Arsenic trioxide for reprogramming the bone marrow microenvironment to eliminate acute myeloid leukemia blasts

Acute myeloid leukemia (AML) is characterized by the uncontrolled proliferation of immature myeloid cells. Traditional chemotherapy has been the cornerstone of AML treatment for decades. ¹ The introduction of targeted therapies, such as FLT3 tyrosine kinase inhibitors (FLT3TKIs) and IDH inhibitors (IDHi), has improved outcomes for specific AML subtypes. Currently, up to 80% of patients achieve morphologic complete remission (CR) at the end of induction. ¹ However, most relapse and die, largely due to the persistence of minimal residual disease (MRD). Understanding the mechanisms behind MRD persistence is paramount for improving survival for these patients. Evidence increasingly shows that the bone marrow microenvironment (BME) plays a central role in the persistence of MRD. ² BME‐dependent drug resistance includes agent‐specific mechanisms (e.g., activation of alternative signaling pathways) ³ and broader resistance (e.g., impaired pharmacokinetics, PKs). ⁴ , ⁵ Currently, eliminating MRD and thus, achieving a cure requires multiple chemotherapy cycles or bone marrow (BM) transplantation. Targeting driver mutations alone is insufficient to eradicate MRD.

A notable exception is acute promyelocytic leukemia (APL), driven by the PML‐RARA fusion oncoprotein, the product of t(15;17). Targeting PML‐RARA with all‐trans retinoic acid (ATRA) and arsenic trioxide (ATO) virtually eliminates relapse, ⁶ even though each agent alone is suboptimal. Both ATRA and ATO bind to different parts of PML‐RARA. ATRA binds RARA, restores retinoid‐dependent transcription, and induces differentiation. ⁷ However, single‐agent ATRA results in “remission without cure,” with nearly 100% relapse rates. ⁸ This is due to BME‐expressed CYP26, an enzyme that degrades ATRA, impairing ATRA PKs in the BM. ⁹ , ¹⁰ , ¹¹ Retinoids directly upregulate stromal CYP26, creating a retinoid‐free niche that allows MRD to persist. ¹² Liposomal ATRA, which improves ATRA tissue distribution, has been more effective in reducing relapse, highlighting the role of ATRA PKs in MRD persistence in APL. ¹³ The biochemical barrier created by the BME can also be bypassed using CYP26‐resistant retinoids or inhibiting CYP26. ¹¹ , ¹² , ¹⁴ , ¹⁵

ATO also directly binds PML‐RARA, leading to ubiquitination and degradation of the fusion protein. ¹⁶ However, single‐agent ATO does not fully eliminate MRD, with relapse rates around 30%. ¹⁷ , ¹⁸ By binding distinct PML‐RARA regions, ATRA and ATO synergistically eliminate leukemic blasts to achieve CR. The mechanism of this synergy within the “retinoid‐free” BM niche is unclear. A possible explanation is that ATO enhances ATRA PKs, improving retinoid bioavailability in the BM. Clinical observations such as increased transaminitis and a higher incidence of pseudotumor cerebri in APL patients receiving ATRA + ATO compared to ATRA + chemotherapy suggest increased systemic ATRA levels. ⁶ , ¹⁹

To test if ATO alters ATRA PKs, we studied four consecutive patients diagnosed with standard risk APL at The Johns Hopkins Hospital (Table S1). The initial treatment included single‐agent ATRA until APL was confirmed, after which ATO was added. We compared systemic ATRA PKs before and after ATO initiation (Figure 1A) and found significantly higher ATRA plasma exposure during ATRA + ATO treatment even though the ATRA dose was not modified during these time point (Figures 1B and S1). Complete experimental methods can be found in Supplementary Information. Published data from the treatment of healthy volunteers suggest that ATRA levels typically decline over time due to hepatic CYP26 upregulation. ²⁰ However, in our study, ATRA levels increased during ATO administration. These findings prompted us to test if ATO suppresses hepatic CYP26 levels. Using human hepatocytes, we confirmed that ATRA induces hepatic CYP26B1 expression and concomitant treatment with ATO blunts this ATRA effect (Figure 1C). ATO had no effect on CYP26A1 expression. Thus, the effect of ATO on the ATRA‐induced hepatic CYP26B1 expression may explain the higher systemic ATRA PKs and the higher incidence of ATRA‐related toxicities seen in patients treated with combination therapy and underscores the need for heightened vigilance when using ATRA/ATO in the clinic—including close monitoring for retinoid‐associated adverse events.

