Loss of the stress sensor GADD45A promotes stem cell activity and ferroptosis resistance in LGR4/HOXA9-dependent AML

Acute myeloid leukemia (AML) is a heterogeneous and malignant clonal disease initiated by self-renewing leukemia stem cells (LSCs) arising from the transformation of hematopoietic stem cells (HSCs) or committed progenitor cells.¹ Preleukemic cells transformed from HSCs by oncogenes (e.g., MLL-AF9) can develop a highly aggressive subtype of AML with chemoresistance and expressing genes associated with inferior survival in patients with AML.^2,3 Current chemotherapy is insufficient in targeting the intrinsic self-renewal mechanism in LSCs, which are a source of therapy resistance and relapse in AML.⁴ We have previously shown that the heterogeneity of LSC self-renewal potential can be driven by distinct or partially overlapping regulatory mechanisms, where the interaction of G-protein–coupled receptor LGR4 and its ligand R-spondin 3 (RSPO3) is required for regulation of HSC-derived LSCs through β-catenin activation in AML that depends on HOXA9, a predictor of poor prognosis in a wide variety of human malignancies.^5-10 High LSC activity in a heterogeneous LSC pool is associated with poor clinical response to therapy.^11,12 It is thus of critical importance to identify key regulators that control leukemia-initiating cell activity essential for acquisition of an aggressive leukemia phenotype and therapeutic resistance.

Here, we found that LGR4 negatively regulates GADD45A, which acts as a crucial stress sensor in response to replication stress-induced DNA damage and genotoxic stress-induced ferroptosis. Inhibition of Lgr4 in HSC-derived leukemic cells induces Gadd45a expression, and deletion of Gadd45a in AML LSCs prevents the reduction of β-catenin induced by Lgr4 depletion, thus impacting a key regulator of self-renewal in hematopoietic development and malignancies. Gadd45a-deleted mice have a normal hematologic phenotype but display genomic instability, impairment in DNA repair, and accumulation of DNA damage and mutations in HSCs.^13,14 This is in line with our observation in LSCs, where deletion of Gadd45a causes mutations in DNA repair genes and replication stress while upregulating antioxidants to attenuate stress-induced reactive oxygen species (ROS) and iron production, leading to resistance to ferroptosis (an iron- and ROS-dependent cell death).¹⁵ DNA methylation of GADD45A is predictive of poor overall survival in AML and is associated with DNMT3A and IDH1/2 mutations, while downregulation of GADD45A via FLT3-ITD mutation contributes to the myeloid differentiation block.^16,17 In contrast, elevated GADD45A expression in normal HSCs induces differentiation via activation of p38 mitogen-activated protein kinase signaling that leads to the removal of damaged HSCs preventing malignant transformation.^14,18 GADD45A thus acts in a cell context–specific manner and is differentially required in normal and malignant hematopoiesis. However, the functional consequence of GADD45A loss in AML, especially in LSCs, has been largely unexplored.

Additional details are provided in the supplemental Methods (available on the Blood website).

Generation and use of Gadd45a knockout (Gadd45a^−/−) and wild-type (Gadd45a^+/+) mice have been described previously.¹⁹ MLL-AF9 AML mouse models and patient-derived xenografts (PDXs) were established by transplanting leukemic cells into female C57BL/6 or NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice (Australian BioResources), as previously described.^5,8,10,20 All animal studies were in strict compliance with the Institutional Animal Care and Ethics Committee–approved protocols.

Using a stringent cutoff, our microarray data identified 15 genes upregulated by short hairpin RNA (shRNA)–mediated Lgr4 knockdown (Lgr4-sh1)⁵ in MLL-AF9–induced AML. Briefly, leukemic cells were generated by transducing HSC-enriched LSK cells (Lin^–Sca-1⁺c-Kit⁺)¹ with MLL-AF9–green fluorescent protein (GFP) and transplanting resultant preleukemic cells into sublethally irradiated C57BL/6 syngeneic mice. After developing AML, GFP⁺ leukemic cells flow sorted from mouse bone marrow were lentivirally transduced with Lgr4 shRNA and selected for the microarray experiment (Figure 1A; supplemental Figure 1A).

