Osimertinib Covalently Binds to CD34 and Eliminates Myeloid Leukemia Stem/Progenitor Cells Open Access

Acute myeloid leukemia (AML) is a lethal hematological malignancy, with a 5-year overall survival (OS) rate of less than 30% with chemotherapy (1, 2). Although 60% to 70% of patients with AML achieve complete remission after induction regimen (combination of cytarabine and anthracycline), most of them relapse within 3 years due to the outgrowth of chemotherapy-resistant leukemic stem/progenitor cells (1), highlighting the urgent need to develop new therapeutic approaches for treating this disease, especially for chemotherapy-resistant and relapsed patients.

The discovery of EGFR activating mutation as an effective therapeutic target and the development of EGFR tyrosine kinase inhibitors (EGFRi) have revolutionized the treatment of non–small cell lung cancer (NSCLC; ref. 3). First generation drugs, represented by erlotinib and gefitinib, reversibly interfere with ATP binding to EGFR and inhibit kinase activity in a noncovalent way. Second- and third-generation drugs, including afatinib, osimertinib, and rociletinib, covalently and irreversibly inhibit kinase by binding to an active cysteine residue in the EGFR (4, 5). In 2007 and 2008, two independent cases reported EGFRi erlotinib treatment induced complete remission of AML in patients with concomitant lung cancer (6, 7). Following in vitro studies revealed pro-differentiation effects of erlotinib against AML (8–12). Because EGFR is not expressed in AML cells (6, 8, 10, 13), “off-target” effects of EGFRi in AML have been proposed and a couple of nonreceptor tyrosine kinases including spleen tyrosine kinase (SYK) and Janus kinase 1 (JAK1) have been revealed to function downstream of EGFRi (8, 10).

Despite much efforts have been implemented in treating AML with EGFRi, multiple clinical trials employing erlotinib in AML yielded conflicting and mostly negative results (14–17). The discrepancy between preclinical and clinical data underscores the necessity to identify the patient subset that might optimally benefit from this approach. Moreover, it remains unknown whether the covalent EGFRi AZD9291 (osimertinib) has activity against AML. We show herein that osimertinib selectively induced apoptosis in CD34⁺ AML cells. High CD34 expression featured a subset of patients who might gain the greatest benefit from treatment with osimertinib.

Bone marrow samples were collected from leukemia patients diagnosed according to French-American-British classification at Department of Hematology of the Second Hospital of Dalian Medical University. Written informed consents were obtained from all patients in accordance with the Declaration of Helsinki, and all manipulations were approved by the Medical Science Ethic Committee of Dalian Medical University. Donors for allogeneic bone marrow transplantation were used to purify normal healthy CD34⁺ cells. KG-1 (CCL-246.1, RRID:CVCL_0374), Kasumi-1 (CRL-2724, RRID:CVCL_0589), U937 (CRL-1593.2, RRID:CVCL_0007), THP1 (TIB-202, RRID:CVCL_0006), K562 (CCL-243, RRID:CVCL_0004), Molm13 (RRID:CVCL_2119) and HT-29 (HTB-38, RRID:CVCL_0320) were purchased from ATCC and renewed every 2 years. Mycoplasma contamination was routinely tested. Details are available in Supplementary Material.

