Nascent transcript and single-cell RNA-seq analysis defines the mechanism of action of the LSD1 inhibitor INCB059872 in myeloid leukemia

Abstract

Drugs targeting chromatin-modifying enzymes have entered clinical trials for myeloid malignancies, including INCB059872, a selective irreversible inhibitor of Lysine-Specific Demethylase 1 (LSD1). While initial studies of LSD1 inhibitors suggested these compounds may be used to induce differentiation of acute myeloid leukemia (AML), the mechanisms underlying this effect and dose-limiting toxicities are not well understood. Here, we used precision nuclear run-on sequencing (PRO-seq) and ChIP-seq in AML cell lines to probe for the earliest regulatory events associated with INCB059872 treatment. The changes in nascent transcription could be traced back to a loss of CoREST activity and activation of GFI1-regulated genes. INCB059872 is in phase I clinical trials, and we evaluated a pre-treatment bone marrow sample of a patient who showed a clinical response to INCB059872 while being treated with azacitidine. We used single-cell RNA-sequencing (scRNA-seq) to show that INCB059872 caused a shift in gene expression that was again associated with GFI1/GFI1B regulation. Finally, we treated mice with INCB059872 and performed scRNA-seq of lineage-negative bone marrow cells, which showed that INCB059872 triggered accumulation of megakaryocyte early progenitor cells with gene expression hallmarks of stem cells. Accumulation of these stem/progenitor cells may contribute to the thrombocytopenia observed in patients treated with LSD1 inhibitors.

1. Introduction

The first drugs that affected histone modifying enzymes were the histone deacetylase inhibitors (HDACi) that were first identified based on their ability to trigger cellular differentiation of myeloid leukemia cells (Richon et al., 1996). Recognition that compounds such as SAHA inhibited the deacetylases to induce apoptosis in AML cells (Tabe et al., 2007) led to the possibility that inhibiting other histone modifying enzymes could have therapeutic potential in AML. As the genetics of AML were defined over the past 25 years, it is now clear that many AML are initiated by chromosomal translocations that affect DNA-binding transcription factors that recruit histone deacetylases and acetyltransferases, as well as histone methyltransferases, demethylases, and enzymes that control histone ubiquitination (Yang et al., 2017). In addition to transcription factors, components of the trithorax-polycomb axis are frequent targets of mutation, with the H3K4 methyltransferase MLL1 being a frequent target for translocation. Polycomb repression complex 2 component EZH2 that mediates H3K27me3 is often inactivated (Cao et al., 2002; Sashida et al., 2014), while the polycomb deubiquitinase (PR-DUB) component ASXL1, and less frequently ASXL2, are mutated to affect H2AK119ub (Nagase et al., 2018). Drugs targeting some of these histone-modifying enzymes have shown potential in clinical trials (San-Miguel et al., 2016; Stein et al., 2018). Given the prevalence of deregulation of these enzymes in hematological malignancies (Hu and Shilatifard, 2016), AML is a logical place in which to explore the therapeutic efficacy of these drugs.

Lysine-Specific Demethylase 1 (LSD1, also known as KDM1A) was the first lysine demethylating enzyme identified and acts as a flavin-dependent monoamine oxidase to remove mono- and di-methyl groups from histone H3K4 and H3K9 (Shi et al., 2004; Metzger et al., 2005). Prior to discovery of its enzymatic activity, it was first purified as part of the NuRD (nucleosome remodeling and deacetylase) complex, which contains histone deacetylases HDAC1/2, helicases CHD3/4, and other factors including MTA2, MBD3, RBBP4, and RBBP7 (Tong et al., 1998; Wang et al., 2009). This complex decommissions enhancers during the differentiation process of embryonic stem cells (Whyte et al., 2012). LSD1 was also discovered in the CoREST transcription repression complex, which contains HDAC1/2, the corepressor RCOR1, and ZNF217, PHF21A, and HMG20B (You et al., 2001; Maiques-Diaz and Somervaille, 2016). Additional transcriptional regulators, such as CtBP1, colocalize with variant forms of the CoREST complex (Wang et al., 2007; Mulligan et al., 2011). However, it is the tissue-specific DNA-binding factors that recruit CoREST that determine the genes that are silenced (Laugesen and Helin, 2014). Inhibitors of LSD1 are expected to modulate gene expression by de-repressing genes regulated by these complexes. The lysine demethylase inhibitors are showing efficacy, yet in many instances the molecular mechanism of action is still poorly defined and the rate-limiting toxicities such as thrombocytopenia are also unexplained at the molecular or gene expression level.