Among 15 upregulated genes, Gadd45a has been reported to negatively regulate β-catenin activity in epithelial tumors through promoting Gsk3β activation via dephosphorylation of Ser9.²³ We found that Lgr4 depletion increased Gadd45a expression in leukemic cells (Figure 1B). We have recently documented an indispensable role for Lgr4 in regulating self-renewal through cooperation with Hoxa9 in aggressive AML.⁵ Overexpression of Lgr4 (Lgr4^OE) in HOXA9/MEIS1-transformed LSK cells not only produces leukemia with a shortened latency, similar to that in MLL-AF9–induced AML, but also upregulates β-catenin, reaching a level comparable to that in MLL-AF9–transformed LSK cells.⁵ Consistently, overexpression of Lgr4 downregulated Gadd45a in HOXA9/MEIS1 leukemic cells, reaching a level similar to that in MLL-AF9 leukemic cells (Figure 1C; supplemental Figure 1B). This is in line with a negative correlation between LGR4 and GADD45A in 244 patients with AML from The Cancer Genome Atlas (TCGA) dataset²¹ that was mainly attributed to patients with unfavorable outcome (Figure 1D). These data indicate LGR4 as a negative regulator of GADD45A in poor-prognosis AML.

We next investigated the mechanism for Lgr4-mediated downregulation of Gadd45a. The TCGA dataset revealed a positive correlation between FOXO3 and GADD45A (Supplemental Figure 1C), which was consistent with a negative correlation between LGR4 and GADD45A in patients with AML with an unfavorable outcome (Figure 1D). This implicates FOXO3-mediated regulation of GADD45A in poor-prognosis AML. Knockdown of Lgr4 decreased the phosphorylation of FoxO3a in MLL-AF9 LSCs (Lin^–Sca-1^–c-Kit^highCD16/32^highCD34⁺)¹ (supplemental Figure 1D). The phosphorylation of FOXO3A, and its consequent cytoplasmic localization and inactivation, is a poor prognostic factor in AML, particularly in those with FLT3 mutations.²⁴ Consistent with the role of p-PKAc as a downstream effector of Lgr4 pathway,⁵ the specific inhibitor of PKA, PKI, reduced the phosphorylation of FoxO3a, leading to increased Gadd45a in MLL-AF9 LSCs and in human MLL-AF9 AML (THP-1) cells (supplemental Figure 1E-F). These results support p-PKAc/p-FOXO3A–mediated downregulation of GADD45A.

To define a direct role for Gadd45a loss in Lgr4/β-catenin signaling, we tested the effect of Gadd45a deletion on Wnt/β-catenin activity in MLL-AF9-GFP transformed HSC/progenitor-enriched c-Kit⁺ cells,¹⁰ which were isolated from the bone marrow of Gadd45a knockout mice.¹⁹ Compared with wild-type Gadd45a (Gadd45a^+/+), deletion of Gadd45a (Gadd45a^−/−) in GFP⁺ preleukemic MLL-AF9 cells (pre-MLL^c-Kit+) increased endogenous β-catenin and inactivated/phosphorylated Gsk3β (p-Gsk3β^Ser9), while upregulating key self-renewal signature genes, such as WNT/β-catenin targets (c-Fos, Tcf7l2, and Ccnd1)⁵ and MLL-fusion targets (Mef2c and HoxA cluster genes Hoxa7 and Hoxa11)¹ (Figure 1E). In contrast, knockdown of β-catenin repressed expression of self-renewal signature genes and reduced colony formation in Gadd45a^−/− cells (supplemental Figure 1G). Gadd45a^−/− cells exhibited enhanced serial-replating capacity, indicative of increased self-renewal (Figure 1F), whereas overexpression of Gadd45a inhibited leukemic cell growth (supplemental Figure 1H). Likewise, knockdown of LGR4 in human AML (THP-1) cells elevated endogenous GADD45A expression, whereas deletion of GADD45A via CRISPR/Cas9 increased inactive p-GSK3β^Ser9 and nuclear nonphosphorylated (active) β-catenin as well as key WNT/self-renewal target genes (supplemental Figure 2A-D). To further determine the role of Gadd45a loss as a downstream effector of LGR4 signaling, we generated Gadd45a^+/+ vs Gadd45a^−/− LSCs expressing or depleting Lgr4, by transplanting pre-MLL^c-Kit+ cells into syngeneic mice to develop AML followed by lentiviral transduction of LSCs with Lgr4 shRNAs. We observed that knockdown of Lgr4 with 2 independent shRNAs (sh1 and sh2)⁵ decreased β-catenin expression in Gadd45a^+/+ LSCs, but not in Gadd45a^−/− LSCs (Figure 1G). These data suggest that loss of GADD45A sustains β-catenin levels and WNT/self-renewal activity through phosphorylation of GSK3β, acting downstream of LGR4 signaling.

