Pharmacological inhibition of the transcription factor PU.1 in leukemia

PU.1 knockdown decreases cell growth and clonogenic capacity and increases apoptosis of murine and human AML cells.

To determine whether PU.1 inhibition may be a suitable strategy in AML, we used an established model of AML driven by reduced PU.1 levels, PU.1 URE^–/– AML, in which PU.1 expression is reduced to approximately 20% of normal levels by disruption of an upstream enhancer (URE) (12, 17). The PU.1 URE^–/– AML cell line has been established from a leukemic mouse with homozygous deletion of the URE of the PU.1 gene, which has been previously described (17). We selected 3 shRNAs that decreased PU.1 expression in mouse and human cells (Supplemental Figure 1, A and B; supplemental material available online with this article; https://doi.org/10.1172/JCI92504DS1). Knockdown of PU.1 in PU.1 URE^–/– AML cells by the 3 different shRNAs led to significantly decreased cell growth and colony formation (Figure 1, A and B). Likewise, the percentage of apoptotic cells was substantially increased upon shRNA-mediated PU.1 knockdown in PU.1 URE^–/– AML cells (Figure 1C). The degree of inhibition of growth and clonogenicity, as well as apoptosis induction, were greater with the shRNA PU.1_2, leading to more efficient PU.1 knockdown (Supplemental Figure 1A). Knockdown of PU.1 in an immature murine hematopoietic cell line with normal levels of PU.1 (BaF3) did not have significant effects on proliferation or apoptosis (Supplemental Figure 1, C–F).

We next investigated the effect of PU.1 knockdown on human leukemic cell lines with different PU.1 levels (Supplemental Figure 1G). MOLM13 and Kasumi-1 cell lines harbor anomalies associated with low PU.1 levels [FLT3-ITD mutation for MOLM13 and t(8;21) for Kasumi-1] (3, 4), while THP1 cells have higher PU.1 levels. Decreased PU.1 led to a strong inhibitory effect on the growth and clonogenic capacity of MOLM13 and Kasumi-1 cells, whereas it had no or very modest effects on THP1 cell growth (Figure 1, A and B). Accordingly, the apoptotic fraction was increased following PU.1 knockdown in MOLM13 and Kasumi-1 cells, but we observed no major effect in THP1 cells (Figure 1C). To confirm that this phenotype was correlated with PU.1 levels, we overexpressed PU.1 in MOLM13 cells upon PU.1 knockdown with shPU.1_2 and found a rescue of the phenotype, with increased clonogenic capacity and decreased apoptosis (Supplemental Figure 1, H–J).

To determine whether PU.1 inhibition has an effect on primary cells from AML patients, we seeded mononuclear cells (MNCs) from AML patients in semisolid media for 2 weeks and assessed the number of colonies, the number of viable cells, and the proportion of apoptotic cells. Knockdown of PU.1 significantly decreased the number of viable cells (mean decrease of 18% for shPU.1_1 and 74% for shPU.1_2) (Figure 1D), as well as the colony-forming capacity of primary human AML cells (mean decrease of 27% for shPU.1_1 and 60% for shPU.1_2, compared with shCtrl) (Figure 1E). At the same time, the proportion of apoptotic cells increased on average by 2-fold upon knockdown of PU.1 in primary human AML cells with the more efficient shRNA (shPU.1_2) (Figure 1F).

Taken together, these data show that inhibition of PU.1 decreases cell growth and clonogenic capacity and increases apoptosis, in murine as well as human AML, and thus provide proof of concept for PU.1 inhibition as a possible therapeutic strategy in AML.

Development of small-molecule PU.1 inhibitors abrogating DNA binding by PU.1.

Classical heterocyclic diamidines, typified by furamidine (DB75), are highly selective and potent ligands for AT-rich DNA, a feature responsible for their success as inhibitors of kinetoplast DNA in trypanosomiasis (25, 26). As with other ETS family transcription factors, DNA site recognition by PU.1 involves contact with the major groove at the 5′-GGAA-3′ consensus, as well as an indirect readout of the minor groove backbone, where sequence-dependent geometry confers selectivity among ETS paralogs. PU.1 is distinguishable from other paralogs by the former’s strong preference for AT-rich sequences upstream of the 5′-GGAA-3′ consensus (27). We screened a number of designed AT-targeting heterocyclic diamidines and found that they efficiently competed with PU.1 for DNA binding. DNA footprinting experiments verified that the compounds targeted the flanking sequences found in cognate PU.1 sites, using the λB motif of the murine Igλ2-4 enhancer (5′-ATAAAAGGAAGTG-3′) (28) as a model. On the basis of the general structural features of our initial PU.1 inhibitors (23), we designed and prepared a focused library of larger derivatives for screening using the solution and cellular analytical methods (vide infra). DB2115 and DB2313 were selected as the 2 best compounds (Figure 2A). Both compounds harbor the amidine-benzimidazole-phenyl platform that provides excellent recognition for expanded AT sites. The concept for the units of these compounds is that the amidine cation imparts solubility, interacts favorably with phosphate groups that are close in the narrow DNA minor groove, and can hydrogen bond with acceptor groups on AT base pairs (carbonyl of T, T=O, and N3 of A) at the floor of the minor groove. The benzimidazole-NH points into the minor groove and can also H-bond with AT base pair acceptors. In addition, all 3 groups (amidine, benzimidazole, and phenyl) make excellent van der Waals contacts with the walls of the minor groove and edges of AT base pairs at the floor of the groove. The compounds also have the appropriate curvature to match the DNA minor groove in AT sequences. Together, they make an excellent AT recognition module. Linking 2 of these modules, as in DB2115 and DB2313, provides a molecule for expanded and very strong AT site recognition. A docked structure shows that DB2313, for example, inserted itself precisely and deeply into the DNA minor groove along the AT-track of the λB motif, interacting with the DNA base pair edges at the minor groove floor over almost 10 consecutive base pairs (Figure 2B). Thus DB2313 and DB2115, along with DB1976, which is the best compound from our initial studies of PU.1 inhibition (23), were evaluated in more detail.