In hematopoietic cells, LSD1 is recruited by the transcriptional repressors Growth Factor Independent 1/1B (GFI1 and GFI1B) to control lineage commitment (Saleque et al., 2007). Gfi1 knockout mice live to adulthood but accumulate immature myelomonocytic cells and display extreme neutropenia (Karsunky et al., 2002) and Gfi1 mutations in humans cause severe congenital neutropenia (Person et al., 2003). Gfi1b deletion in mice was embryonic lethal, and fetal liver cells from these mice were unable to complete erythropoietic or megakaryocytic differentiation (Saleque et al., 2002). Conditional deletion of Gfi1b from adult mouse bone marrow was lethal within 3 weeks, as these mice had severely reduced hemoglobin levels and platelet counts (Foudi et al., 2014). Consistent with these mouse data, mutations in GFI1B were detected in familial platelet disorders (Stevenson et al., 2014; Monteferrario et al., 2014). GFI1 and GFI1B interact with LSD1 through an N-terminal SNAG domain that is highly conserved between the two family members, and mutations within the SNAG domain of Gfi1 mimics genetic deletion of Gfi1 (Grimes et al., 1996; Fiolka et al., 2006). Apart from LSD1, there is no other known protein that associates with the SNAG domain of GFI1/1B (Moroy et al., 2015), suggesting that LSD1 is essential for Gfi1 and Gfi1b to repress transcription.

Genetic deletion of Kdm1a in hematopoietic stem cells of mice caused multi-lineage cytopenia with a general defect in maturation across all cell types (Kerenyi et al., 2013). Moreover, the long-term stem cells were also affected, indicating that LSD1 was required for hematopoietic stem cell self-renewal. ChIP-seq studies have revealed that the majority of LSD1 binding sites are at enhancers (Whyte et al., 2012; Maiques-Diaz et al., 2018), so a role in cell fate determination was expected. A conditional mouse model of LSD1 knockdown exhibited severely reduced platelet production, that was similar to that found in the Gfi1b knockout (Sprussel et al., 2012). Though shRNAs directed at LSD1 expanded progenitor cells of multiple blood lineages (Sprussel et al., 2012), the stage of megakaryopoiesis affected by LSD1 inhibition has not been determined. These genetic studies indicated that inhibition of LSD1 could have therapeutic potential in acute leukemia, but also highlighted potential toxicities.

INCB059872 is a selective, irreversible LSD1 inhibitor that has recently entered the clinic in early clinical trials. It is potent (18 nM) and highly selective, but its biological effects are yet to be described. Early studies showed that decreased LSD1 activity could synergize with all-trans retinoic acid to induce differentiation of myeloid leukemia cells (Schenk et al., 2012). Because leukemias driven by MLL translocations were especially sensitive to LSD1 knockdown and inhibition (Harris et al., 2012; Feng et al., 2016), we focused on MLL-rearranged myeloid leukemia cells for deep characterization of the transcriptional effects of INCB059872. In this study, we have combined analysis of nascent transcription with single-cell RNA-seq to capture a better picture of the differential effects of INCB059872 on leukemic and non-leukemic blood lineages. Single-cell expression profiling provides more detailed insights into the differentiation trajectories of hematopoiesis (Hamey et al., 2017; Dahlin et al., 2018) and allows for a more detailed phenotypic analysis of stem and progenitor cells. Our studies reveal that INCB059872 causes the rapid expression of GFI1/GFI1B repressed genes and causes the accumulation of megakaryocyte-like stem and progenitor cells.

2. Materials and Methods

Cell lines

THP-1 (Cat# TIB-202, RRID:CVCL_0006) and MV-4–11 (Cat# HTB-189, RRID:CVCL_0064) cell lines were obtained from ATCC. THP-1 were cultured in RPMI containing 10% FetalPlex (Gemini) and 50 μM 2-mercaptoethanol (Gibco). MV-4–11 were cultured in IMDM containing 10% FetalPlex. S2 cells (RRID:CVCL_Z232, provided by Emily Hodges lab) were cultured in Schneider’s Drosophila Medium containing 10% FetalPlex.

Mice

All animal experiments were conducted in accordance with guidelines approved by the IACUC at Vanderbilt University Medical Center. For the in vivo studies, female 6–8 week old C57BL/6J mice (The Jackson Laboratory, IMSR Cat# JAX:000664, RRID:IMSR_JAX:000664) were treated daily with 10mg/kg INCB059872 via oral gavage for up to six days. INCB059872 was dissolved in N,N- dimethylacetamide (DMAC; Sigma) and diluted in 5% methylcellulose (Sigma). For peripheral blood analysis, blood was collected via tail vein into an EDTA microtainer and analyzed for complete blood counts using a Hemavet (Drew Scientific).