We assessed the impact of Gadd45a loss on AML pathogenesis in vivo, and observed a significantly shortened latency for the mice transplanted with Gadd45a^−/− compared with Gadd45a^+/+ pre-MLL^c-Kit+ cells (primary AML; Figure 2A-B; supplemental Figure 3A). This suggests that Gadd45a loss promotes leukemia progression. To test the role of Gadd45a loss in regulating the self-renewal and frequency of LSCs in vivo, we performed serial transplantation and limiting-dilution assays. GFP⁺ MLL-AF9 leukemic cells flow sorted from mice with primary AML were serially transplanted into syngeneic recipient mice. Gadd45a^−/− leukemic cells gradually shortened disease latencies on serial transplantation in mice (secondary/tertiary/quaternary AML; Figure 2C-E) compared with Gadd45a^+/+ cells, consistent with a progressively increasing self-renewal potential and an aggressive phenotype similar to that of relapsed human MLL-rearranged AML. In vivo limiting-dilution assay showed that deletion of Gadd45a caused a 10-fold increase in LSC frequency in Gadd45a^−/− compared with Gadd45a^+/+ mice (1/142 and 1/1650; P = .00114; Figure 2F). Likewise, immunophenotypic analysis revealed a 2-fold increase of Gr-1^-/lowc-Kit^high LSC-enriched population in Gadd45a^−/− compared with Gadd45a^+/+ leukemic cells (Figure 2G; supplemental Figure 3B). We have previously documented that the Gr-1^-/lowc-Kit^high population is >100-fold enriched for LSCs compared with the Gr-1⁺c-Kit^high population in a heterogeneous LSC pool of AML.¹⁰ Loss of GADD45A thus results in leukemia-initiating cell enrichment and increased LSC frequency, consistent with a correlation between high LSC frequency at diagnosis and poor prognosis in patients with AML.¹¹ 

As Gadd45a^−/− mice exhibit decreased DNA repair and increased mutation frequency,¹³ we examined whether deletion of Gadd45a promoted AML progression by inducing genomic instability and mutations. Our whole-genome sequencing of LSCs from primary and tertiary AML showed that deletion of Gadd45a induced substantial mutations (Figure 2H-I), particularly missense mutations involved in DNA repair (Ddx11, Rad21),²⁵ self-renewal (HoxA clusters: Hoxa4, Hoxa5, Hoxa9, and Hoxa10),¹ and DNA methylation (Dnmt1)²⁶ (supplemental Figure 3C). Sanger sequencing confirmed persistent missense variants of Hoxa5 and Hoxa9 induced by Gadd45a deletion in LSCs from primary and tertiary AML (supplemental Figure 3D), demonstrating the accuracy of whole-genome sequencing data. Notably, Gadd45a deletion induced not only a missense mutation in exon 21 of the Ddx11 gene but also multiple synonymous mutations in exons 19 to 26, leading to decreased expression of DEAD/H-box helicase 11 (Ddx11) (supplemental Figure 3E). Because Ddx11 is a DNA helicase essential for double-stranded break repair during DNA replication,²⁵ loss of Gadd45a likely causes replication stress through DNA repair gene variants, leading to genomic instability and DNA damage accumulation in LSCs. Because no additional mutations were found in key Gadd45a targets over serial transplantation, the enhancement of the leukemic potential in Gadd45a^−/− mice was likely because of increased self-renewal, associated with HoxA cluster gene variants, which endowed LSCs with a competitive advantage. This finding aligns with our recent discovery demonstrating a positive cooperation between HOX and LGR4 pathways in promoting self-renewal and leukemic potential in LGR4/HOXA9-dependent AML.⁵ 