We measured the affinity of these heterocyclic diamidines for the λB motif, a typical native DNA-binding site for PU.1, following our biosensor surface plasmon resonance (SPR) procedure (29, 30). Briefly, duplex DNA harboring the λB motif was immobilized on a streptavidin-functionalized sensor chip via a 5′-end–labeled biotin. The affinity of each compound for the λB motif was determined from the concentration-dependent increase in the SPR sensorgrams at steady state, as shown in Figure 2C for DB2313. The specificity of the compounds was probed by substituting the λB motif with a model cognate site (5′-GCCGGAAGTG-3′, termed SC1) that is recognized by ETS1. As shown by DB2313, a very large excess of compound, at concentrations far in excess of those used to saturate the λB motif, failed to bind the SC1 site to detectable levels (gray sensorgram in Figure 2C). With the λB motif as a target, all 3 compounds bound their target DNA tightly, with 1:1 stoichiometry and dissociation constants in the 10^–8 to 10^–9 M range (Figure 2D).

To define the potential of the diamidines to inhibit cognate PU.1 binding, we loaded the immobilized λB site with the purified, recombinant PU.1 ETS domain (Figure 2E). In the absence of diamidine, PU.1 formed a 1:1 complex with the immobilized λB motif with a dissociation constant of 5.4 nM, in agreement with solution fluorescence anisotropy measurements (31). Control titrations with the ETS domain of ETS1, which we have shown to be structurally superimposable with PU.1, showed a 3-log selectivity of the λB site for PU.1 and attested to the specificity of our SPR methodology. Immobilized λB DNA was loaded to greater than 95% capacity and then titrated with graded concentrations of each compound (32). Displacement of PU.1 was detected by the drop in SPR signal as a function of the compound concentration. All 3 compounds efficiently inhibited PU.1 binding, with IC₅₀ values in the 10^–8 to 10^–9 M range (Figure 2F). The rank order of the IC₅₀ values (DB2115 < DB2313 < DB1976) corresponded with the rank order of affinities, suggesting that the 3 compounds inhibited PU.1 binding to the λB motif via the same mechanism.

To gain structural insight into the mechanism by which the diamidines inhibit PU.1 binding to the λB site, we performed a molecular dynamics simulation of the DB2313-λB complex. The λB-bound complex showed the expected binding mode, including the typical water-mediated contacts between the 2 diamidiniumsand the DNA backbone (Supplemental Figure 2) (33). Comparison of the equilibrated DB2313-DNA complex with the PU.1-DNA co-crystal structure (34) showed that the bound compound induced the double helix into a conformation that was incompatible with PU.1 binding. The co-crystal structure of the PU.1-DNA complex shows that extensive perturbations of the DNA groove structure occur in PU.1 binding. Specifically, compression of the minor groove in both width and depth at the AT-rich flanking sequences is required for indirect readout of the DNA backbone by PU.1. Crucially, insertion of DB2313 gives rise to a minor groove profile, in terms of groove width and depth, that deviates strongly from the requirement of PU.1 for binding (Figure 2G). The incompatible effects of DB2313 and PU.1 on DNA structure therefore suggested that denial of PU.1 occupancy through conformational perturbations of the target DNA represents a major component of the efficacy of our compounds as PU.1 inhibitors. We confirmed experimentally that the diamidines induced incompatible DNA conformations for PU.1 binding by probing the drug-bound DNA with DNase I, a minor groove probe that is highly sensitive to local DNA structure (Figure 2H). Although each diamidine induced a unique DNase I footprint, as would be expected from their different structures, all were highly dissimilar to the characteristic footprint produced by PU.1. In summary, computational and experimental evidence revealed mutually exclusive effects on DNA conformation as a fundamental mechanism of PU.1 inhibition by the identified diamidines.

Functional inhibition of PU.1 by heterocyclic diamidines.