Figure 1.

The impact of arsenic trioxide (ATO) on all‐trans retinoic acid (ATRA) pharmakinetics, CYP26 expression, and stroma‐mediated resistance. (A) Experimental design aimed to assess the impact of ATO on systemic ATRA levels in patients diagnosed with acute promyelocytic leukemia (APL); blue horizontal lines depict doses of ATRA, and orange horizontal lines depict doses of ATO. Plasma was collected on Day 1 and Day 6 of treatment, before the second dose of ATRA (0 h) and at indicated time points (1, 2, 4, 6, or 8 h) after the second dose of ATRA. (B) Area under the curve (AUC) was determined based on plasma levels of ATRA from patients treated with single‐agent ATRA (blue bar) and patients treated with ATRA + ATO (red bar). Data represent average ± STD, N = 4 patients. All groups were compared using Student's t‐test. Those with P < 0.01 were marked with **. (C) Changes in CYP26B1 messenger RNA (mRNA) expression in hepatocytes treated daily for 5 days with 10⁻⁶ M ATRA ± 0.5 × 10⁻⁶ M ATO in vitro. Data represent the average ± STD of three independent experiments. All groups were compared using Student's t‐test. Those with P < 0.01 were marked with **. (D) Changes in CYP26 mRNA expression in primary human bone marrow‐derived mesenchymal stromal cells (hBMSCs) treated daily for 5 days with 10⁻⁶ M ATRA ± 0.5 × 10⁻⁶ M ATO in vitro. Data represent the average ± STD of three independent experiments. All groups were compared using Student's t‐test. Those with P < 0.05 were marked with *. (E) AUC of ATRA on Day 5 of coculture with hBMSC after daily treatment with 10^‐6 M ATRA ± 0.5 × 10⁻⁶ M ATO. Data represent average ± STD of three independent experiments. All groups were compared using Student's t‐test. Those with P < 0.01 were marked with **. (F) Inhibitory concentration (IC)₅₀ of ATRA defined by CFU recovery of NB4R cells treated for 72 h either in suspension (left two columns) or in the presence of hBMSCs (right two columns) with or without 0.5 × 10⁻⁶ M ATO. Data represent the average ± STD of three independent experiments. All groups were compared using Student's t‐test. Those with P < 0.01 were marked with **.

While hepatic CYP26 regulates systemic ATRA PKs, BM stromal CYP26 controls local retinoid levels. Previous data show that ATRA upregulates stromal CYP26B1, resulting in paradoxically lower retinoid levels in the BME. ¹² Using primary human BM‐derived mesenchymal stroma cells (hBMSCs) (Table S2), we observed that, similar to data from hepatocytes, ATO blunts the ATRA‐induced upregulation of stromal CYP26B1 expression (Figure 1D). Consequently, addition of ATO enhances ATRA PKs in the presence of hBMSCs in vitro (Figures 1E and S2), suggesting that ATO improves not only systemic ATRA PKs but also BM retinoid levels by decreasing CYP26B1‐mediated ATRA metabolism.

We tested whether ATO's effects on stromal CYP26B1 and ATRA PKs could overcome stroma‐mediated resistance (Figure S3). We used NB4R cells, an APL cell line resistant to ATO via acquisition of the A216V mutation in PML‐RARA. ²¹ In stroma‐free cultures, NB4R cells remained sensitive to ATRA‐induced differentiation (Figures 1F and S4A,B). In these conditions, ATO had no impact on the effects of ATRA. In contrast, hBMSCs protected NB4R cells from ATRA‐induced differentiation, a protection that was reversed by the addition of ATO (Figures 1F and S4A,B). Thus, even though NB4R cells are intrinsically resistant to ATO, in stroma coculture conditions ATO sensitizes these cells to ATRA‐induced differentiation.