Patient sample

Experiments were conducted on a primary patient sample, which was provided by the Vanderbilt-Ingram Cancer Center Hematopoietic Malignancies Repository and in accordance with the tenets of the Declaration of Helsinki and approved by the Vanderbilt University Medical Center Institutional Review Board.

Differentiation analyses

For flow cytometry, THP-1 or MV-4–11 cells were stained with BD Pharmingen anti-CD11b (clone ICRF44) conjugated to PE (cat# 301306, RRID:AB_314158) or APC (cat# 301310, RRID:AB_314162) using 18 μl antibody per 10⁶ cells, following manufacturer’s staining protocol. Data were analyzed using FlowJo software (RRID:SCR_008520). For Wright-Giemsa staining of cytospins, Fischer Hema 3 Stat Pack (cat# 123–869) was used according to manufacturer’s instructions.

Co-immunoprecipitation

HEK293T cells were transfected with CMV-GFI1 or empty CMV vector then treated with DMSO or 250 nM INCB059872. After 48 hours, cells were lysed in NETN buffer (100 mM NaCl, 2 mM Tris pH 8.0, 0.5 mM EDTA, 0.5% NP-40). Each lysate was divided into two IP samples. For IgG control IPs, 3 μg rabbit IgG (Invitrogen, cat# 02–6102, RRID:AB_2532938) was added to each sample. For LSD1 IPs, 3 μg anti-LSD1 (Abcam, ab17721, RRID:AB_443964) was added to each sample. Lysates were incubated with antibody for 3 hours while rotating at 4°C. Protein A (Invitrogen cat# 10001D) and Protein G (Invitrogen cat# 10003D) Dynabeads (15 μl of each) were added to each IP sample, and tubes were rotated at 4°C for 1 hour. Beads were washed three times with 0.5X NETN buffer by rotating 4°C for 5 min. Beads were resuspended in SDS loading buffer (40% glycerol, 240 mM Tris pH 6.8, 8% SDS, 0.4 mg/mL bromophenol blue, 5% beta-mercaptoethanol) and boiled for 10 min. Samples were run on 8% SDS-PAGE gel and transferred onto 0.45 μm PVDF for Western blotting.

RNA-seq and analysis

For each sample, 0.75 million cells were resuspended in 1 mL TRIzol, and RNA was isolated according to manufacturer’s instructions. Samples were submitted to Vanderbilt Technologies for Advanced Genomics (VANTAGE) for polyA-enriched library preparation and sequencing on Illumina NextSeq500. Reads were aligned to hg19 genome using TopHat (RRID:SCR_013035). Cuffdiff (Cufflinks software suite, RRID:SCR_001647) was used to calculate differential gene expression (Trapnell et al., 2010). Gene Set Enrichment Analysis software v3.0 (RRID:SCR_003199) was used to identify gene signatures associated with expression changes (Subramanian et al., 2005).

PRO-seq and analysis

Nuclear run-on reactions and library preparation were performed as described (Mahat et al., 2016). Samples were submitted to VANTAGE for sequencing on Illumina NextSeq500. After adapter trimming with Trimmomatic (RRID:SCR_011848) and removal of low-quality sequences, reads were reverse-complemented. Reads were aligned to hg19 genome with Bowtie2 (RRID:SCR_005476), and reads with mapping quality <10 were removed. NRSA package (Wang et al., 2018) was used to calculate polymerase density based on DESeq-normalized read counts. Motif analysis of enhancers was done using HOMER software (Heinz et al., 2010; RRID:SCR_010881).

ChIP-seq and analysis

Immunoprecipitation and library preparation are described in Supplemental Methods. Libraries were submitted to VANTAGE for sequencing on Illumina NextSeq500. Trimmomatic was used to remove adapters and low-quality sequences. Reads were aligned to a combined genome file containing both Human hg19 and Drosophila dm3 using Bowtie2. Duplicate reads and reads with mapping quality <10 were removed using samtools. Reads aligned to dm3 were separated into their own files, and read counts of these files were used to calculate normalization factors for each sample. MACS2 (RRID:SCR_013291) was used to call peaks with q-value cutoff of 0.001. Count tables were generated using R program DiffBind (Stark and Brown, 2011; RRID:SCR_012918). Normalization factors were used with DESeq2 (RRID:SCR_015687) to calculate changes in peak size.