We next examined the role of Gadd45a loss as a determinant of therapy resistance in LSCs. Compared with Gadd45a^+/+ LSCs, Gadd45a^−/− LSCs exhibited a lower level of ROS and a higher proportion of quiescent G0 cells (Figure 2J; supplemental Figure 3F), which are critical features associated with increased resistance to therapy.²⁷ Consistently, treatment with the chemotherapeutic agent doxorubicin (a mainstay of AML therapy, producing high levels of ROS) had little effects on Gadd45a^−/− LSCs but reduced the colony-forming capacity of Gadd45a^+/+ LSCs (Figure 2K). Subsequent cell proliferation assay at 12 days after injection of mice, following ex vivo pretreatment with doxorubicin, revealed impaired proliferation of Gadd45a^+/+ LSCs, but not Gadd45a^−/− LSCs (Figure 2L). This is consistent with aberrant ROS production induced by doxorubicin or menadione (ROS inducer) but rescued by N-acetylcysteine (ROS scavenger) in Gadd45a^+/+ LSCs, rather than Gadd45a^−/− LSCs (Figure 2M), underlining ROS-specific and Gadd45a-dependent effects. These data support a protective role of GADD45A loss against chemotherapy-associated oxidative stress in LSCs, and underscore the importance of GADD45A expression in determining LSC properties essential for leukemia progression and therapy resistance.

We have recently demonstrated that RSPO3-LGR4 activation is required for LSC survival in PDX mouse model of relapsed AML harboring mutations, including DNMT3A, RUNX1, and K/NRAS.⁵ Here, we investigated the impact of GADD45A loss on AML progression in established PDXs expressing firefly luciferase and mCherry. We found that deletion of GADD45A by CRISPR/Cas9 caused increased AML burden, as determined by firefly luciferase–based bioluminescence imaging in PDX mice (Figure 3A-B; supplemental Figure 4A-B). Subsequent secondary transplantation in NSG mice showed that GADD45A deletion enhanced the long-term leukemia-initiating activity of AML cells, which was associated with increased WNT/self-renewal signature (Figure 3C-D). This supports the role for GADD45A loss in promoting in vivo engraftment of human AML cells and leukemia-initiating activity.

GADD45A loss upregulates pathways involved in self-renewal and antioxidant defense at the single-cell level

To understand how loss of GADD45A contributes to therapy resistance in aggressive AML, we performed CITE-seq (cellular indexing of transcriptomes and epitopes by sequencing)²⁸ that used DNA-barcoded antibodies to tag cell-surface markers of human LSCs, enabling simultaneous quantification of mRNA and antibody-derived tag on a given single cell. Our results showed that the LSC compartment (CD34⁺CD38^–CD90^–CD45RA⁺)²⁹ in human AML PDX cells was enriched specifically in cluster 2 (supplemental Figure 5A-B). Deletion of GADD45A resulted in a 3-fold increase in the LSC-enriched CD34⁺ population and a 2-fold increase in the LSC compartment (Figure 4A-B). Single-cell transcriptomic profiling identified 43 differentially expressed genes (DEGs) upregulated in GADD45A-deleted LSCs compared with CRISPR-control LSCs, including genes associated with LSC survival and self-renewal (e.g., DUSP1 and c-FOS)^30,31 and iron detoxification and storage (e.g., ferritin heavy/light chain [FTH1/FTL])^32,33 (Figure 4B; Supplemental Figure 5C). Pathway enrichment analysis of DEGs using MSigDB uncovered several self-renewal pathways, including the HOXA9 pathway (e.g., c-FOS) (Figure 4C). Strikingly, c-FOS, as a key WNT/β-catenin target that directly interacts with β-catenin for transcriptional activation of downstream components,³⁴ was present in most of the top 9 oncogenic pathways, implicating its crucial role in GADD45A loss-mediated β-catenin self-renewal activation. Subsequent Gene Ontology analysis showed that the DEGs were enriched in molecular function (Figure 4D), particularly in iron ion binding (e.g., FTH1 and FTL). Likewise, our fast preranked gene set enrichment analysis of the Reactome pathway database identified GADD45A loss-induced enrichment of antioxidant-related gene sets, including iron uptake and transport (e.g., FTH1 and FTL) and ROS sensing/detoxification (eg, PRDX1, GSTP1)^35,36 (Figure 4E). These observations agree with reduced ROS levels and increased self-renewal in GADD45A-deleted LSCs, underlining a possible involvement of GADD45A loss in antioxidant defense against iron and ROS accumulation (a defining feature of ferroptosis).¹⁵ 