To confirm that CYP26B1 downregulation is necessary for ATO's effects, we created MSCs lacking CYP26 (CYP26KO) or MSCs that overexpress only CYP26B1 (CYP26TG) (Supplementary Methods). CYP26KO MSCs failed to protect NB4R cells from ATRA‐induced differentiation, and ATO provided no additional benefit (Figure S4C,D). Conversely, CYP26TG MSCs protected NB4R cells, and ATO did not reverse this effect (Figure S4C,D). Thus, ATO's ability to downregulate stromal CYP26B1 is essential for its effects on ATRA‐induced differentiation in the presence of BM stroma.

Similar mechanisms may be at play in non‐APL AML, where BM MSCs protect leukemia cells from targeted therapies. ³ Notably, retinoids enhance sensitivity to FLT3 TKIs and IDHi, ²² , ²³ raising the question of whether stromal CYP26 impairs FLT3 TKI and IDHi efficacy in the BM and thus, contributes to persistence of MRD during treatment with these agents. Because prior studies showed that pharmacological ATRA synergizes with sorafenib (a FLT3 TKI) to eliminate FLT3‐mutant AML cells, ²³ we tested if hBMSCs disrupt this synergy. We found that, as reported, sorafenib and ATRA are synergistic in eliminating FLT3‐mutant AML cells in stroma‐free conditions (combination index [CI] = 0.5), but the presence of hBMSCs renders ATRA inefficient and breaks this synergy (CI = 1.1) (Figure 2A). Inhibition of stromal CYP26 restores sensitivity to and the synergy between sorafenib and ATRA (CI = 0.75) (Figure 2A). We sought to investigate whether ATO‐mediated downregulation of stromal CYP26 could enhance FLT3 TKI efficacy and thus, may decrease MRD burden in FLT3‐mutant AML. Since gilteritinib is the only FLT3 TKI shown to reduce MRD burden, ²⁴ we used it for further studies. To eliminate the direct effects of ATO on AML cells, we generated ATO‐resistant MV4‐11 (MV4‐11R) cells by gradually exposing them to increasing concentrations of ATO over 3 months (Supplementary Methods). ATO did not alter the gilteritinib inhibitory concentration (IC)₅₀ of MV4‐11R cells in stroma‐free cultures (Figures 2B and S5). In stroma cocultures, MV4‐11R cells gain relative resistance to gilteritinib and ATO overcame this acquired, cell extrinsic resistance and synergizes with gilteritinib (CI = 0.5) (Figures 2B and S5). In CYP26KO or CYP26TG MSC cocultures, ATO had no effect on gilteritinib sensitivity of MV4‐11R cells, confirming that CYP26B1 downregulation is required for ATO's synergy with FLT3 TKIs (Figure S6) in the presence of hBMSCs.

Figure 2.

The impact of stromal CYP26 and arsenic trioxide (ATO) on FLT3 activity in vitro and in vivo. (A) Absolute CFU of MV4‐11 cells treated for 72 h with 10⁻⁸ M sorafenib and/or 10⁻⁷ M all‐trans retinoic acid (ATRA) in stroma‐free conditions (suspension: left panel), in the presence of human bone marrow‐derived mesenchymal stromal cells (hBMSCs; middle panel) or in the presence of hBMSCs and 10⁻⁶ M R115866, a potent CYP26 inhibitor (hBMSC + CYP26 inhibitor: right panel). Data represent the average ± STD of three independent experiments. All groups were compared using Student's t‐test. Those with P < 0.01 were marked with **. (B) Inhibitory concentration (IC)₅₀ of gilteritinib defined by CFU recovery of MV4‐11R cells treated for 72 h with gilteritinib ± 0.5 × 10⁻⁶ M ATO in the presence or absence of hBMSCs. Data represent the average ± STD of three independent experiments. All groups were compared using Student's t‐test. Those with P < 0.05 were marked with *. (C) Kaplan–Maier curve of survival of MV4‐11R‐Luc acute myeloid leukemia (AML) mouse xenografts treated with either a carrier (no treatment), ATO alone, gilteritinib alone, or gilteritinib + ATO for 28 days. Mice treated with gilteritinib or gilteritinib + ATO were compared using the log‐rank (Mantel–Cox) test. P value is displayed on the graph. (D) IC₅₀ of gilteritinib defined by recovery of alive primary AML blasts treated for 72 h with gilteritinib ± ATO in the presence of hBMSCs. Lines connect individual patients. Paired t‐test was used to compare the two conditions. Those with P < 0.01 were marked with **.