Osimertinib (HY-15772), erlotinib (HY-50896), ibrutinib (HY-10997), daunorubicin (HY-13062A), saracatinib (HY-10234), stattic (HY-13818), necrostatin-1 (HY-15760) were obtained from MedChemExpress. Puromycin was from Merck/Millipore. Antibodies against the following proteins were used: CD34 (Abcam, ab81289, RRID:AB_1640331), cleaved caspase-3 (Cell Signaling Technology, #9661, RRID:AB_2341188), RIPK1 (Cell Signaling Technology, #3493, RRID:AB_2305314), p-S166-RIPK1 (Cell Signaling Technology, #65746, RRID:AB_2799693), MLKL (Cell Signaling Technology, #14993, RRID:AB_2721822), phospho-MLKL (Cell Signaling Technology, #91689, RRID:AB_2732034), EGFR (Abcam, ab52894, RRID:AB_869579), phosphorylated Y1068-EGFR (Abcam, ab40815, RRID:AB_732110), Src (Proteintech, 60315, RRID:AB_2881426; Abcam, ab231081, RRID:AB_2917962), phosphor-Src (Cell Signaling Technology, #6943, RRID:AB_10013641), STAT3 (Cell Signaling Technology, #9139, RRID:AB_331757), phospho-STAT3 (Cell Signaling Technology, #9145, RRID:AB_2491009), AKT (Proteintech, 10176, RRID:AB_2224574), phospho-AKT (Cell Signaling Technology, 4056, RRID:AB_331163), ERK (Proteintech, 11257, RRID:AB_2139822), phospho-MAPK (Cell Signaling Technology, #4370, RRID:AB_2315112), Lyn (Proteintech, 60211, RRID:AB_10949077), Flag (Cell Signaling Technology, #14793, RRID:AB_2572291), HRP-labeled Streptavidin (Beyotime, A0303) and β-actin (Cell Signaling Technology, #4970, RRID:AB_2223172). Secondary antibodies for immunofluorescence staining are Alexa Fluor 488 and Alexa Fluor 555 (Invitrogen Molecular Probes, A11008, RRID:AB_143165 and A31572, RRID:AB_162543). Recombinant of human protein CD34 was from Abcam (ab182830, hCD34) or Sino Biological (10103, hCD34). BeyoMag Streptavidin Magnetic Beads was from Beyotime (P2151). Necroptosis Inducer Kit with TSZ was from Beyotime (C1058). Caspase Inhibitor Z-VAD-FMK was from Beyotime (C1202).

Primary CD34⁺ AML or chronic myeloid leukemia (CML) cells were cultured with osimertinib for 48 hours and were transplanted via tail vein into female 8- to 10-week-old sublethally irradiated (2.5Gy) NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ NSG mice (The Jackson Laboratory, RRID:BCBC_4142). Human cells were assessed using anti-human CD45 (Invitrogen, 11045942, RRID:AB_10852703), anti-mouse CD45 (Invitrogen, 12045182, RRID:AB_465668), anti-human CD34 (Invitrogen, 17034942, RRID:AB_2016672), anti-human CD33 (Invitrogen, 48033742, RRID:AB_2016671) by flow cytometry. Animal handling was approved by the committee for humane treatment of animals at Shanghai Jiao Tong University School of Medicine.

Survival analysis was performed on the bulk RNA sequencing (RNA-seq) from The Cancer Genome Atlas (TCGA; RRID:SCR_003193) and BeatAML cohorts. The gene expression and survival data were downloaded at cbioportal website (http://www.cbioportal.org/). We divided AML samples into the high and low expression groups based on the optimal cutpoint determined by R function “surv_cutpoint” in survminer package (v0.4.9, RRID:SCR_021094). The survival risk and statistical significance were determined according to the HR and log-rank P values reported by R package survival (v3.2–11, RRID:SCR_021137). Univariate analysis was performed using the Cox proportional hazard model. Kaplan-Meier survival curve was plotted by R package survminer (v0.4.9, RRID:SCR_021094).

DNA and total RNA were extracted from cryopreserved mononucleated cells, and libraries were prepared according to the protocol of the TruSeq RNA/TruSeq DNA Sample Preparation Kit (Illumina). Then, targeted DNA library was captured by using a panel of 38 commonly mutated genes in myeloid hematologic malignancies. Whole transcriptome RNA and target DNA sequencing (DNA-seq) were performed on a NovaSeq platform with paired-end 150 bp read-length by the Novogene Company (Beijing, China).

For gene expression, we mapped the sequencing data to the reference genome (hg38) using STAR (18) and obtained gene expression abundances reported as reads per kilobase per million mapped reads using the Cufflinks package (19) guided by the gene annotation format file (GTF file) from GENCODE (Release 27, GRCh38). For the target DNA-seq data, we mapped the sequencing data to the reference genome (hg38) using Burrows–Wheeler Aligner tool (20), and identified the variants under the GATK pipeline (21) and annotated the variants using ANNOVAR (22).

Statistical analyses between the control and treatment groups were performed by standard two-tailed Student t test. All experiments were repeated at least three times. A value of P < 0.05 was considered to be statistically significant.