Our analysis of TCGA AML dataset²¹ revealed a correlation between high expression of antioxidants FTH1 and PRDX1 with unfavorable outcomes in a cohort of 172 patients with AML (Figure 4F). This is in line with a recent report showing that high levels of FTH1 and FTL in a cohort of 525 patients with AML are associated with chemoresistance and enrichment of genes related to oxidative stress and iron pathways.³⁷ Likewise, our functional studies showed that knockdown of GADD45A induced endogenous FTH1 expression (supplemental Figure 6A). Overexpression of FTH1 reduced intracellular ROS and iron, whereas knockdown of FTH1 elevated ROS levels in human AML (THP-1) cells (supplemental Figure 6B-D). These data suggest FTH1 acting downstream of GADD45A loss to provide antioxidant defense against iron and ROS accumulation.

We next explored the role for GADD45A loss in preventing ferroptosis through upregulation of FTH1. RSL3 (a selective ferroptosis inducer)³⁸ triggered ferroptotic/nonapoptotic cell death and reduced in vitro colony-forming capacity and in vivo cell proliferation in Gadd45a^+/+ LSCs, rather than Gadd45a^−/− LSCs, from mice with tertiary AML (Figures 2D and 5A-C). The sensitivity of Gadd45a^+/+ LSCs to RSL3 treatment could be attributed to the ability of RSL3 to induce Gadd45a expression and reduce Fth1 in Gadd45a^+/+ LSCs, but not Gadd45a^−/− LSCs (Figure 5D). In support of this notion, knockdown of Fth1 promoted RSL3-induced ferroptosis in both Gadd45a^−/− and Gadd45a^+/+ LSCs, revealing increased ROS and ferrous iron (Fe²⁺) accompanied by decreased cell viability (Figure 5E-G). This is in line with the role of Fth1 acting downstream of Gadd45a loss to counteract RSL3-induced ROS and iron and consequently prevent genotoxic stress-induced DNA damage and ferroptosis. The protective effect of Gadd45a deletion against RSL3-induced exogenous DNA damage was demonstrated by the result of RSL3-induced increase in phosphorylated H2AX (γ-H2AX, a biomarker for DNA double-stranded breaks)³⁹ in Gadd45a^+/+ LSCs, but not in Gadd45a^−/− LSCs, and by a similar result obtained from the alkaline comet assay (genotoxicity testing for DNA-strand breaks)⁴⁰ (Figure 5H-I). Notably, a higher level of endogenous DNA damage was observed in Gadd45a^−/− than Gadd45a^+/+ LSCs (Figure 5I), consistent with the role of Gadd45a deletion in increasing replication stress and stress-induced DNA damage (Figure 2H). These data demonstrate GADD45A loss as a key determinant in preventing stress-induced DNA damage and RSL3-induced ferroptosis, primarily by upregulating FTH1-mediated antioxidant defense that scavenges excessively produced ROS and iron.