Having established the biophysical basis of PU.1 inhibition by DB2313, DB2115, and DB1976, we proceeded to define their cellular properties and potential to inhibit PU.1-dependent gene transactivation. To assess the potential for functional inhibition of PU.1 transactivation by the 3 diamidines, we tested their effects on the expression of a cellular EGFP reporter under the control of a minimal PU.1-dependent promoter harboring a triple tandem copy of the same λB enhancer site used in the SPR screen (Figure 3A). This cell-based system was designed such that ectopic PU.1 and reporter expression can be independently tracked by fluorescence to filter out any background activation of the reporter. In PU.1-negative HEK293 cells, which express several endogenous ETS paralogs including ETS2, GABPA, and members of the ELF subfamily (Human Protein Atlas), transfection of the λB-based reporter was silent in flow cytometry, indicating specificity of the λB enhancer to PU.1 (Figure 3B). All 3 compounds inhibited PU.1-dependent transactivation of the reporter in a dose-dependent manner, with similar IC₅₀ values between 2 and 5 μM (Figure 3C). The relatively high IC₅₀ values should be considered in light of the strong CMV promoter used to drive the expression of ectopic PU.1. To establish the specificity of diamidine action to the inhibition of PU.1, we tested whether the diamidines could inhibit transactivation at an analogous minimal promoter in which the λB enhancer sites were replaced by the SC1 sequence. Although PU.1 can bind the SC1 site, it is not found in native PU.1 target genes (35) and is therefore functionally nonspecific. None of the diamidines significantly inhibited transactivation at the SC1-based enhancer (Figure 3C), consistent with the lack of PU.1 binding to the SC1 site we observed by SPR (Figure 2C). Thus, heterocyclic diamidines appeared to maintain their strong AT selectivity in live cells and specifically inhibited PU.1 activity at functional DNA sites associated with PU.1 target genes.

Small-molecule PU.1 inhibitors decrease cell growth and the clonogenic capacity of AML cells and lead to increased apoptosis.

To determine whether our PU.1 inhibitors had functional effects on AML cells, we treated murine PU.1 URE^–/– AML and human MOLM13 cell lines with compounds at different concentrations and compared them directly with the effects on murine WT bone marrow (BM) cells or human CD34⁺ cord blood cells, respectively. Treatment with compounds led to a profound decrease in the growth of PU.1 URE^–/– AML cells (IC₅₀ for DB2115: 3.4 μM < IC₅₀ for DB2313: 7.1 μM < IC₅₀ for DB1976: 105 μM), while showing little effect on normal hematopoietic cells at similar concentrations (calculated IC₅₀ for DB2115: 192 μM; DB2313: 240 μM; DB1976: 334 μM) (Figure 4A). We made similar observations with human cells, with strong inhibitory effects on MOLM13 cells (PU.1 low), substantially lesser effects on THP1 cells (PU.1 high), and negligible effects on healthy CD34⁺ cells (Figure 4B). We confirmed the inhibitory effect on the proliferation of PU.1 URE^–/– AML and MOLM13 cells during a treatment time course and observed no effect on AML cells with high PU.1 levels (THP1) (Supplemental Figure 3A). To confirm that the drug effects were related to PU.1 levels, we lentivirally expressed PU.1 in MOLM13 cells and observed a rescue of the cytotoxic effects of the compounds with increased cell viability upon treatment with DB1976, DB2115, or DB2313 (Supplemental Figure 3B).

Like our observations with PU.1-directed shRNAs, treatment with pharmacological PU.1 inhibitors led to a 1.6-, 2-, and 3.5-fold increase in apoptotic cells with DB1976, DB2115, and DB2313, respectively, in murine PU.1 URE^–/– AML cells (Figure 4C), and we observed similar effects in human MOLM13 cells (Supplemental Figure 3C).

PU.1 inhibitors also significantly decreased the colony-forming capacity of PU.1 URE^–/– AML and MOLM13 cells, but not of THP1 cells (Figure 4D and Supplemental Figure 3D). To assess the effect of PU.1 inhibitors on the long-term clonogenic capacity (in vitro self-renewal) of PU.1 URE^–/– AML cells, we performed serial replating assays. Strikingly, we found that treatment with DB2313 led to a significant decrease in clonogenicity in the second and third rounds of plating and a complete disruption of clonogenic capacity in the fourth and higher rounds of plating (Figure 4E).

We also explored the effects of the small-molecule PU.1 inhibitors on primary human AML cells and treated 13 AML patients’ samples (Supplemental Table 1) with PU.1 inhibitors. PU.1 inhibitors led to a significant decrease in the number of viable cells (mean decrease of 81% for DB1976, 68% for DB2115, and 72% for DB2313) (Figure 4F) and clonogenic capacity (mean decrease of 36% for DB1976, 45% for DB2115, and 60% for DB23313) compared with vehicle-treated cells (Figure 4G). The apoptotic cell fraction increased on average by 1.5-fold with DB1976, 2.2-fold with DB2115, and 2.5-fold with DB2313 (Figure 4H). The apoptosis-inducing effect was also confirmed in short-term liquid culture for a few samples (Supplemental Figure 3E). Of note, we observed that PU.1 inhibition had an effect on the majority of samples, which harbored various genetic and cytogenetic anomalies.

Taken together, treatment with PU.1 inhibitors leads to decreased cell viability, colony formation, and increased apoptosis in PU.1^lo-induced AML cell lines as well as in a majority of primary AML cell samples from patients.

Inhibitors show on-target PU.1-inhibitory activity in AML cells.