Single-cell RNA-seq- mouse bone marrow

Harvested bone marrow was stained with BD Pharmingen Biotin Mouse Lineage Panel (cat# 559971, RRID:AB_10053179) and streptavidin-PacBlue (Invitrogen, cat# S11222). Zombie-NIR Fixable Viability Dye (BioLegend cat# 423105) was added to each sample at 1:1000 dilution before sorting alive lineage-negative population (PacBlue-/NIR-). Sorted cells were submitted to VANTAGE for single-cell RNA-seq library prep and processing with 10X Genomics Chromium Controller. Samples were demultiplexed and single-cell gene expression matrices were generated using Cell Ranger software (RRID:SCR_017344). Seurat R package (Butler et al., 2018; RRID:SCR_016341) was used for data analysis (see Supplementary file for R code).

Single-cell RNA-seq- human AML

Cells were thawed and cultured in IMDM media containing 10% FBS (Stasis), 2 mM L-glutamine, 10 ng/mL IL-3, 10 ng/mL Flt3, 10 ng/mL SCF, 10 ng/mL TPO, 5 ng/mL IL-6, 10 μM 2-mercaptoethanol, and 4 μg/mL LDL. After 48hr in culture with drugs, apoptotic cells were removed from the samples using Miltenyi Annexin V Microbead Kit (cat# 130–090-201) with MACS LS columns (cat# 130–042-401). The Annexin V-negative cells were submitted to VANTAGE for single-cell RNA-seq as described above.

Flow cytometry- mouse bone marrow

Mouse BM cells were obtained from femurs and tibias. Red blood cells were removed by incubation with Qiagen Buffer EL (cat# 79217) according to manufacturer’s protocol. Cells were resuspended in PBS with 0.5% BSA at a density of approximately 10⁷ cells/mL before proceeding to staining. For lineage staining, BD Pharmingen Biotin Mouse Lineage Panel (cat# 559971, RRID:AB_10053179) was used according to manufacturer’s instructions and followed by staining with 0.5 μl streptavidin-Pacific Blue (Invitrogen, cat# S11222) per 10⁶ cells. Additional antibodies are listed in Supplemental Table S3. Data were analyzed using FlowJo (Becton Dickinson and Company, Ashland, OR, USA; RRID:SCR_008520) and GraphPad Prism (GraphPad Software, La Jolla, CA, USA; RRID:SCR_002798).

RNA-seq- mouse megakaryocyte progenitors

Lin- CD41+ CD200r3- bone marrow cells from 6 mice (2 replicates of 3 conditions) were flow-sorted directly into Buffer RLT from Qiagen RNeasy Mini Kit (cat# 74104), and RNA was isolated according to manufacturer’s protocol. Samples were submitted to HudsonAlpha Genomic Services Laboratory for library preparation with Nugen Ovation RNA-Seq System V2 kit (cat# 7102–08) and sequencing on Illumina NovaSeq6000. Data analysis was performed as described above.

Data availability

Genomics datasets generated as part of this study have been deposited in GEO under the following accession numbers: GSE145071, GSE145106, GSE145166, GSE145211, GSE145279, GSE145410.

3. Results

AML cell lines are sensitive to INCB059872, a selective LSD1 inhibitor

INCB059872 was developed as a potent, irreversible LSD1 inhibitor and initially tested for activity using ten hematopoietic cell lines (Lee et al., 2016). INCB059872 triggered differentiation of myeloid leukemia cell lines containing MLL translocations, so we explored the molecular mechanism of action of INCB059872 against THP-1 (MLL-AF9) and MV-4–11 (MLL-AF4). These two cell lines had different responses to INCB059872, as THP-1 showed a growth defect within one cell doubling time or approximately 3 days (Fig. 1A). Coincident with the slowing of cell growth, THP-1 cells began expressing CD11b and showed morphological changes indicative of myeloid differentiation. Interestingly, even if the drug was removed after the first 24 hr of cell culture, differentiation was still induced by 72 hr, (Supplementary Fig. S1A), which confirms that INCB059872 is an irreversible inhibitor. Conversely, INCB059872-treated MV-4–11 cells continued to grow at the same rate as DMSO-treated cells for multiple cell divisions (Fig. 1A) and showed no overt signs of differentiation (Fig. 1B, C). To confirm that the phenotypes observed with INCB059872 were on-target, we used shRNAs targeting KDM1A (LSD1) and factors that interact with LSD1 in THP-1 cells. Knockdown of either KDM1A (LSD1) or RCOR1, but not non-targeting controls, caused a dramatic increase in the percentage of cells expressing CD11b (Figs. 1D, S1B), suggesting that the effects of INCB059872 were mediated by inactivation of CoREST rather than by inhibiting the NuRD complex.

