Posttranscriptional activity of eIF4E contributes to HoxA9-driven leukemogenesis Open Access

TO THE EDITOR:

More than 50% of patients with acute myeloid leukemia (AML) are characterized by twofold to eightfold elevation of homeobox A9 (HoxA9) protein which predicts poor outcomes.^1-8 The pathogenic activities of HoxA9 are attributed to its transcriptional activity.⁷ However, HoxA9 acts beyond transcription but its relevance to leukemogenesis was unknown. For example, HoxA9 physically interacts with the eukaryotic translation initiation factor 4E (eIF4E), also dysregulated in AML.⁹ In many patients, eIF4E is highly elevated and nuclear-enriched corresponding to inappropriate elevation of eIF4E-dependent nuclear export of oncogenic messenger RNA (mRNAs), an activity tied to its oncogenic capacity.^10-12 eIF4E has been targeted in AML clinical trials with ribavirin, a competitor of its normal ligand the m⁷G RNA cap.^11,13,14 Ribavirin targeted eIF4E in patients corresponding to objective clinical responses, including complete remissions.^11,13,14 Interestingly, eIF4E and HoxA9 interacted in vitro, in specimens from patients of primary AML, and HoxA9 stimulated eIF4E activities in mRNA export and translation in AML.⁹ However, the relevance to HoxA9-mediated leukemogenesis was not examined. Here, we demonstrated that HoxA9 relies on eIF4E for leukemogenesis revealing a new therapeutic vulnerability in HoxA9-dependent AML.

To disrupt the HoxA9-eIF4E axis, we targeted its conserved eIF4E-binding motif (11-YVDSFLL-17),⁹ mutating residues 11 and 16 to alanine (11-AVDSFAL-17, HoxA9AA; Figure 1A). Purified wildtype glutathione S-transferase (GST)–HoxA9 bound to endogenous eIF4E in K562 cell lysates, whereas HoxA9AA only exhibited background binding for GST (Figure 1B). No interaction was observed with the negative control, eIF4E binding-protein 1. In K562 cells stably overexpressing the respective HoxA9 constructs, purified GST-eIF4E bound to HoxA9 wild type (HoxA9wt), whereas HoxA9AA was much weaker (Figure 1C). HoxA9 and HoxA9AA proteins were produced at equivalent levels which were approximately threefold more than endogenous HoxA9 present in vector (Figure 1C, left panel). GST-eIF4E bound to eIF4E-bindingprotein 1 (4E-BP1), a positive control, but not GST (Figure 1C). Bait protein loading was equivalent (supplemental Figure 1A-B). HoxA9AA and HoxA9wt had comparable binding to cognate HoxA9 DNA-binding site by electrophoretic mobility shift assays (Figure 1D). Super-shift assays using FLAG antibody confirmed HoxA9wt and HoxA9AA complexed to DNA with equivalent affinity. The HoxA9 homeodomain N51S mutant did not bind DNA, as expected¹⁵ (Figure 1D). Taken together, HoxA9AA was impaired in eIF4E binding without impacting its association with cognate DNA.

Next, we monitored the mutant’s impact on eIF4E-dependent mRNA export by measuring RNA content in nuclear and cytoplasmic compartments of K562 cells by quantitative reverse transcription polymerase chain reaction (RT-qPCR) (Figure 1E; supplemental Figure 1C).^9,18,19 HoxA9wt stable overexpression stimulated export of MCL1 and MYC by approximately twofold vs vector, whereas HoxA9AA was inactive (Figure 1E). Consistently, HoxA9wt stimulated MYC and MCL1 production by more than twofold, whereas Hox9AA only produced marginal increases (Figure 1F). Neither HoxA9 construct altered levels of target mRNA consistent with arising from posttranscriptional effects (Figure 1E, lower panel). The negative control glyceraldehyde-3-phosphate dehydrogenase was insensitive to eIF4E, HoxA9wt or HoxA9AA (Figure 1E-F). Given HoxA9AA was deficient in binding and stimulation of eIF4E, we utilized it to explore the relevance of eIF4E for HoxA9-dependent leukemogenesis.

For this purpose, we utilized a well-established AML mouse model that employs co-overexpression of HoxA9 and MEIS1 based on their frequent co-overexpression in patients with poor prognosis AML.^3,6,20 First, we monitored the ex vivo effects of the mutation on plating efficiency of sorted primary murine human stem and progenitor cells (HSPCs) Lin^–/Sca-1⁺, c-Kit⁺ cells by performing serial replating assays in cytokine supplemented methylcellulose after retroviral transduction of freshly isolated murine HSPCs with HoxA9wt or HoxA9AA and Meis1. Importantly, the colony forming potential of HoxA9AA-transduced cells was reduced by 2.8-fold relative to HoxA9wt after 4 rounds of replating indicative of a major reduction in leukemic capacity (Figure 2A).

Next, we evaluated the eIF4E contribution to HoxA9-induced leukemogenesis in vivo. CD45.2⁺ Lin^–/Sca-1⁺, c-Kit⁺ cells overexpressing HoxA9wt-IRES-GFP or HoxA9AA-IRES-GFP and MEIS1-IRES-YFP were transduced into lethally irradiated CD45.1⁺ recipient mice (Figure 2B). Transduction efficiency was similar for HoxA9wt and HoxA9AA, 22.9% and 24.5%, respectively (supplemental Figure 2A). Both groups showed comparable engraftment with similar kinetics based on fractions of CD45.2 and green fluorescent protein (GFP) positive peripheral blood cells at 4 weeks posttransplantation (supplemental Figure 2B). Both eventually developed myeloid leukemia with 97% to 99% CD45.2⁺/GFP⁺ in the bone marrow (Figure 2C). Diseased mice exhibited myeloid (Mac1/Gr-1) but not B- or T-lymphoid markers and maturation profiles were indistinguishable between groups (supplemental Figure 2C-D).

