WT1 and DNMT3A play essential roles in the growth of certain patient AML cells in mice

TO THE EDITOR:

Patients with acute myeloid leukemia (AML) experience poor prognosis, and precision oncology represents an attractive therapeutic option, applying targeted therapies against so-called dependencies.^1-4 Dependencies are essential components required for cell growth and survival; they represent attractive therapeutic targets as their inhibition reduces tumor burden.^1-4 

Many genes recurrently mutated in AML contribute to oncogenesis,^5,6 which may imply a role as dependency and allow precision therapy, based on genetic profiling. Examples already in routine clinical practice include AML with mutated FMS related receptor tyrosine kinase 3 treated with midostaurin and AML with mutated isocitrate dehydrogenase responding to ivosidenib.² Herein, we asked whether additional recurrently mutated genes might represent dependencies in established AML.

Previous efforts to identify dependencies used established cell lines, including large-scale functional genomic screens; WT1 and DNMT3A were shown to be dispensable in AML cell lines.⁷ As a limitation, cell lines might acquire nonphysiologic alterations, and discrepant results have been described (eg, between cell lines and organoids).^8,9 To approximate the clinical situation, we studied patient-derived xenograft (PDX) models^10,11 and mimicked the complex in vivo situation by performing CRISPR/CRISPR associated protein 9 (Cas9) knockout (KO) studies in mice. Using this highly patient-related in vivo approach, we identified WT1 and DNMT3A as yet unknown dependencies in a subset of patients' AML tumor cells.

From our toolbox of serially transplantable AML xenografts,¹² models derived from 7 patients were selected for the study (supplemental Tables 1-3, available on the Blood website). Genetically engineered PDX (GEPDX) models were generated that stably expressed recombinant Cas9 (supplemental Figure 1A).

We had recently established in vivo CRISPR/Cas9 dropout screens in GEPDX models of acute lymphoblastic leukemia¹³; herein, we transferred the technique to AML, which resulted in favorable quality controls (Figure 1A; supplemental Figure 2A). The 34 most frequently mutated genes in AML were studied, restricted to gain-of-function or change-of-function mutations.⁵ A library was designed containing 5 single-guide RNAs per target gene, together with positive and negative controls (supplemental Tables 4 and 5); the library was cloned into a lentiviral vector that coexpressed recombinant markers to enrich successfully transduced cells, using our custom library multiplexed cloning (CLUE) technique (supplemental Figures 1 and 2A; supplemental Tables 3 and 4).¹⁴ 

A CRISPR/Cas9 dropout screen was performed with 5 GEPDX models. KO resulted in dropout in about half of all genes from the screen, albeit to varying degrees, and most KO induced similar effects across the PDX samples (Figure 1B; supplemental Figure 2B; supplemental Tables 6-8). Confirming the robustness of our technical approach, genes with known common essential function or genes required for the hematopoietic system were strongly depleted in the KO screen. Among them, NPM1 was a dropout hit and served as a positive control, as it is known to have a broad essential function in malignant cells (Figure 1B).⁷ Another expected hit was KRAS, which is one of the genes most frequently mutated across all cancers and known to represent a dependency in numerous tumor types, including AML.^15,16 

Hits from dropout screens require validation, and single-KO experiments were performed as competitive in vivo assays where all cell populations are studied under identical conditions within the same mouse, giving robust results at low resources.¹⁸ Recombinant fluorochromes enabled an unbiased differentiation of cell populations by flow cytometry (Figure 1C-D; supplemental Figure 3). For each gene of interest as well as for nontargeting controls, 3 different, highly efficient single-guide RNAs were tested in 3 independent mixtures (supplemental Figures 4 and 5). From the 7 PDX models studied, up to 5 PDX models gave reliable results for each gene.