To assess if ATO deepens FLT3 TKI‐induced remission in vivo, we developed a mouse AML xenograft model using luciferase‐expressing MV4‐11R cells (MV4‐11R‐Luc) (Supplementary Methods). Mice with equal tumor burden as determined by bioluminescence were assigned to four groups: no treatment, ATO alone, gilteritinib alone, and gilteritinib + ATO. By Day 28, no treatment and ATO‐alone groups had significant tumor burden, and they were moribund. At this time, the groups treated with gilteritinib and gilteritinib + ATO showed no clinical signs of disease, and their bioluminescence levels were undetectable, thus achieving a state comparable to clinical CR. The treatment was stopped, and treatment‐free survival was measured as a surrogate of MRD burden. Despite the small group size, mice receiving gilteritinib + ATO had significantly prolonged treatment‐free survival, indicating lower MRD burden at treatment discontinuation (Figure 2C).

To extend our findings to patient‐derived models, we assessed FLT3‐mutant AML blasts from newly diagnosed or relapsed patients (Table S3, Figure S7). Primary blasts exhibited variable gilteritinib IC₅₀ values, with relapsed‐disease samples tending toward greater sensitivity. Importantly, ATO cotreatment in hBMSCs coculture consistently reduced gilteritinib IC₅₀ across all primary samples (Figure 2D).

Taken together, our findings suggest ATO downregulates CYP26B1 and increases retinoid tissue distribution and thus, synergizes with targeted therapy in the BM niche. This may explain APL's low relapse rates with ATRA + ATO and the high incidence of ATRA toxicities in these patients. Interestingly, ATO's effect on the BME also restores FLT3‐mutant AML sensitivity to gilteritinib, potentially reducing MRD. With the availability of nontoxic targeted therapies and precise MRD quantification methods, the field is ripe for clinical interventions aimed at preventing relapse through the eradication of MRD. Targeting CYP26B1‐mediated retinoid metabolism, whether by repurposing Food and Drug Administration (FDA)‐approved agents like ATO or by developing novel inhibitors, could offer a promising strategy to enhance the efficacy of existing targeted therapies.

AUTHOR CONTRIBUTIONS

David Kegyes: Data curation; formal analysis; writing—review and editing; writing—original draft. Patric Teodorescu: Investigation; writing—original draft; methodology; writing—review and editing; formal analysis; data curation. Teodora Supeanu: Writing—review and editing; methodology; investigation; writing—original draft; formal analysis; data curation. Yuya Nagai: Conceptualization; writing—review and editing; methodology; formal analysis. Guo Zhong: Writing—review and editing; methodology; formal analysis; data curation. Vikram Mathews: Resources; writing—review and editing. Nina Isoherranen: Investigation; conceptualization; writing—review and editing; validation; methodology; formal analysis; data curation. Gabriel Ghiaur: Conceptualization; investigation; funding acquisition; writing—original draft; methodology; validation; writing—review and editing; formal analysis; project administration; supervision; resources.

CONFLICT OF INTEREST STATEMENT

Gabriel Ghiaur received research funding from AbbVie Inc., Menarini Richerche, Kinomica Inc., and Arcellx Inc. Gabriel Ghiaur served on the advisory board of Syros Inc. Gabriel Ghiaur holds a patent for the use of IRX195183. All other authors have no conflicts to declare.

ETHICS STATEMENT

The study was performed under a research protocol approved by the Johns Hopkins Institutional Review Board. Patient consent statement: Patients were consented in accordance with the Declaration of Helsinki and under a research protocol approved by the Johns Hopkins Institutional Review Board.

FUNDING

D.K. was funded by a research scholarship of the Romanian Ministry of Research, Innovation and Digitalization (Bursa Henri Coandă). G.G. was supported by P01CA225618, R01 CA253981, and P30 CA006973‐57S2, and a Break Through Cancer Award. G.Z. and N.I. were supported by R01 GM111772.

Supporting information

Supporting Information.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.