To investigate the role of GADD45A loss in regulating ferroptosis resistance in human AML cells, we treated mCherry⁺ PDX cells with RSL3. CRISPR/Cas9-mediated deletion of GADD45A prevented RSL3-induced ferroptosis; conversely, CRISPR-control PDX cells that expressed low levels of GADD45A responded to RSL3 treatment, causing ∼40% cell death (Figure 6A; supplemental Figure 7). RSL3-induced ferroptosis could be rescued by pretreatment with a ferroptosis inhibitor, ferrostatin-1 that inhibits ferroptosis by reducing iron levels,¹⁵ which suppressed RSL3-induced expression of endogenous GADD45A (Figure 6B). These results support GADD45A as a positive regulator of ferroptosis, and deletion of GADD45A promotes ferroptosis resistance in human AML cells.

We further assessed the association of GADD45A expression with ferroptosis in primary AML patient specimens that express high levels of LGR4/HOXA9 (9p deletion, MLL-AF10, MLL-AF9) vs relatively low levels of LGR4/HOXA9 (acute promyelocytic leukemia, or APL; AML with normal karyotype, or AML-NK), as shown in our recent studies.⁵ Lower expression of GADD45A was shown in LGR4/HOXA9-high than LGR4/HOXA9-low cells (Figure 6C; supplemental Figure 8A-B). Lack of endogenous GADD45A in AML with 9p deletion or MLL-AF10 rendered LGR4/HOXA9-high cells resistant to RSL3 treatment; conversely, expression of GADD45A in APL or AML-NK, even at much lower levels than paired remission samples, sensitized LGR4/HOXA9-low cells to RSL3 treatment, leading to ferroptosis with increased iron and ROS accompanied by reduced cell viability (Figure 6C-F; supplemental Figure 9A-B). RSL3 induced ferroptosis by upregulating GADD45A that could be suppressed by ferroptosis inhibitor ferrostatin-1 in LGR4/HOXA9-low cells; however, RSL3 had no impact on LGR4/HOXA9-high cells that intrinsically lacked GADD45A (Figure 6C-G). This suggests a threshold level of endogenous GADD45A required for ferroptosis induction. These data underline a proactive role of GADD45A in RSL3-induced ferroptosis; loss of GADD45A provides a shield against this mechanism of cell death in aggressive AML that is dependent on LGR4/HOXA9 expression.

LSCs exhibit functional heterogeneity, particularly in HSC-derived AML with adverse/intermediate-risk cytogenetics, characterized by high self-renewal and frequency of leukemia-initiating cells alongside low response to chemotherapy.^2,3,5,41 High LSC activity is associated with elevated relapse risk and poor survival in patients with AML.^11,12 We have recently demonstrated an indispensable role for RSPO3-LGR4 signaling in the formation and function of HSC-derived LSCs through aberrant activation of p-PKAc/p-CREB/CBP-p300/β-catenin pathway, contributing to aggressive phenotypes in HOXA9-dependent AML.⁵ Here, we identify GADD45A as a key negative regulator of leukemia-initiating activity and therapy resistance, acting downstream of LGR4/p-PKAc/p-FOXO3A. Loss of GADD45A in a heterogeneous LSC pool increases LSC self-renewal and frequency and decreases chemosensitivity, leading to a more aggressive AML phenotype. Several mechanisms may act in concert to promote GADD45A loss-associated leukemogenic potential: deletion of GADD45A increases WNT/β-catenin self-renewal activity (e.g., c-FOS and HOXA cluster) through GSK3β phosphorylation (inactivation), and induces replication stress and mutations in DNA repair and self-renewal genes contributing to enhanced cooperation of LGR4 and HOX pathways in promoting stemness and self-renewal, while conferring resistance to oxidative stress and ferroptosis via upregulation of FTH1-mediated antioxidant defense that stores iron in a nontoxic form and scavenges excessive ROS production induced by replication stress or therapy-induced genotoxic stress.