There were highly significant differences in latency of disease onset and penetrance between HoxA9wt and HoxA9AA groups. All animals (n = 10) in the HoxA9wt group succumbed to aggressive leukemia within 120 days (median, 90 days), corresponding to full penetrance of HoxA9wt/Meis1 leukemia (Figure 2C). By contrast, the median latency in the HoxA9AA/Meis1 group (n = 14) was 280 days. Two HoxA9AA/Meis1 mice never developed leukemia. One mouse had a low frequency of GFP positive cells in peripheral blood and bone marrow when sacrificed 413 days after transplantation. The second had detectable GFP⁺ peripheral blood cells until day 175 after transplantation with no remaining GFP⁺ cells in the bone marrow at sacrifice (day 190). HoxA9AA protein levels were marginally higher than HoxA9wt in mouse leukemic cells indicating that reduced leukemia is not a result of lower HoxA9AA production nor due to differences in eIF4E levels between groups (Figure 2A, right panel).

In secondary recipients, retransplantation of bone marrow cells from leukemic animals demonstrated an even more pronounced difference in latency and penetrance (Figure 2D). Although secondary recipients transplanted with 5/5 HoxA9wt/Meis1 developed myeloid leukemia and died within 60 days after transplantation, only 2 of 5 in the HoxA9AA/Meis1 group developed disease even a year after transplantation. The remaining 3 recipients had no detectable GFP⁺ cells in peripheral blood or bone marrow. Thus, the eIF4E interaction is continuously required for maintenance of leukemic transformation by HoxA9. Furthermore, eIF4E-dependence is not overcome by leukemic evolution.

Given these findings, we examined the sensitivity of HoxA9-dependent cells to therapeutic targeting of eIF4E. Treatment with a clinically achievable dose of ribavirin, 10 uM,^11,13,14 reduced proliferation in HoxA9wt leukemic murine cells approximately threefold with only modest impacts on HoxA9AA cells suggestive of an oncogene addiction to eIF4E in the HoxA9wt group (Figure 2E). To explore this in a human context, we compared human AML cells with established HoxA9 elevation and dependency (MV4-11, SHI-1) to cells characterized by reduced HoxA9 sensitivity and low levels (Kasumi-1, HL60) confirming expression differences by droplet polymerase chain reaction (∼10 mRNA copies per cell MV4-11 and SHI-1 vs <0.1 copies per cell for Kasumi and HL60) (https://www.proteinatlas.org/ENSG00000078399-HOXA9/cell+line).^22-25 Genetic reduction of eIF4E with 2 different short hairpin RNAs reduced eIF4E levels in both groups (Figure 2F). Proliferation of HoxA9-dependent cells was reduced up to 10-fold by eIF4E knockdown relative to short hairpin RNA controls. Reduction in low-HoxA9 cells only marginally impacted growth, and indeed, HL60 cells continued to proliferate (Figure 2F). Ribavirin treatment displayed similar trends with HoxA9-dependent cells approximately eightfold more sensitive to ribavirin than low-HoxA9 cells as seen by comparison of 50% inhibitory concentrations, which were in clinically achievable ranges (Figure 2G).^11,13,14 Thus, high-HoxA9 cells develop an oncogene dependency to eIF4E providing a new therapeutic vulnerability.

To examine the impact on transcription, we analyzed our primary leukemic murine cells and observed that both HoxA9AA and HoxA9wt induced the expected HoxA9 gene signature (GSE288783). A comparison of HoxA9wt and HoxA9AA transcriptomes revealed only minor differences with 163 downregulated and 247 upregulated genes of 23 589 transcripts observed (>1.5 fold, P <.05) (Figure 2H). Genes in HoxA9-related pathways were not altered (Figure 2H). In all, the reduced leukemogenic capacity of HoxA9AA is not caused by driving an alternative transcriptional program.

In conclusion, we discovered a posttranscriptional mechanism underpinning HoxA9-mediated leukemogenesis. HoxA9AA elicited a highly similar transcriptional program and bound DNA indistinguishably from HoxA9wt; however, this mutation no longer stimulated eIF4E, which correlated to substantially extended disease latency and reduced penetrance in mice. eIF4E interaction plays an essential role in induction and maintenance of HoxA9-dependent leukemogenesis. Strategies to directly target HoxA9 in patients remain elusive. We discovered a new therapeutic vulnerability whereby high-HoxA9 AML patients could benefit from eIF4E inhibitors like ribavirin.

All animal studies were approved by the Beth Israel Deaconess Medical Center/Harvard Medical School and were used in accordance with protocols approved by the institutional animal care and use committee.

Acknowledgments: The authors are grateful for assistance from Laurent Volpon for purified protein production.

This work was supported by the National Institutes of Health (RO1CA098571 and RO1CA080728 to K.L.B.B.; 1P01HL131477-6 A1 to D.G.T.); the Canada Research Chair programme (K.L.B.B.); and the AGA/Jenzabar Foundation (D.G.T.).

Contribution: F.Z., C.B., L.F., Y.M., B.C.-K., K.L.B.B., and D.G.T. designed the experiments; B.C.-K., C.B., L.F., and Y.M. performed experiments; and F.Z., C.B., L.F., B.C.-K., K.L.B.B., and D.G.T. wrote the manuscript.

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

Correspondence: Daniel G. Tenen, Harvard Stem Cell Institute, Center for Life Science, Room 437, Boston, MA; email: dtenen@bidmc.harvard.edu; and Katherine L. B. Borden, Department of Pharmacology and Robert H. Lurie Cancer Center, Northwestern University, Chicago, IL. 60611; email: katherine.borden@northwestern.edu.