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org, RRID:SCR_004055) via the iProX partner repository (23) with the dataset identifier PXD023328. The public AML expression and clinical data analyzed in this study were obtained from AML (TCGA, NEJM 2013) and AML (OHSU, Nature 2018) corresponding to TCGA and BeatAML cohorts at [http://www.cbioportal.org/]. Targeted next-generation sequencing data generated by this work are available at National Center for Biotechnology Information (NCBI) Sequence Read Archive (BioProject no. PRJNA1046209). The RNA-seq data have been deposited in Gene Expression Omnibus, and the accession number is GSE249138. All other raw data generated in this study are available upon request from the corresponding author.

To test the effects of osimertinib on leukemic stem/progenitor cells, we isolated CD34⁺ cells from patients with AML by FACS and then treated with osimertinib. The percentage of apoptosis was measured using Annexin V/PI assay. As shown in Fig. 1A, treatment with osimertinib from 0.5 to 2 μmol/L dose-dependently induced apoptosis of CD34⁺ leukemic stem/progenitor cells. Lower concentration (0.1–0.3 μmol/L) of osimertinib displayed apoptosis-inducing effect from 0.3 μmol/L (Fig. 1B; Supplementary Fig. S1A and S1B; refs. 3, 4). In contrast, the first generation noncovalent EGFRi erlotinib, which was reported to have activity against AML, and another covalent inhibitor ibrutinib for Bruton's tyrosine kinase did not achieve significant cytotoxicity until 10 μmol/L in both primary (Supplementary Fig. S1C) or CD34⁺ AML cell lines (Fig. 1C), suggesting a different mechanism between the noncovalent and covalent EGFRi (8, 9, 12). Notably, osimertinib also displayed remarkable apoptosis-inducing effects on primary CD34⁺ cells sorted from patients with CML (Fig. 1D), which population is believed to contribute to relapse of CML due to its insensitivity to the BCR-ABL inhibitor imatinib (24). Furthermore, CD34⁺CD38⁻ leukemia stem cells (LSC), which is the more immature compartment within CD34⁺ population and chemotherapy-resistant, displayed sensitivity to osimertinib (Fig. 1E). Combination of osimertinib with daunorubicin, a first-line chemotherapy drug in AML treatment, exhibited synergistic effect in killing LSC (Fig. 1E).

To discriminate whether osimertinib induces apoptosis or direct necrotic cell death, we examined the markers for apoptosis and necrotic cell death (necroptosis) in osimertinib-treated AML cells by Western blot. We observed cleavage of caspase-3 but not activation of either RIPK1 or MLKL by osimertinib as evidenced by absence of phosphorylated-RIPK1 and phosphorylated-MLKL (Fig. 1F; refs. 25, 26). We took classic necroptosis inducer TSZ (combination of TNFα, Smac mimetic, and a pan-caspase inhibitor z-VAD-FMK) treated HT-29 as a positive control. In addition, co-treatment of inhibitor of caspase-3 (Supplementary Fig. S1D) but not inhibitor for necrotic cell death necrostatin-1 suppressed osimertinib-induced cell death in KG-1 cells (Fig. 1G; Supplementary Fig. S1E). These data further validated the apoptosis-inducing effect of osimertinib.

We next evaluated the effect of osimertinib on CD34⁻ leukemia cells purified from AML and CML patients. As depicted in Fig. 2A, exposure of CD34⁻ cells, which was confirmed to be part of malignant blasts but not normal lymphocytes (Supplementary Fig. S2A), to osimertinib did not induce comparable cell death in contrast to its CD34⁺ counterparts. Moreover, we administrated osimertinib to the bone marrow mononuclear cells (BMMC) from a number of AML samples (Fig. 2B; Supplementary Fig. S2B) or cell lines (Fig. 2C) with different CD34 expression level and observed that CD34⁺ cells within the bulk population were more susceptible than CD34⁻ population. Collectively, these findings strongly suggested a specificity of osimertinib towards CD34⁺ leukemia progenitors.

To reveal the mechanism of EGFRi, we first examined the expression of EGFR on CD34⁺ leukemia stem/progenitor cells from AML and CML patients. Taking the EGFR⁺ lung carcinoma cell lines A549 and H1299 as positive controls, we conducted Western blot, immunofluorescent staining and flow cytometry analysis using antibodies against EGFR or Y1068-phosphorylated EGFR, an active form of EGFR (27). In accordance with previous observations (6, 10, 13), EGFR protein in CD34⁺ leukemia stem/progenitor cells were barely detectable (Supplementary Fig. S3A–S3C).