NPM1 was included as a positive control, and KO of NPM1 completely eliminated AML GEPDX cells in all GEPDX models tested in vivo (Figure 1E). KRAS was studied in PDX models carrying mutant KRAS at variant allele frequencies of either 0 or close to 0.5, avoiding intrasample heterogeneity. KRAS KO revealed a strong dropout in all GEPDX models studied, which was significantly more pronounced in KRAS^mutant PDX models than KRAS^wildtype PDX models (Figure 1F; supplemental Figure 6). Thus, our PDX models strengthen previously published data showing that KRAS represents a dependency and attractive therapeutic target in AML, especially in tumors carrying a KRAS mutation.¹⁶ 

Next, we examined 2 genes with poorly defined roles in oncogenes and for which we had suitable PDX models with appropriate variant allele frequencies at hand (supplemental Table 1). Although data on WT1 as an oncogene are controversial,^19,20,DNMT3A mainly represents a tumor suppressor, required for hematopoietic differentiation.^21-24 

Reproducing published data with our own tools,⁷ we found no evidence that either WT1 or DNMT3A might play a role as dependencies in AML cell lines, with trends toward slightly increased proliferation rates on gene KO (supplemental Figures 7-9; supplemental Table 1). In contrast and surprisingly, in in vivo GEPDX models, we discovered a pronounced dropout of either of both genes on KO in certain PDX models (Figure 2A). Thus, WT1 and DNMT3A represent dependencies in a subset of PDX AML models in vivo, indicating an obvious discrepancy with their function in cell lines in vitro (Figure 2B), without any meaningful impact on the immunophenotype (supplemental Figure 10). PDX models showed dropout of WT1 or DNMT3A exclusively in the in vivo environment on which PDX cells depend as opposed to cell lines, suggesting that in vivo approaches are required to unmask certain dependencies in AML (Figure 2C). There was no correlation between dependency on DNMT3A and presence of a somatic hot spot mutation in DNMT3A in the GEPDX models (supplemental Figure 6D). In the transcriptome, KO of WT1 or DNMT3A was accompanied by regulation of biological processes, such as apoptosis and oxidative phosphorylation (Figure 2D; supplemental Figure 11).

When characterizing in vivo essentiality in more detail, we found that KO of WT1 induced a certain increase in the antitumor effect of cytarabine, an important drug in routine treatment of AML (supplemental Figure 12). WT1 KO reduced the capacity of AML-346 cells to home to the bone marrow environment on either intrafemural or intravenous cell injection followed by early in vivo growth disadvantage, suggesting an impaired tumor-niche interaction (supplemental Figures 13 and 14). KO of either WT1 or DNMT3A reduced the numbers of leukemia-initiating cells in competitive limiting dilution transplantation assays and prevented reengraftment of AML-346 cells into secondary recipient mice, with and without prior cell enrichment, indicating that stem cell surrogates were depleted on WT1 or DNMT3A KO (Figure 2E; supplemental Figure 15). Taken together, our data reveal that WT1 and DNMT3A represent dependencies in a subset of AML GEPDX models in vivo, suggesting that they might represent therapeutic targets.

Our study identified WT1 and DNMT3A as dependencies in a subset of patient AML PDX samples growing in vivo, although less pronounced and less frequent compared with KRAS. KO of WT1 and DNMT3A impaired PDX AML growth in vivo, attenuated the tumor-niche interaction, eradicated AML stem cells, and increased treatment response.

Although cell lines did not reveal the phenotype, PDX models proved valuable tools to identify dependency on WT1 and DNMT3 and might more closely resemble patient's tumors.^10,11 Our technique now allows studying gene dependencies in patient PDX models in vivo (eg, to personalize pharmacologic precision therapy). Our data encourage testing additional genes recurrently mutated in AML for their essentiality in PDX models in vivo (eg, additional dropout candidates from our screens).

The essential function of WT1 identified herein fits with its previously described oncogenic function,¹⁹ whereas different phenotypes between different PDX models might mirror conflicting data on WT1 obtained during leukemogenesis.^19,20 

For DNMT3A, a prevailing tumor suppressor function was described,^21-23 making a dependency function unlikely. Amid complexity, a tumor-supportive function of mutant DNMT3A was reported in specific AML subsets (eg, AML driven by a partial tandem duplication in KMT2A).²⁵ AML-388 harbors a KMT2A-AFDN translocation (supplemental Table 2), indicating that KMT2A-driven AML might preferably depend on DNMT3A.

Taken together, our molecular PDX AML in vivo studies allowed identifying WT1 and DNMT3A as dependencies and putative therapeutic targets in defined subsets of AML, warranting further evaluation.