To assess the on-target activity of our inhibitors in AML cells, we measured transcript levels of well-known PU.1 targets in PU.1 URE^–/– AML cells. It has been shown that PU.1 positively regulates CSF1R and JUNB and autoregulates itself, whereas it represses E2F1 (12, 16, 36, 37). In line with this, we found a decrease in Csf1r, Junb, and PU.1 transcript expression and an increase in E2f1 expression upon treatment with DB2115 or DB2313 (Figure 5A, left). As Csf1r is one of the most sensitive PU.1 targets in myeloid cells, we assessed its expression at different time points (1 h, 4 h, 8 h, and 24 h) after drug treatment. Interestingly, Csf1r expression significantly decreased as early as 4 hours after treatment (Figure 5A, right), in line with a direct effect of the drugs on PU.1 transcriptional activity. Furthermore, treatment of BM MNCs isolated from a PU.1-GFP–knockin reporter mouse model (38) led to a decrease in GFP reporter expression after treatment with DB1976, DB2115, or DB2313, further confirming a direct effect of the compounds on PU.1 transactivation (Figure 5B and Supplemental Figure 4A) and consistent with PU.1 positive autoregulation (36). ChIP assays revealed that treatment of AML cells with DB1976, DB2115, or DB2313 decreased PU.1 occupancy on E2f1, Junb, and Csf1r promoters, confirming that the compounds were directly interfering with PU.1 binding to chromatin in vivo (Figure 5C).

To gain insight into the genome-wide transcriptional effects following treatment of AML cells with our inhibitors, we performed gene expression analysis. We found dysregulation of 1,648 transcripts (of 34,472 total) by at least 1.2-fold after DB2313 treatment of PU.1 URE^–/– AML cells, with 867 probe sets upregulated and 781 probe sets downregulated. We found highly significant enrichment of known genes directly downstream of PU.1 (Figure 5D). Interestingly, enrichment of genes regulated by other ETS family transcription factors, such as ETS1, GABPΑ, SPI-B, or FLI-1, was much lower or nonsignificant, suggesting a preferential inhibitory effect of our compounds on PU.1 binding, with selectivity even within the ETS family. We confirmed the dysregulation of some known PU.1 target genes such as Ly96, Clec5a, Cdkn1a, Itgb2, Fcgr3, and Gfi1 by quantitative reverse transcription PCR (qRT-PCR) (Supplemental Figure 4B). Top canonical pathways and biological functions with significant enrichment identified by Ingenuity Pathway Analysis (IPA) included “hematological system development and function,” “cell death and survival,” “cellular development,” and “cellular growth and proliferation” (Supplemental Figure 4C), consistent with known functions of PU.1.

We compared the differentially expressed genes upon PU.1 inhibitor treatment of PU.1 URE^–/– AML cells with publicly available data in which the PU.1-regulatory transcriptional network had been identified by tamoxifen-mediated induction of PU.1 expression in engineered PU.1-null immature hematopoietic cells (PUER cells) (Gene Expression Omnibus [GEO] GSE13125) (39). Of the 1,334 genes dysregulated after PU.1 inhibitor treatment of PU.1 URE^–/– AML cells, 36% (484 cells) overlapped with previously identified canonical PU.1 targets in PUER cells (Figure 5E), which was highly significant (P < 0.0001). Furthermore, comparative pathway analysis revealed a significant inverse correlation between PU.1 URE^–/– AML cells after treatment and PU.1 overexpression in PUER cells, suggesting that inhibitor treatment antagonizes canonical PU.1-regulated pathways (Figure 5F). Gene set enrichment analysis (GSEA) confirmed that DB2313 treatment leads to a change in expression of genes regulated by PU.1 in HSCs from the PU.1 URE^–/– mouse model (12) (Supplemental Figure 4D). Moreover, we found an enrichment in genes regulated by transcription factors of the AP1 family, which are known PU.1 targets (Supplemental Figure 4E).

Furthermore, we performed interactome analysis of 3 independent large series of primary AML patients’ samples including those from The Cancer Genome Atlas (TCGA), the MILE (Microarray Innovations in LEukemia) study, and GEO GSE1159 to identify the PU.1 regulon specifically in AML cells using the ARACNe algorithm (40–42), and then tested the activity of PU.1 using this gene set in the DB2313-induced transcriptional signature using GSEA. In these analyses, the PU.1 “regulon” generated by the ARACNe algorithm was partitioned in a positive and a negative component on the basis of a positive or negative Spearman correlation (P < 0.05) between expression levels of PU.1 and the target gene across all samples of the data set used to generate the network. Strikingly, for each of these patient cohorts, we found that the PU.1 positive regulon (“PU.1 UP,” genes upregulated when PU.1 activity was high) was enriched under vehicle-treated conditions, consistent with a negative effect of DB2313 on PU.1 transcriptional activity in AML cells (Figure 5, G and H, and Supplemental Figure 4, F–I).

In addition, we performed a comparative analysis with publicly available PU.1 ChIP-sequencing (ChIP-seq) data (GEO GSE63317) (20) and found significant enrichment of PU.1 binding in the promoters of genes deregulated after treatment with PU.1 inhibitor (Figure 5I). Up- and downregulated genes were similarly affected, consistent with an interference of our drugs with PU.1-chromatin interaction and independent of downstream transactivating or repressive mechanisms.