Figure 1.

LSD1 inhibition impairs proliferation and induces differentiation of AML cell lines. (A) Growth curves of THP-1 (left) or MV-4–11 (right) following treatment with INCB059872. 25nM in THP-1; 100nM in MV-4–11 (B) Flow cytometry analysis of CD11b expression in THP-1 (left) or MV-4–11 (right) at 3 days after treatment with INCB059872. (C) Wright-Giemsa staining of THP-1 and MV-4–11 cells after 3-day INCB059872 treatment; images taken at 400X. (D) Flow cytometry analysis of CD11b expression in THP-1 cells 10 days following lentiviral infection with shRNA targeting LSD1 or interacting proteins. shNT, non-targeting control. **p<0.01, ***p<0.001

PRO-seq identifies genes and enhancers regulated by INCB059872

To begin to characterize the changes in gene expression triggered by INCB059872, we performed RNA-seq analysis of the early (3hr) and intermediate (24hr) effects of LSD1 inhibition in these cell lines (Fig. 2A). At the 3hr timepoint, there were relatively few changes in expression that met the significance cutoff (q-value <0.05). In THP-1, only 31 transcripts were upregulated and 92 downregulated at least 1.5-fold (Supplementary Fig. S2A). In contrast, MV-4–11 cells did not have any significantly downregulated genes, but 194 genes were upregulated at least 1.5-fold (Supplementary Fig. S2A).

Figure 2.

RNA-seq analysis of LSD1i in AML cell lines. (A) Clustered heatmaps showing ln(RPKM +1) values with Pareto scaling and row-centering for genes with FDR < 0.05 in THP-1 24hr analysis. (B) Volcano plots showing gene expression changes in each cell line after 24hr INCB059872 treatment. Transcripts with fold-change > 1.5 and q-value < 0.05 are indicated by red dots (increased) or blue dots (decreased). (C) KEGG overrepresentation analysis of 178 genes commonly upregulated in THP-1 and MV-4–11 cells at 24hr timepoint. (D) Ranked list gene set enrichment analysis shows downregulation of MYC target genes in THP-1 (left) and MV-4–11 cells (right) after 24hr INCB059872. (E) Ranked list gene set enrichment analysis shows gene expression changes in THP-1 cells after 24hr INCB059872 are similar to changes caused by knockdown of HOXA9 in MOLM-14 cells (Faber et al., 2009). NES; normalized enrichment score.

By 24hr after drug treatment in both cell lines, RNA-seq revealed hundreds of significant changes in gene expression. The clustered heatmap in Fig. 2A displays relative expression levels of genes with significant changes after 24hr treatment in THP-1 cells (left) and the relative expression values of those genes in MV-4–11 cells (right). Remarkably, more than twice as many genes were induced at least 1.5-fold in MV-4–11 (1338) as compared to THP-1 (448; Fig 2A, 2B), even though THP-1 cells showed a far more dramatic phenotype in response to LSD1i (Fig. 1). In fact, for THP-1, slightly more genes were down-regulated (459) than induced (Fig. 2B, upper panel). Nevertheless, of the genes induced in THP-1 cells, 40% were also induced in MV-4–11 (Supplementary Fig. S2B), suggesting a commonly regulated set of genes. KEGG pathway analysis of the 178 genes upregulated in both cell lines indicates that differentiation-related pathways were affected by INCB059872; categories identified as enriched encompass processes associated with monocyte/granulocyte functions, such as “antigen processing & presentation”, “phagosome”, and “toll-like receptor signaling pathway” (Fig. 2C). Upregulated genes in the “hematopoietic cell lineage” pathway include CD1C/D, CSF1R/2RA, HLA-DMB, and ITGAM. Interestingly, MYC target genes were downregulated in both cell lines, while only MV-4–11 cells had decreased expression of oxidative phosphorylation genes in response to LSD1i (Figs. 2D, S2C). Gene set enrichment analysis showed that the expression changes induced by INCB059872 in THP-1 cells (Fig. 2E) were similar to the changes caused by knockdown of HOXA9 (Faber et al., 2009). Gene expression changes in MV-4–11 did not correlate with the HOXA9 signature (Supplementary Fig. S2D). Given that HOXA9 drives a gene expression pattern that promotes self-renewal (Vijapurkar et al., 2004), reduced expression of these target genes likely contributed to myeloid differentiation. Additionally, KIT expression was decreased in both cell lines, supporting the evidence that INCB059872 shifts cells toward later stages of hematopoiesis.