Next, we investigated EGFR-independent mechanisms underlying the inhibitory effects of EGFRi in CD34⁺ leukemia stem/progenitor cells. We carried out quantitative phosphotyrosine(Y) proteomic analyses to explore intracellular signaling events in response to osimertinib. CD34⁺ cells isolated from 3 AML individuals were pooled and treated with osimertinib, followed by analysis of tyrosine-phosphorylated sites targeted by osimertinib using LC/MS-MS. Previously reported large-scale phosphoproteomic studies indicated that off-targets of EGFR inhibitors are preferentially intracellular non-receptor tyrosine kinases including Syk and Lyn (10, 28, 29). As expected, the group of proteins annotated as tyrosine protein kinases was significantly overrepresented as being under-phosphorylated in the phosphoproteomes of osimertinib-treated cells (Fig. 3A; Supplementary Table S1). Notably, the phosphorylation of a significant portion of proteins was inhibited upon osimertinib exposure, which was further supported by STRING analysis (Fig. 3A and B; Supplementary Table S2). Meanwhile, motif analysis revealed the osimertinib targeted “-E-X-X-pY” motif, which is enriched (fold increase 4.66 vs. 3.88) in downregulated phosphotyrosine sites compared with all regulated sites (Fig. 3B, right). Previous study has showed that overexpression of myeloid Src family kinases (SFK) contributed to AML survival and is associated with poor prognosis (30). Among the 7 protein tyrosine kinases containing the “-E-X-X-pY” motif (Supplementary Table S3), there are 4 proteins, Src, Lyn, Fyn and Hck, belonging to the SFKs, indicating a pivotal role of SFK in responding to osimertinib (31–34). A remarkable loss of Y329 phosphorylation of CD34, the surface marker of stem/progenitor cells that lacks tyrosine kinase activity, was observed after osimertinib exposure (Fig. 3A; Supplementary Table S1; refs. 35, 36). Next we validated the downregulation of Y329 phosphorylation of CD34, but no changes in the protein level of CD34 (Fig. 3C and D) by osimertinib treatment using parallel reaction monitoring-based targeted MS (PRM-MS) in 5 AML patient samples as well as CD34-expressing KG-1 and Kasumi-1 leukemia cell lines (Fig. 3C-E; Supplementary Fig. S4-S10). This finding was recapitulated in a set of AML specimens using a custom-designed antibody against the phosphorylated Y329 residue of CD34 (CD34-pY329; Supplementary Fig. S11). Taken together, these data demonstrate that treatment with osimertinib inhibits the phosphorylation of Y329 in CD34.

We next investigated how EGFR inhibitors associate with CD34. One common feature of leukemia stem/progenitor cells from AML and CML patient samples is the expression of CD34. Moreover, EGFRi do not kill CD34⁻ leukemia cells. We proposed that CD34 might be a non-tyrosine kinase target of EGFRi. To test this hypothesis, a chemical probe biotin-tagged osimertinib (hereafter named biotin-osimertinib; Fig. 4A; Supplementary Fig. S12A), which retained the ability to induce CD34⁺ leukemia cell apoptosis (Supplementary Fig. S12B), was synthesized through click reaction. Incubation of two different purified recombinant CD34 (rCD34) with biotin-osimertinib showed a dose-dependent binding starting from 10 nmol/L of compound, which could be competed away by high concentrations of unlabeled osimertinib (Fig. 4B; Supplementary Fig. S12C), indicating a specific binding of osimertinib with CD34 in vitro. Cellular target engagement of CD34 was further confirmed through incubating biotin-osimertinib with KG-1 cell lysates followed by blotting the precipitates with antibody against CD34 (Fig. 4C).

