax-based regimens — and timing require rigorous definition in new trials. Subgroup selection on the basis of niche biology (secondary, therapy-related, refractory, and high-risk disease) is likely to be an important determinant of clinical signal.

9. Conclusions

The BMM has a major role in AML biology and pathogenesis, sustaining leukemic cell survival, proliferation, and drug resistance through a network of cellular and molecular interactions. The vascular and osteoblastic niches, together with adipocytes and tumor-associated macrophages, are remodeled by AML blasts into a permissive compartment that favors disease progression. Metabolic reprogramming — including enhanced fatty-acid oxidation and mitochondrial transfer from stromal cells — contributes to chemoresistance, while exosome-mediated communication allows AML cells to further reshape the niche, suppress normal hematopoiesis, and reinforce leukemic fitness.

The tumor microenvironment, although composed of non-neoplastic cells, must therefore be considered in any comprehensive model of AML biology. In selected contexts — stromal-initiated murine models, donor cell leukemia, and DCL after cord-blood transplantation — it may also contribute to leukemogenesis.

For therapy, the implication is not to replace clone-directed treatment but to combine it with strategies that disrupt protective niche interactions, quiescence, and metabolic plasticity. Preliminary clinical signals with CXCR4, E-selectin, IDO1, and Hedgehog-directed approaches are encouraging in some settings, particularly refractory, secondary, or therapy-related AML, but randomized confirmation, rational combinations, and better patient selection remain necessary.