As LSD1 binds to intergenic and intronic regions as well as promoters and can control H3K4 methylation, we expected that INCB059872 treatment would affect enhancer activity, but we reasoned that it could also affect transcriptional elongation through accumulation of H3K4me3 (Benayoun et al., 2014). Therefore, we used precision nuclear run-on transcription and sequencing (PRO-seq) at 6, 12, and 24hr after addition of INCB059872 to the culture medium to measure genome-wide nascent transcription, as this is one of the best methods to identify active enhancers and RNA polymerase pausing and elongation (Wang et al., 2018). In THP-1 cells, INCB059872 caused upregulation of only about 200 genes by 24hr, with several of these changes being detectable at earlier time points as well (Fig. 3A). By mapping the relative polymerase density in drug-treated versus control cells, we determined that those genes affected by INCB059872 had altered polymerase initiation, rather than affecting promoter-proximal paused polymerase and/or polymerase elongation (Fig. 3B). Using a ranked list of gene expression changes, gene set enrichment analysis identified several differentiation-related gene sets that were upregulated after the 24hr drug treatment of THP-1 cells. These pathways included NFkB signaling, toll-like receptor signaling, cytokine receptors, inflammation, and myeloid differentiation (Fig. 3C). MV-4–11 cells also showed significant increases in transcription for genes within these categories (Supplementary Fig. S3A), implying that the cells were blocked from fully committing to a differentiation program made available by LSD1 inhibition.

Figure 3.

PRO-seq analysis identifies genes and enhancers regulated by INCB059872 in THP-1 cells. (A) Venn diagrams of genes with > 1.5-fold change (padj < 0.05) in gene body transcription at 6, 12, or 24hr after INCB059872 treatment. Left, upregulated genes; right, downregulated genes. (B) Heatmap of polymerase density surrounding TSSs of genes meeting 1.5-fold change cutoff at 24hr. Yellow indicates higher density of active polymerase at a locus in treated cells relative to control cells. (C) Ranked list gene set enrichment analysis. Individual plots show gene lists that are upregulated by 24hr LSD1i. (D) Venn diagrams of enhancers with > 1.5-fold change (padj < 0.05) in transcription at 6, 12, or 24hr after INCB059872 treatment. Left, upregulated enhancers; right, downregulated enhancers. (E) Homer motif analysis of the 1,278 enhancers that are upregulated by 24hr LSD1i. *Transcription factor identified by Homer as having DNA recognition sequence most similar to the identified motif. (F) IGV screenshot of PRO-seq signal at GLIPR1 gene and nearby enhancers. Asterisk indicates GFI1 binding motif.

Transcriptional changes at enhancers (identified by intergenic bidirectional transcription) outnumbered changes within gene bodies. At 24hr after INCB059872 treatment, there were nearly 1300 enhancers with at least 1.5-fold increase in transcription at the enhancer “start site” in THP-1 cells (Fig. 3D). Motif analysis of the sequences within these enhancers identified the GFI1/GFI1B recognition sequence as the most highly enriched (Fig. 3E). This consensus sequence was also enriched in enhancers changed after INCB059872 treatment in MV-4–11 cells (Supplementary Fig. 3B). Thus, the primary targets of INCB059872 appeared to be sites at which LSD1 cooperated with GFI1/GFI1B to repress transcription. Within this dataset we pinpointed GFI1 as the important factor, as GFI1B was not expressed in THP-1 cells (Supplementary Fig. 3C). Interestingly, GFI1B was not expressed in untreated MV-4–11 cells but was transcribed at 24hr after drug treatment (Supplementary Fig. S3C). An example genome browser track (Fig. 3F) from INCB059872-treated THP-1 cells shows gradual increases in active polymerase over time throughout the gene body as well as at two upstream enhancers of GLIPR1, a gene often silenced in AML (Xiao et al., 2011).