To determine the specific residue modified by osimertinib in CD34 molecule, we incubated purified rCD34 with osimertinib followed by MS analysis. We got 44% sequence coverage of CD34 protein and identified all 6 cysteines within the cysteine-rich domain. The m/z ratio of the C199-containing peptide TSSCAEFKK was measured as 1,056.49 in the absence of osimertinib and 1,498.74 in the presence of osimertinib (Fig. 4D). The calculated mass shift was consistent with the addition of one molecule of osimertinib. Similar mass shift was also observed on the C177 residue (Supplementary Fig. S12D), while the other four cysteines showed negative binding with osimertinib. These data indicated that C199 and C177 are the covalent binding sites for osimertinib.

To understand the structural basis of the C199 and C177 binding with osimertinib, we predicted 3D structure of CD34 (aa169–285) using AlphaFold Protein Structure Database and performed molecular docking (37, 38). Three linkages (C177-C199, C215-C229, and C190-C242) could be formed respectively by the six cysteines with both C215-C229 and C190-C242 inaccessibly buried (Supplementary Fig. S12E; refs. 37, 39). A cavity surrounding the C177-C199 linkage on the surface of CD34 is readily accessible to form covalent bond with compounds. Osimertinib was then docked into this binding surface, forming a covalent bond via it's α,β-unsaturated carbonyl in the proximity of both C199 and C177 through Michael addition reaction. In contrast, no favorable interaction could be formed between ibrutinib and C199 or C177 due to the distance between the β-carbon and both cysteines (Fig. 4E), providing a molecular basis for selectivity of osimertinib to CD34.

To validate that CD34 is the endogenous target mediating the effect of osimertinib, we depleted CD34 in primary CD34⁺ AML cells, KG-1 and Kasumi-1 cells via the CRISPR-Cas9 system with specific guide RNAs (Fig. 4F). As shown in Fig. 4G and H, knockout of CD34 rendered primary leukemia cells less sensitive to osimertinib, which could be restored by re-expression of CD34^WT but not CD34^C199S/C177S, indicating that the binding to C199 and C177 of CD34 is essential for fully mediating the effect of osimertinib. Of note, single point mutation of C199 rescued most but not the entire effect of CD34^WT, suggesting that both cysteine residues are required to mediate the signaling of osimertinib-CD34 (Fig. 4H).

Next, we attempted to delineate the connection between osimertinib modification of extracellular cysteines and downregulation of the intracellular phosphorylation of Y329 of CD34 molecule. It is speculated that the covalent binding of osimertinib may interfere the interaction between CD34 and its kinases. We pooled CD34⁺ cells from 3 AML individuals, treated with/without osimertinib and conducted endogenous immunoprecipitation–mass spectrometry (IP-MS) assays using an antibody against CD34. Among the CD34-interacting proteins, Src was identified to be downregulated upon osimertinib treatment (Supplementary Table S4) and was further validated in KG-1 and Kasumi-1 cells as well as primary CD34⁺ cells using antibody against CD34 (Fig. 5A–C). The association between Src and CD34 was further confirmed through IP assay using an antibody against Src (Fig. 5A and B) and immunofluorescent staining (Fig. 5D). More importantly, the interaction between endogenous CD34 or exogenous CD34^WT but not CD34^C199S mutant with Src was impaired upon osimertinib treatment (Fig. 5E–G). These data indicate that extracellular modification by osimertinib may induce conformational change that interferes CD34-Src interaction.

We then investigated whether Src functions as the protein kinase for CD34. The SFKs is a family of proto-oncogenic nonreceptor tyrosine kinases comprising nine members (Src, Lck, Hck, Fyn, Blk, Lyn, Fgr, Yes, and Yrk) that presented redundant activity (31). Among them, Lyn-associated phosphopeptides were also significantly downregulated upon osimertinib treatment (Fig. 3A), although CD34-IP-MS did not identify Lyn as CD34-interacting protein (Supplementary Table S4). In an in vitro kinase assay, where recombinant kinases were incubated with Y329-containing CD34 peptide followed by MS analysis, Src and Lyn, but not previously identified Syk (data not shown), could directly phosphorylate CD34 at Y329 (Fig. 5H; refs. 10, 40). We overexpressed WT or Y329F in sgCD34 KG-1 cells and observed that Y329F-expressing cells grew slower than WT cells (Fig. 5I), suggesting a growth-promoting role of Y329 phosphorylation. Overall, these data suggested that the engagement of osimertinib on CD34 contributes to its apoptosis-inducing effect.