ROS is an important determinant of stem cell fate. Low ROS closely correlates with increased self-renewal and quiescence in normal and malignant stem cells,^27,42 and a slight increase of intracellular ROS induces LSC differentiation.⁴³ Consistently, deletion of GADD45A in LSCs reduces ROS levels and promotes self-renewal in a mouse model of AML development; however, it also causes replication stress and DNA damage accumulation, which have the potential to increase ROS production.⁴⁴ This raises a critical question of how low ROS is sustained in LSCs under persistent DNA damage. Our single-cell cellular indexing of transcriptomes and epitopes by sequencing data on patient-derived LSCs and subsequent functional studies directly link GADD45A deletion to aberrant activation of antioxidant defense against ROS and iron accumulation. GADD45A loss serves as a safeguard by upregulating FTH1-mediated antioxidant defense, which attenuates ROS generated by replication stress-induced DNA damage and thereby maintains ROS at a basal nontoxic low level to protect LSC compartment from oxidative stress.

Antioxidant defense upregulated by GADD45A loss also acts as the protective mechanism against genotoxic stress-induced iron and ROS accumulation in response to chemotherapeutic agent doxorubicin and ferroptosis inducer RSL3, leading to therapy resistance. Ferroptosis is an iron (Fe²⁺)-dependent oxidative cell death caused by ROS generated through Fenton reaction and subsequent lipid peroxidation.¹⁵ Ferroptosis resistance remains a significant challenge to therapeutic exploitation of the iron-dependent cell death mechanism. In this study, we identify GADD45A loss as a key regulator of ferroptosis resistance. Deletion of GADD45A in murine LSCs or lack of GADD45A in LGR4/HOXA9-dependent patient cells (indicative of poor outcome) renders cells resistant to ferroptosis induced by RSL3. Our pathway enrichment analysis uncovers abnormal activation of iron detoxification driven by ferritin (FTH1/FTL), an iron storage protein complex.^32,33 FTH1 has ferroxidase activity to convert Fe²⁺ to Fe³⁺ for iron storage in a nontoxic form via FTL that prevents excess Fe²⁺ to produce ROS via the Fenton reaction or by forming the doxorubicin-Fe²⁺ complex.^33,45 Upregulation of FTH1 is correlated with unfavorable outcomes in patients with AML, and FTH1 modulates intracellular iron and ROS levels acting downstream of GADD45A in AML LSCs. These findings underline a crucial role of GADD45A loss in therapy resistance by enhancing FTH1-mediated antioxidant defense that limits iron-dependent ROS production, allowing LSCs to survive under persistent genotoxic stress without surpassing a deadly threshold in a subgroup of leukemia with poor prognosis.

In addition to ferritin (FTH1/FTL) as an antioxidant defense system, our single-cell studies also uncover aberrant activation of pathways related to LSC survival (e.g., DUSP1) and antioxidants (e.g., PRDX1 and GSTP1) in GADD45A-deleted human LSCs. Like FTH1, high expression of PRDX1, which can scavenge excess ROS and decreases damage from oxidative stress,³⁵ correlates with unfavorable prognosis in patients with AML. The positive correlation between FTH1 and DUSP1 expression in patients with AML from TCGA (supplemental Figure 10) implies a possible cooperation between FTH1 and DUSP1 pathways in ferroptosis resistance and LSC survival, acting downstream of GADD45A. Thus, loss of GADD45A may play a crucial role in protecting leukemia-initiating cells against oxidative stress and ferroptosis by simultaneous modulation of multiple pathways, suggesting that targeting of individual downstream effectors may not be sufficient to fully restore GADD45A tumor-suppressor function. In this regard, therapeutic approaches to induce GADD45A expression in LSCs likely represent a more robust strategy to compromise LSC function and modulate cell sensitivity to anticancer agents associated with oxidative stress and ferroptosis.