Transcriptional changes caused by INCB059872 are consistent with loss of LSD1:CoREST activity at GFI1 binding sites

LSD1 controls H3K4 methylation, but global changes in histone methylation were not detected after INCB059872 treatment of THP-1 cells using western blot analysis (Supplementary Fig. S4A). Nevertheless, we anticipated that inhibition of LSD1 would cause an accumulation of H3K4 mono- and di-methylation at specific, regulated loci. Therefore, we performed ChIP-seq for these marks 48hr after INCB059872 treatment and included an analysis of H3K27ac 24hr after treatment (Fig. 4A). This experiment included a spike-in control of D. melanogaster S2 cells to ensure that normalization did not minimize the drug effect. For the H3K4me1 and H3K4me2 analysis, only 5 peaks showed a significant change (H3K4me2, Fig. 4A), yet there was a trend of increased H3K4 dimethylation in treated versus control cells. In contrast, ChIP-seq for H3K27ac showed locus-specific increases with 111 peaks significantly upregulated over 1.5-fold (Fig. 4A; padj = 0.05). Over half of the significantly upregulated H3K27ac peaks located in intergenic regions overlapped with enhancers identified as upregulated in our PROseq datasets (Supplementary Fig. S4B).

Figure 4.

Transcriptional changes caused by INCB05982 are consistent with loss of LSD1:CoREST activity at GFI1 binding sites. (A) MA plots of H3K4me1 or H3K4me2 ChIP-seq changes after 48hr INCB059872 or H3K27ac ChIP-seq changes after 24hr INCB059872 in THP-1 cells. Red dots indicate FDR<0.1. (B) Histograms showing ChIP-seq coverage (per bp per peak) for H3K4me2 (left) or H3K27ac (right) at 1,278 intergenic enhancers upregulated by 24hr INCB059872. (C) IGV screenshot of CYBB locus and upstream enhancers; top: PRO-seq signal +/− 24hr INCB059872; middle: H3K27ac ChIP-seq signal +/− 24hr INCB059872; bottom: H3K4me2 ChIP-seq signal +/− 48hr INCB059872. Asterisk indicates GFI1 binding motif. (D) Homer motif analysis of the 111 H3K27ac peaks with >1.5-fold change (padj<0.05) after 24hr LSD1i. *Transcription factor identified by Homer as having DNA recognition sequence most similar to the identified motif. (E) 293T cells were transfected with CMV-GFI1 and treated with DMSO or 250nM INCB059872 for 48 hr. Cell lysates were immunoprecipitated with anti-LSD1 or IgG control. Western blots representative of three experiments are shown.

Histograms of the ChIP-seq signal at the 1278 enhancers that were upregulated in the PROseq analysis demonstrated a more dramatic shift in H3K27ac than in H3K4me2 (Fig. 4B). For example, CYBB encodes a component of the oxidase system of phagocytes and has a GFI1 binding motif ~27 kb upstream of its start site, coinciding with an enhancer. Genome browser tracks of CYBB show increases in transcription and acetylation throughout gene body and at the upstream enhancers compared to histone methylation that was unchanged (Fig. 4C). Overall, the significantly increased H3K27ac peaks were enriched for GFI1/1B consensus DNA binding motifs (Fig. 4D).

Our PRO-seq findings are consistent with the increases in H3K27 acetylation observed when THP-1 cells were treated with OG86, an LSD1 inhibitor previously shown to disrupt the LSD1:GFI1 interaction (31). Gene expression changes caused by INCB059872 as early as 6hr post-treatment showed strong positive or negative enrichment of genes up- or down-regulated, respectively, after OG86 treatment (Supplementary Fig. S4C). These data imply that INCB059872 disrupted the GFI1:LSD1 interaction, prompting us to test this directly by co-immunoprecipitation. LSD1 and GFI1 were co-expressed in HEK293T cells and treated for 48hr with INCB059872. GFI1 co-purified with LSD1 in the anti-LSD1-immunoprecipitate, but showed substantially less association in the presence of INCB059872 (Fig. 4E). These data support the notion that INCB059872 increased enhancer activity and gene expression by disrupting the association of GFI1:LSD1:CoREST at loci where this complex would normally repress transcription via histone deacetylase activity.

Single-cell RNA-seq analysis of AML patient bone marrow reveals gene expression changes caused by INCB059872