Given that sgCD34 or CD34^C199S/C177S mutation could not fully abolish osimertinib-induced cell death (Fig. 4G and H), we speculated that osimertinib may also function in a CD34-independent way in CD34⁺ AML cells. Previous studies have identified SFK as one of the key constitutively activated tyrosine kinases in AML cells (12, 31). Our phosphotyrosine (Y) proteomic data revealed downregulation of SFK members including Src, Lyn, Fyn, and Hck in response to osimertinib (Fig. 3A). Subsequently, we tested the activity of SFK signal pathway in CD34⁺ leukemia cells. Primary CD34⁺ AML and CML cells were treated with osimertinib and several major signaling pathways were probed. Phosphorylation of Src and its downstream STAT3 (41), but not ERK or AKT, were remarkably inhibited in both primary and cultured AML cells (Fig. 5J, left, and K). Supporting this finding, SFK inhibitor saracatinib and STAT3 inhibitor stattic induced synthetic lethality with osimertinib in primary CD34⁺ cells (Fig. 5L and M). Notably, osimertinib could downregulate Src phosphorylation in CD34⁻ cells from AML-#11 and CML-#2 (Fig. 5J, right), indicating that osimertinib could induce cell death in a CD34-independent way. Intriguingly, osimertinib did not affect protein level (Supplementary Fig. S13A and S13B) and localization (Supplementary Fig. S13C) of WT or C199S mutant CD34. Taken together, we proposed that osimertinib inhibited the activity of SFK and its downstream STAT3, which contributes to AML survival.

It has been shown that CD34⁺ expression and CD34⁻ expression occurs in AML and indicates outcome of leukemias (42, 43). To understand the prognostic significance of CD34 expression in AML, we analyzed CD34 in publicly available data from BeatAML (n = 200; ref. 44) and TCGA (n = 173; ref. 45) cohorts. Firstly, CD34 high expression exhibited a significant correlation with refractory patients and was observed in adverse patient group according to 2017-ELN (European Leukemia Net) classification in BeatAML (Supplementary Fig. S14A and S14B; ref. 46). Secondly, univariate Cox regression analysis revealed an association of high expression of CD34 genes with decreased OS in BeatAML and TCGA cohorts (Fig. 6A). Notably, CD34 overexpression remained significantly associated with worse OS when survival analysis was applied to CD34 plus FLT3-ITD, NPM1, CEBPA, and CBF fusion (Fig. 6B; ref. 43). Overall, we proposed that CD34 serves as an adverse prognostic marker for AML.

To delineate the molecular pathogenic features of patients with high level of CD34 (CD34^high), we performed targeted DNA-seq and RNA-seq analysis in a larger AML population (n = 94). On the basis of the median expression of CD34, 94 patients with AML were categorized into CD34^high and CD34^low group. Targeted DNA-seq illustrated higher frequency of mutations in CEBPA (P = 0.0008), BCOR (P = 0.0485), and SF3B1 (P = 0.01405) in CD34^high group, in contrast to DNMT3A (P = 2.483E-6), NPM1 (P = 3.808E-6), and FLT3 (P = 0.01323) in CD34^low group, indicating distinct mutation patterns (Fig. 6C). The correlation of CEBPA, DNMT3A, NPM1, and FLT3 mutations with CD34 were further validated in TCGA and BeatAML cohorts (Supplementary Fig. S14C). To investigate the expression features associated with high level of CD34, we purified CD34⁺ cells from patients with AML and normal donors and performed RNA-seq and quantitative proteomics analysis. RNA-seq analysis revealed that compared with normal donors, 32 genes were upregulated, among which, CD34 resided in the top 10 (Fig. 6D; Supplementary Fig. S14D). Higher expression of CD34 in AML was also seen in proteomics data (Fig. 6E), indicating that CD34 is transcriptional overexpressed in AML cells. Strikingly, gene set enrichment analysis (GSEA) revealed high consistency between RNA-seq and proteomics data, with MYC targets, DNA repair, and oxidative phosphorylation as the top upregulated pathways, while complement and inflammatory response as the top downregulated pathways (Fig. 6F and G), suggesting that the driving mechanism in leukemogenesis occurs at the transcriptional level.