We performed PU.1 ChIP-seq experiments and analyzed PU.1-bound sites in the vehicle-treated sample and PU.1 sites in the PU.1 inhibitor–treated (DB2313-treated) sample in PU.1 URE^–/– cells. Comparison of PU.1 ChIP-seq of vehicle versus DB2313 treatment as the background showed a strong enrichment of PU.1 motifs, while the opposite comparison showed enrichment of motifs that are not related to ETS transcription factors (data not shown). This is in line with better PU.1 binding under control conditions. The most common motif in the PU.1 ChIP was AAAGAGGAAGTG, and the respective position weight matrix (PWM) found by Homer was present in 41% of peaks in the vehicle sample versus only 16% of the peaks in the DB2313-treated sample, which is again in line with PU.1 inhibitor treatment causing decreased binding of PU.1 to its canonical target sites. When we looked specifically at the peaks with a PU.1 motif in the vehicle-treated samples and subtracted peaks with a PU.1 motif in the DB2313-treated samples, we arrived at 268 bona fide PU.1 sites that were “lost” upon drug treatment. A complete list of these peaks/sites is provided in Supplemental Table 2. Examples of peaks from this analysis are shown in Supplemental Figure 5. Overall, the results from the ChIP-seq analyses are consistent with a genome-wide inhibitory effect of DB2313 on PU.1 binding to its canonical sites in AML cells.

PU.1 inhibitors have limited effects on normal hematopoiesis.

In order to determine the effect of our PU.1 inhibitors on normal hematopoietic differentiation, we sorted immature Lin^–Sca1⁺c-Kit⁺ (LSK) cells from WT mice and treated them with either DB1976, DB2115, or DB2313. After treatment with PU.1 inhibitors, the total numbers of formed colonies were only slightly reduced, however, we saw a substantial reduction of the more mature myelomonocytic colony types (CFU-granulomonocyte [CFU-GM], CFU-granulocyte [CFU-G], and CFU-monocyte [CFU-M]) and a relative increase in the number of burst or CFU-erythroid (B/CFU-E), and small immature cell colonies (Figure 6A), consistent with the known important role of PU.1 in myelomonocytic differentiation. Of note, the initial compound DB1976 appeared to show toxicity on LSK cells, with a majority of colonies unidentifiable and a majority of dead cells, contrary to what we observed with the 2 newer compounds (DB2115 and DB2313), which we therefore prioritized for further investigation. Overall, the numbers of viable cells per colony were reduced upon treatment with PU.1 inhibitor (Supplemental Figure 6A), again consistent with an effect of PU.1 inhibition on more mature, differentiating cells. Consistently, cytomorphologic analysis revealed almost no macrophages, significantly fewer mature granulocytes, and an increase in immature cells, characterized by a higher nucleocytoplasmic ratio and basophilic cytoplasm after treatment (Figure 6B). This was confirmed by flow cytometric analysis, which revealed a reduction in mature monocytes (CD11b⁺Gr1^–) and granulocytes (CD11b⁺Gr1⁺) and a slight increase in immature granulocytes (CD11b^–Gr1⁺) and CD41⁺ cells, but no significant difference in erythrocyte (Ter119⁺CD41^–) generation (Figure 6C and Supplemental Figure 6B). To further explore the granulocyte population and pinpoint the exact stage of maturation at which the small molecules impact differentiation, we used a combination of markers described by Guibal et al. (43) and further separated the cell populations into the myeloblast/promyelocyte-metamyelocyte/mature stages (Supplemental Figure 6C). After treatment, we observed a slight increase in the proportion of cells at the myeloblast/promyelocyte stage (stage 1), a strong increase of metamyelocytes (stage 2), and a decrease in more mature cells (stage 3) (Supplemental Figure 6D), indicating that our compounds primarily inhibit later stages of granulocytic differentiation.

As we had observed a reduction in mature myelomonocytic colonies upon treatment with PU.1 inhibitors, but persistence or even slight increases in more immature cells (CFU-granulocyte/erythrocyte/macrophage/megakaryocyte [GEMM] and immature colonies), we wanted to test whether these immature cells were still functional after drug removal. We focused on the compound with the lowest IC₅₀ and the least effect on WT cells (DB2313). Interestingly, the production of mature granulocytes and monocytes increased significantly by 4-fold and 22-fold, respectively, after removing the treatment in the second plating, showing that the effects of PU.1 inhibition on granulocyte/monocyte (G/M) generation are reversible (Figure 6, D and E). In addition, we performed serial replating assays with continuous treatment with PU.1 inhibitors and found no significant differences in terms of serial replating capacity, indicating that treatment with PU.1 inhibitors does not significantly affect the long-term colony-forming potential or in vitro self-renewal capacity of immature hematopoietic stem and progenitor cells (Figure 6F).

We also explored the effect of the inhibitors on lymphoid cells. We treated B cells in liquid culture for 3 days and found that the compounds had no effect on the apoptotic fraction (Supplemental Figure 6G). To assess T cells, we performed culturing on OP9-Delta1 cells and found that compound treatment led to slightly delayed, but not completely blocked, T cell differentiation (Supplemental Figure 6, E and F). Finally, we assessed PU.1 inhibitor–treated BM cells in congenic transplantation assays and found no changes in T cell numbers and only modestly decreased B cell numbers after 20 weeks (Supplemental Figure 6H).

Taken together, our results indicate that treatment with our PU.1 inhibitors leads to effects on normal hematopoiesis that are consistent with fundamental roles of PU.1 function during hematopoiesis in myelomonocytic lineages, with limited effects on T and B cell lymphopoiesis. The effects on myeloid cells are reversible upon treatment discontinuation and seem to primarily affect more mature cells. However, in future studies, it will be of interest to perform longer-term in vivo treatment with further optimized compounds (once available) for a more comprehensive and definitive evaluation of the more subtle effects on normal hematopoiesis.

Treatment with PU.1 inhibitors decreases leukemia progression in vivo.