Ferroptosis inducers may be 1 approach to promote GADD45A expression in AML cells expressing this gene even at low levels, but are unable to restore GADD45A in cells lacking its expression. This is because a threshold level of endogenous GADD45A is required for ferroptosis induction. Alternative approaches are needed for AML cells lacking GADD45A. Given our findings that LGR4 pathway inhibition induces GADD45A expression and GADD45A acts downstream of LGR4/p-PKAc/p-FOXO3, restoring GADD45A in LSCs can be achieved by blocking this pathway in a significant group of AML. In particular, we observe the lack of GADD45A in primary patient specimens expressing high LGR4/HOXA9, which critically depend on RSPO3-LGR4 activation for LSC survival and maintenance of myeloid undifferentiated state.⁵ We have recently demonstrated that a clinical-grade anti-RSPO3 antibody specifically targets LGR4 signaling via disruption of the RSPO3-LGR4 interaction, which compromises in vivo leukemia-initiating activity in primary AML PDX mouse models that harbor high-risk cytogenetics (e.g., MLL-AF9 or 9p deletion)⁵ and lack GADD45A. Notably, we found that in vivo anti-RSPO3 antibody treatment induced endogenous GADD45A in NSG mice engrafted with primary MLL-AF9 AML patient cells (supplemental Figure 11A-C), thereby offering a potential strategy to overcome resistance to ferroptosis. Nevertheless, we cannot exclude the possibility of other factors, such as DNA methylation, in coordinately regulating GADD45A. We have previously shown that LGR4 decreased intracellular ROS,⁵ which affects DNA methylation by acting on the activity/expression of DNA methyltransferases (DNMTs).⁴⁶ The DNMT inhibitor, decitabine, which predominantly inhibits DNMT1 activity,⁴⁷ induced GADD45A expression in MLL-AF9 LSCs (supplemental Figure 11D). This indicates a coordinated downregulation of GADD45A in LGR4/HOXA9-dependent LSCs by multiple mechanisms, including LGR4/p-PKAc/p-FoxO3a and DNA methylation.

In summary, we have identified a key regulatory role for GADD45A loss in promoting LSC self-renewal and stemness and in driving oxidative resistance, including ferroptosis resistance, in LGR4/HOXA9-dependent AML. Understanding the mechanisms by which GADD45A manipulates iron-dependent ROS accumulation provides additional insights on ferroptosis regulation and unlocks a new arsenal to reverse resistance by restoring the GADD45A activity in combination with ferroptosis induction or chemotherapy for treating this poor prognosis subset of leukemia lacking GADD45A.

The authors thank Irmela Jeremias and Karsten Spiekermann for providing acute myeloid leukemia patient-derived xenograft cells; Basit Salik, Karthik B. Polpaya, Gargi Kulkarni, Jayvee Datuin, Sheng X. F. Chen, Jonason Yang, Sayali Gore, and the staff from the Ramaciotti Centre for Genomics at the University of New South Wales for technical assistance; and the Sydney Children’s Tumor Bank Network for providing primary patient samples and related clinical information for this study.

This work was supported by grants from the Leukemia & Lymphoma Society, United States (TRP-23664-23); Cancer Council NSW, Australia (RG 22-06); Anthony Rothe Memorial Trust, Australia (RRE/0700:sz); and Tour de Cure, Australia (RSP-187-2020) (to J.Y.W.).

Contribution: N.H. and B.M. performed drug treatment and ferroptosis-related experiments with input from X.D.Z., T.L. and J.W.; H.Y. performed murine model experiments; J.S., A.B., and N.H. performed single-cell experiments with input from D.R.C.; L.G.-B. analyzed cellular indexing of transcriptomes and epitopes by sequencing data with assistance from J.S. and T.C; A.D. performed in vivo bioluminescence imaging; S.E.S., D.A.C., R.J.D., and D.A.L. supplied mouse Gadd45a^−/− bone marrow cells and provided input on murine model experiments; R.J.D, D.A.L., S.E.S., D.A.C, J.W., G.M.M., M.N., M.H., M.K., and B.B.C. reviewed the manuscript; and J.Y.W. conceived, designed, and analyzed the study and wrote the manuscript.

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

Correspondence: Jenny Y. Wang, Cancer and Stem Cell Laboratory, School of Medical Sciences, Faculty of Medicine and Health, University of Sydney, Kolling Institute, Sydney, NSW, Australia; email: jenny.wang@sydney.edu.au.