Next, the in vivo effects of osimertinib on CD34⁺ leukemia stem/progenitor cells were further evaluated using AML and CML patient-derived xenograft (PDX) models. We sorted CD34⁺ cells from AML or CML patients and exposed them to osimertinib for 48 hours before transplantation into sublethally irradiated NSG mice (Fig. 7A). Flow cytometry analysis revealed the presence of human CD45⁺ cells in the bone marrow 16 weeks posttransplantation, indicating successful engraftment of human LSCs (Fig. 7B and C). Osimertinib-treated CD34⁺ cells displayed decreased engraftment as evidenced by reduced numbers of human CD45⁺ (hCD45⁺) and human CD33⁺ (hCD33⁺) cells in the bone marrow of both AML and CML PDX models (Fig. 7B and C; Supplementary Fig. S15A).

To further assess the clinical relevance of our findings, we continued to investigate whether posttransplant treatment with osimertinib could improves survival in mice. Samples from patients with different CD34 expression levels were injected into sublethally irradiated NSG mice (Fig. 7D). Six weeks posttransplantation, we treated the mice daily for a period of 4 weeks with vehicle (ctrl) or osimertinib (5 mg/kg/day; ref. 4). Administration of osimertinib significantly extended the survival of recipient mice injected with CD34^high AML samples but not those with CD34^low samples (Fig. 7E, right). Importantly, reduced percentage of hCD34⁺ cells was observed in osimertinib-treated CD34^high PDX models (Fig. 7E, middle). Consistently, the AML burden, as evaluated by the infiltration of leukemia cells into liver and spleen with tissue damage as well as the proportion of hCD45⁺ cells in the bone marrow, were remarkably decreased (Supplementary Fig. S15B–S15D).

Importantly, osimertinib treatment on normal CD34⁺ cells displayed less toxicity as evaluated by Annexin V/PI and colony formation assays compared with its AML counterpart (Fig. 7F–H; Supplementary Fig. S15E). Osimertinib inhibits AML CD34⁺ cells at significantly lower doses compared with normal BM CD34⁺ cells, reflecting a substantial therapeutic window in favor of AML cells (AML IC₅₀ 0.76 μmol/L vs. normal 5.54 μmol/L; Fig. 7I). Furthermore, we demonstrated in vivo that osimertinib showed no impairment to the engraftment of normal CD34⁺ cells, as evidenced by comparable percentage of hCD45⁺ cells between osimertinib-treated and control group (Fig. 7J). In support of this finding, clinical administration of osimertinib in patients with NSCLC did not yield significant hematopoietic toxicity (5, 47).

In preliminary clinical studies, we tested the efficacy of osimertinib (80 mg once daily) in two CD34^high individuals who have failed to respond to any available therapies. Patient 1, with 80.3% CD34⁺ cells in BMMC (Fig. 2B, AML-#7), responded to osimertinib treatment as evidenced by the recovery of white blood cells (dropped sharply from 24.9×10⁹/L to 7.1×10⁹/L at day 2 and remained below 5×10⁹/L till day 17) accompanying with an improvement of red blood cells (Fig. 7K and L; Supplementary Table S5 and S6). A bone marrow biopsy demonstrated a reduction in blast percentage from 11.1% at day 0 to 5% at day 14 (Fig. 7L). Similarly, Patient 2 displayed sharp drop of white blood cell count from 32.8×10⁹/L to 7.7×10⁹/L on day 3 and remained below 5×10⁹/L till day 9 (Fig. 7M), accompanying with an improvement of platelet count and blast percentage drop from 19.5% to 4% (Fig. 7M and N; Supplementary Table S5). Collectively, these data suggest an opportunity to evaluate osimertinib's usage for treating patients with AML from the subgroup with high level of CD34 and warrant clinical investigation in the treatment of refractory/resistant AML.