To assess the effect of PU.1 inhibitors on the growth of AML cells in vivo, we treated PU.1 URE^–/– AML cells with DB2313 for 2 days and injected 2 × 10⁵ viable cells into sublethally irradiated recipient mice (Figure 7A). Mice that received vehicle-treated AML cells succumbed to leukemia, with a median latency of 44 days, whereas mice that received DB2313-treated AML cells survived significantly longer (P = 0.0036), with a median latency of 68 days (Figure 7B). Assessment of tumor burden 6 weeks after transplantation showed a decrease in hepatoplenomegaly after treatment; mean splenic weight was 410 mg for vehicle-treated mice and 243 mg for DB2313-treated mice (Figure 7C); mean liver weight was 2,347 mg for vehicle-treated mice and 1,548 mg for DB2313-treated mice (Figure 7D). AML cell engraftment in the BM was significantly decreased after treatment with DB2313, with a mean chimerism of 55% for the vehicle-treated group and 33% for the DB2313-treated group (Figure 7E and Supplemental Figure 7A). Histological analyses revealed severe blast infiltration, with disruption of the splenic architecture and complete loss of the red pulp in the vehicle-treated group, and significantly less pronounced effects in the DB2313-treated group. Likewise, infiltration of the liver by leukemic blasts was substantially reduced in the DB2313-treated group compared with that detected in the vehicle-treated group (Figure 7F). When we performed the same experiment with human MOLM13 AML cells, we observed a similarly significant decrease in leukemic burden in the BM after treatment, with a mean chimerism of 35% for the vehicle-treated group and 11% for the DB2313-treated group (Figure 7, A, G, and H).

Further, we performed in vivo treatment by i.p. injections of DB2313 following transplantation of PU.1 URE^–/– AML cells (Figure 7I). The effect on AML cells in the BM was modest, with a mean chimerism of 47% for the vehicle-treated group versus 37% for the DB2313-treated group, which was likely due to the low concentration of this prototype compound in the serum (Supplemental Figure 7, B and C). However, we observed decreased splenomegaly (mean weight of 257 mg for vehicle-treated mice and 174 mg for DB2313-treated mice) and hepatomegaly (mean weight of 1,604 mg for vehicle-treated mice and 1,203 mg for DB2313-treated mice) (Figure 7, J and K), reduced leukemic infiltration in the spleen (Figure 7L and Supplemental Figure 7D), and no detectable leukemic infiltration in the liver after DB2313 treatment (Figure 7L).

Taken together, although our compounds represent tool compounds that require further optimization, these data provide proof of concept for the antileukemic effects of PU.1 inhibitors in vivo, including decreased tumor burden and increased survival.

Cells.

THP1, MOLM13, BaF3, Kasumi-1, HEK293, and 293T cells were purchased from ATCC. MOLM13 and Kasumi-1 cells were cultured in RPMI 1640 medium supplemented with 20% heat-inactivated FBS and 1% penicillin/streptomycin (P/S). THP1, BaF3, and HEK293 cells were cultured in RPMI 1640 medium completed with 10% FBS and 1% P/S. BaF3 cells were supplemented with 10 ng/ml IL-3. 293T cells were maintained in DMEM supplemented with 10% FBS and 1% P/S. PU.1 URE^–/– and PU.1 URE^+/–Msh2^–/– leukemia cells (17, 18) were cultured in M5300 medium (STEMCELL Technologies) supplemented with 10% FBS, 15% WEHI supernatant, 15% BHK supernatant, and 1% P/S. PU.1 URE^–/– and PU.1 URE^+/–Msh2^–/– cell lines were established from leukemic URE^–/– and URE^+/–Msh2^–/– mice, respectively.

Mouse cells from peripheral blood (PB), BM, spleen, or liver were isolated, treated with ACK buffer (pH 7.4) to lyse red blood cells, and cultured or stained for FACS analysis. Cells were maintained in Iscove’s Modified Dulbecco’s Media (IMDM) (Gibco, Thermo Fisher Scientific) supplemented with 20% FBS, 20 ng/ml recombinant mouse SCF (rmSCF) (Miltenyi Biotec), 10 ng/ml rmIL-3 (Gemini), 10 ng/ml rmIL-6 (Gemini), 20 ng/ml rmTPO (Miltenyi Biotec), and 1% P/S.

Human MNCs from BM or PB obtained from healthy or leukemic patients were isolated by density ultracentrifugation using Ficoll-Hypaque (GE Healthcare) and cultured in IMDM medium supplemented with 2% FBS, cytokines (10 ng/ml rhIL-3 [Preprotech], 25 ng/ml recombinant human IL-6 [rhIL-6], 50 ng/ml rhSCF, 50 ng/ml rhFLT3 ligand, and 100 ng/ml rhTPO [all from Gemini]), 40 μg/ml hLDL, and 1% P/S. All cell lines and primary cells were maintained in an incubator at 37°C and 5% CO₂.

Small molecules.

The syntheses of DB1976 (23), DB2115 (48), and DB2457 (49) have been previously reported. The synthesis of DB2313 and the experimental details, along with characterization data, are provided in the supplemental materials.

After synthesis, DB1976, DB2115, DB2313, and DB2457 were dissolved in sterile water and stored at –20°C.

Mice.