CD34 is a heavily glycosylated sialomucin-type transmembrane phosphoglycoprotein, first identified on stem/progenitor cells (35, 48). Recent evidences demonstrated a wider context of CD34 expression including muscle satellite cells, corneal keratocytes, interstitial cells, epithelial progenitors, and vascular endothelial progenitors (48). Clinically, sorting CD34⁺ cells is employed to enrich donor hematopoietic stem cells for allogeneic stem cell transplantation. Despite its well-known role as a stem cell marker, the function of CD34 remains poorly understood. The most well-defined role of CD34 is in cytoadhesion as a binding partner of L-selectin (49–51). However, controversial evidence suggests a role of CD34 in blocking cell adhesion (52). Structurally, CD34 possesses an extracellular domain containing mucin domains with multiple O-linked and N-linked carbohydrates, followed by a cysteine-rich domain with undetermined function. The cytoplasmic domain contains a consensus site for tyrosine phosphorylation but lacks enzymatic domain motifs. It is therefore speculated that CD34 transduces signals through binding adaptor proteins such as CRKL to modulate biological processes including adhesion (53). In this study, we described previously unknown characteristics of CD34 featuring with the cysteine-rich domain as the target of covalent tyrosine kinase inhibitors. Recently developed orally administrated covalent inhibitors including ibrutinib, afatinib and osimertinib have displayed multiple advantages versus conventional reversible inhibitors including the potential for more sustained target engagement and prolonged pharmacodynamic effects in cancers (6, 54). To the best of our knowledge, this is the first evaluation of administration of the third generation EGFR inhibitor on AML cells and patients. Importantly, our results provide a rational for targeting this domain in developing CD34-assoiciated drug therapies in various clinical context including stem cell mobilization. CD34 can serve as a therapeutic target for eradicating stem/progenitor cells while sparing the healthy counterpart, likely due to the discrepancy in CD34 expression level between leukemia and normal CD34⁺ cells. Interestingly, the two original cases of EGFR-inhibitor responsive AML that occurred in NSCLC were both diagnosed as M1, with 88% and 81% of blasts CD34⁺ (6, 7). Furthermore, our results suggest that leukemia stem/progenitor cells are vulnerable to SFK-STAT3 perturbation, supporting more preclinical and clinical evaluation of EGFRi in AML treatment. These data collectively demonstrate CD34-dependent and -independent mechanisms of osimertinib (Fig. 7O).

The clinical responses of 2 patients with AML to anecdotal osimertinib monotherapy suggested significant antileukemia activity, particularly as these patients had experienced multiple rounds of chemotherapy or had poor performance status precluding treatment with conventional induction regimens. Osimertinib therapy may serve as a bridge to allogenic hematopoietic stem cell transplantation for patients with refractory AML, although transplant was not performed on the 2 patients in this study. Unexpectedly, EGFR covalent inhibitors also present activity against stem/progenitor cells in CML. Administration of imatinib and second generation of TKI has transformed CML from a fatal malignancy to a manageable disease with lifelong therapy (24, 55). However, CML LSCs are independent of BCR-ABL for survival and resistance to imatinib remains a significant clinical challenge. The selective killing of CD34⁺ cells by osimertinib suggests that synergistic use of imatinib with osimertinib may provide an attractive approach to target BCR-ABL-independent mechanism of resistance.

In summary, our work reveals that osimertinib targets CD34 with significant anti-myeloid leukemia activity. Despite CD34 expression is employed as a marker on hematopoietic stem/progenitor cells, the data presented herein unveil an unanticipated and novel function of CD34 in the transduction of hematopoietic stem/progenitor cells survival signaling and define a biomarker for the selection of the AML patient subset that may benefit from EGFR inhibitor therapy.

No disclosures were reported.

L. Xia: Conceptualization, investigation. J.-Y. Liu: Investigation. M.-Y. Yang: Investigation. X.-H. Zhang: Conceptualization. Y. Jiang: Investigation. Q.-Q. Yin: Methodology. C.-H. Luo: Investigation. H.-C. Liu: Investigation. Z.-J. Kang: Resources. C.-T. Zhang: Resources. B.-B. Gao: Resources. A.-W. Zhou: Resources. H.-Y. Cai: Resources. E.K. Waller: Writing–original draft. J.-S. Yan: Writing–original draft. Y. Lu: Conceptualization, resources, funding acquisition, writing–original draft.

This work was supported by National Natural Science Foundation (82170179, 81970131, 81370652, 81770146 to Y. Lu; 32171431 to L. Xia) and Foundation from Science and Technology Commission of Shanghai Municipality (22S11900400 to Y. Lu).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.