PU.1 URE^–/–, PU.1 URE^+/–Msh2^–/–, and PU.1 heterozygous mice with knockin of a GFP reporter into exon 1 of PU.1 have been previously described (17, 18, 38). Nonobese, diabetic severe combined immunodeficiency IL-2 receptor γ–null (NSG) mice were housed in the animal facility.

Patients’ characteristics.

See the Supplemental Methods and Supplemental Table 1.

Lentiviral vector and transduction.

shRNA sequences for knockdown of PU.1 have been previously described and validated (shPU.1_1 [ref. 50], and shPU.1_2 [ref. 51]); shPU.1_3 and shPU.1_4 were newly designed (shPU.1_3: 5′-AGATGCACGTCCTCGATAC-3′; shPU.1_4: 5′-TCCAGTTCTCGTCCAAGCACAA-3′). These sequences recognize both mouse and human sequences. Control (shCtrl) or shPU.1 sequences were cloned under an H1 promoter and inserted into the lentiviral vector pRRLsin-PGK-eGFP-WPRE (provided by Hana Raslova, Institut Gustave Roussy, Villejuif, France). PU.1 overexpression was performed using the lentiviral vector PCAD-PU.1-IRES-GFP or GFP alone as previously described (18). All viruses were produced using the 293T cell line. Cells were transduced by spin infection at 1,000 g for 60 minutes at 37°C in the presence of Polybrene. Seventy-two hours after transduction, cultures were sorted for GFP⁺ cells and used for further studies.

Molecular dynamics simulations.

See the Supplemental Methods.

Biosensor-SPR assays for compound binding affinity and PU.1-DNA complex inhibition by compound.

See the Supplemental Methods.

DNA footprinting.

See the Supplemental Methods.

Cell proliferation assays.

To evaluate the effect of the small molecules on cell viability, cells were cultured in the presence of vehicle (water) or different compounds. Cells were plated at a density between 0.5 × 10⁵ and 2 × 10⁵ in 100 μl or 500 μl of media and manually counted every 24 hours for 4 days using a trypan blue exclusion assay. For IC₅₀ determination, cell viability was first assessed using a resazurin cell viability assay (Cell Titer Blue; Promega) after 48 hours of culture and confirmed by manual counting using a trypan blue exclusion assay. If not further specified, drugs for the treatment of mouse cells were used at the following concentrations: 25 μM for DB1976, 700 nM for DB2115, and 330 nM for DB2313. For the treatment of human cells, the same concentrations were used, with the exception of DB2313, which was used at a concentration of 660 nM.

Colony formation assays and serial replating assays.

For human cell line colony assays, 1,000 or 3,000 cells were plated per milliliter of methylcellulose in HSC002SF or HSC003 methylcellulose (R&D Systems). For mouse cell colony assays, 1,000 cells per milliliter of methylcellulose were plated in HSC007 methylcellulose (R&D Systems) and scored after 7 days. For human primary AML cell colony assays, 150,000 mononuclear cells were plated per milliliter of methylcellulose in HSC003 media and the colonies scored after 12 to 14 days. For LSK cells, 200 cells per 500 μl methylcellulose were plated in HSC007. Vehicle or inhibitors were added to the methylcellulose at the following concentrations: 25 μM for DB1976, 700 nM for DB2115, and 330 nM for DB2313 for murine cells and 660 nM for DB2313 for human cells. For serial replating of cell lines, the same numbers of cells initially plated were replated. For serial replating of primary LSK cells, 5,000 cells per 500 μl methylcellulose were replated.

Lymphoid cell culture.

See the Supplemental Methods.

Flow cytometric analysis and cell sorting.

See the Supplemental Methods.

Cytomorphology.

See the Supplemental Methods.

Western blot analysis.

See the Supplemental Methods.

Inhibition of the PU.1-DNA complex in live cells (reporter assay).

The functional inhibition of PU.1 transactivation by heterocyclic diamidines in live cells was measured using a fluorescent EGFP reporter. For this assay, a PU.1 expression plasmid was cloned by inserting a fragment encoding full-length PU.1 fused to an infrared RFP (iRFP) (52) reporter between the NheI/BamHI sites of pcDNA3.1(+). The fusion was linked by a sequence encoding a self-cleaving 2A peptide (53). Cultured HEK293 cells, which do not express PU.1, were transfected with the PU.1 expression plasmid for 24 hours. Cells were then retransfected, in the absence or presence of compounds, with an EGFP-based reporter under the control of a minimal promoter that was downstream from a synthetic enhancer harboring 5 tandem repeats of the λB motif (23). After an additional 24 hours, cells were trypsinized and counted by flow cytometry using an Accuri C6 instrument (BD) on 2 spectrally isolated channels for EGFP (488/511 nm) and iRFP (640/>670 nm).

Molecular modeling and docking.

See the Supplemental Methods.

ChIP.

PU.1 URE^+/–Msh2^–/– cells (10 × 10⁶) were treated with vehicle, 50 μM DB1976, 1.4 μM DB2115, or 660 nM DB2313 for 8 hours. ChIP experiments were performed as previously described (54, 55), except that Dynabeads Protein A (Thermo Fisher Scientific) were used for the pulldown. Briefly, cells were cross-linked using a 1% formaldehyde solution for 15 minutes at room temperature. Samples were sonicated for 10 minutes to generate fragments below 1 kb. For immunoprecipitation, 5 μg PU.1 antibody (Santa Cruz Biotechnology; sc-352) or normal rabbit IgG (Santa Cruz Biotechnology; sc-3888) was used. DNA fragments were purified using a QIAquick PCR Purification Kit (QIAGEN) and subjected to qPCR amplification. The percentage of input was calculated using the comparative cycle threshold method. The primers used are listed in Supplemental Table 3.

qRT-PCR and microarray experiments.

RNA was extracted using an RNeasy Mini or Micro Kit (QIAGEN) and, for qRT-PCR, reverse transcription of RNA was performed using an iScript cDNA Synthesis Kit (Bio-Rad). qPCR reactions were performed on a ViiA7 instrument using SYBR Green PCR Master Mix (both from Life Technologies, Thermo Fisher Scientific). Expression levels of genes were normalized to Gapdh. The primers used are listed in Supplemental Table 3. For microarrays, RNA was isolated from PU.1 URE^–/– AML cells after 24 hours of treatment with vehicle or DB2313 (330 nM). After checking RNA quality using a 2100 Bioanalyzer (Agilent Technologies), RNA was labeled with the Affymetrix GeneChip WT Terminal Labeling Kit, hybridized to Affymetrix 2.0 Mouse GeneST microarrays, and scanned by an Affymetrix GeneChip Scanner 3000 7G system using standard protocols. Microarray data are available in the NCBI’s GEO database (GEO GSE77651). Raw data were normalized using Expression Console software (Affymetrix), and differentially expressed genes were determined using a fold-change of 1.2 and a P value of 0.1 after analysis with Transcriptome Analysis Console software (Affymetrix). Pathway analyses were performed using IPA (QIAGEN). A –log (P value) cutoff of 1.3 was used as the significance threshold. GSEA was performed with gene sets from the Molecular Signatures Database (56, 57) and from the PU.1 URE^–/– HSC signature database (12).

Integrative analysis with published data sets.

For comparative analysis, we used the NCBI GEO data set GSE13125 (39), derived from a PU.1 overexpression system with tamoxifen-inducible “PUER” cells. Comparative analysis between differentially expressed genes in this data set and in PU.1 URE^–/– AML cells after treatment was performed using IPA. For GSEA, we used the NCBI GEO data set GSE5654, derived from PU.1 URE^–/– HSC (12). To identify genes directly regulated by PU.1, we used the PU.1 ChIP-seq data obtained from GSE63317 (GSM1545739) (20). Among the genes dysregulated in PU.1 URE^–/– after DB2313 treatment, we identified those with PU.1 occupancy in their promoters, defined as being within 1 kb upstream of the transcription start site (TSS). To determine the significance of correlation of differential gene expression upon PU.1 activation with the presence of PU.1 binding in the gene promoter, we performed a Fisher’s exact test using UCSC Genes (mm9) as a reference (https://genome.ucsc.edu/cgi-bin/hgGateway?db=mm9).

Interactome analyses (ARACNe Networks).

See the Supplemental Methods.

ChIP-seq analysis.

See the Supplemental Methods.

BM transplantation.

See the Supplemental Methods.

Determination of serum compound concentrations.

See the Supplemental Methods.

In vivo leukemia models.

NSG mice were sublethally irradiated (2.0 Gy) and transplanted via retro-orbital vein injection with PU.1 URE^–/– AML cells or via tail vein injection with MOLM13 cells. For pretreatment experiments, AML cells were treated in vitro with vehicle or DB2313, viable cells were counted after 2 days of treatment, and 2 × 10⁵ cells (PU.1 URE^–/– AML) or 1 × 10⁶ cells (MOLM13) per mouse were transplanted. Mice were monitored for signs of leukemia. For survival experiments, live BM aspiration was performed in mice 6 weeks after transplantation of PU.1 URE^–/– cells. Otherwise, mice were sacrificed at 6 weeks (PU.1 URE^–/– AML group) or 3 weeks (MOLM13 group). For i.p. treatment with PU.1 URE^–/– AML cells, mice were transplanted with 1 × 10⁶ cells per mouse. Mice were divided into vehicle-treated and DB2313-treated (17 mg/kg) groups. Treatment started 3 days after transplantation and continued for 3 weeks, with 3 injections per week. Mice were sacrificed 3–4 weeks after transplantation.

After sacrifice, BM, spleens, and livers were harvested from the mice, and engraftment was assessed by FACS using CD45.1, CD43, and c-Kit antibodies for PU.1 URE^–/– AML cells, and CD45.1 and human CD45 antibodies for MOLM13 cells.

Statistics.

Data are presented as the mean ± SD of independent experiments, unless otherwise specified. Statistical analyses were performed with GraphPad Prism 7.0 (GraphPad Software) using 1-way ANOVA with Tukey’s post-test correction for multigroup comparisons or a 2-tailed Student’s t test for 2-group comparisons, unless otherwise specified in the figure legends. A P value of less than 0.05 was considered significant. For survival analyses, a log-rank test was used to determine statistical significance.

Study approval.

Human samples were collected after informed consent was obtained from patients and upon approval of the IRB of Albert Einstein College of Medicine (protocol 2008-842). All mouse experiments were approved by the IACUC of Albert Einstein College of Medicine (protocol 2013-1202).

Version 1. 10/30/2017

Electronic publication

Version 2. 12/01/2017

Print issue publication

Version 2. 11/20/2017

Fixed author's name in authorship note.