INCB059872 is in early phase trials for clinical development. Although few patients have been treated with INCB059872, one patient with new diagnosis, poor risk treatment-related acute myeloid leukemia (t-AML) with TP53 mutation and complex genetics showed a remarkable, unexpected response with INCB059872 + 5’azacitidine (AZA) therapy. This provided a unique opportunity to study the gene expression changes in the bone marrow of an AML patient who responded to treatment with INCB059872 + AZA. Subpopulations within a sample are likely to have variable responses to drug treatment, so we chose scRNA-seq to be able to observe these effects. For this experiment, a pre-INCB059872 treatment bone marrow sample was divided and treated ex vivo in duplicate with vehicle, INCB059872, AZA, or INCB059872 + AZA. Though a longer treatment duration would have been ideal to detect differentiation-related changes, a 48hr timepoint was chosen to avoid measuring drug-independent cell death. Cells were passed over Annexin V depletion columns to exclude dying cells from the experiment. Likely due to leukemic cells being able to withstand culture conditions, the majority of the cells expressed myeloid markers (Supplementary Fig. 5A). Unsupervised clustering of nearly 30,000 cells identified 11 different populations based on mRNA expression of 1500 to 5500 genes (Figs. 5A, S5B). Surprisingly, the majority of INCB059872- and combination-treated cells were assigned to clusters that were not found in control- or AZA-treated samples (Fig. 5B). Within this treatment window, AZA had a negligible effect on gene expression compared to INCB059872. Consistent with the results in THP-1 cells, AML blasts exposed to INCB059872 dramatically upregulated GFI1 and GFI1B (Fig. 5C). Additionally, several of the most highly increased transcripts (e.g. ANXA2, GLIPR1) measured in LSD1i-treated AML cell lines were also upregulated in this patient sample (Fig. 5C), which suggests that the leukemic blasts from this patient were pushed toward myeloid differentiation, likely via the same mechanism as in THP-1 cells.

Figure 5.

Single-cell RNA-seq analysis of AML patient bone marrow reveals gene expression changes caused by INCB059872. (A-C) AML patient bone marrow sample was divided and treated ex vivo with DMSO, 100nM INCB059872, 500nM azacitidine (AZA), or combination of INCB059872 + AZA. (A) UMAP plot shows unsupervised clustering of all 29,278 patient bone marrow cells. (B) UMAP plots, separated by treatment group. (C) Heatmaps of individual gene expression displayed on UMAP plots. Intensity of blue color indicates level of expression.

Single-cell RNA-seq reveals changes in bone marrow progenitor populations following INCB059872 treatment in mice

While the human AML patient sample provided a unique opportunity to gain insights into how INCB059872 affected gene expression to trigger myeloid differentiation, it was limited by the availability of a single responding patient. Therefore, we turned to mice to further explore how INCB059872 affects gene expression and myeloid differentiation, while also assessing the thrombocytopenia found in preclinical studies of INCB059872 in mice. C57BL/6 mice were treated with 10 mg/kg INCB059872 via oral gavage daily and circulating platelet counts began to drop after four days and continued to fall at day six. Therefore, these times were selected for scRNAseq. Bone marrow was harvested and lineage-negative cells isolated for scRNAseq analysis. After quality control filtering, this dataset consisted of 15,046 individual cells, with 750 to 4000 transcripts detected per cell (Supplementary Fig. S6A). Unsupervised clustering divided these cells into 22 unique clusters, which were identified by expression levels of known lineage-defining genes (Fig. 6B). While the distribution of cells into different progenitor populations was largely unaffected by drug treatment, these data revealed a striking increase in the number of cells assigned to a megakaryocyte stem/progenitor cluster (Fig. 6C, cluster 9; Supplementary Fig. S6B). Cells within this expanded cluster expressed stem cell markers such as MYCN and PBX1, but also expressed VWF (Supplementary Fig. S6C). VWF expression defined a population of platelet-primed HSCs, which have multipotent potential in vitro yet exclusively adopted a megakaryocyte lineage upon transplantation into primary recipients (Sanjuan-Pla et al., 2013; Carrelha et al., 2018). Interestingly, these megakaryocyte lineage-restricted cells displayed reduced platelet output when compared to other less restricted HSC populations (Carrelha et al., 2018). The expansion of a progenitor population with lower platelet yield is consistent with the lack of mature platelets in LSD1i-exposed animals. Notably, INCB059872 caused dramatic upregulation of GFI1 and GFI1B in clusters 1, 3, 4, and 9 (Fig. 6D). Additionally, GFI1 expression increased within neutrophil progenitor cells (cluster 7), and GFI1B expression increased within erythrocyte progenitor cells (cluster 16), suggesting that LSD1-mediated control of GFI1B contributed to the phenotype.

Figure 6.

Single-cell RNA-seq reveals changes in bone marrow progenitor populations following INCB059872 treatment in mice. (A) Platelet counts of mice treated daily with 10 mg/kg INCB059872. (B-D) Single-cell RNA-seq analysis of bone marrow from wild-type mice treated daily with 10 mg/kg INCB059872 for 0, 4, or 6 days. (B) UMAP plot of all 15,046 mouse bone marrow cells; colored according to unsupervised clustering. HSPC, hematopoietic stem progenitor cell; GMP, granulocyte-monocyte progenitor; MkP, megakaryocytic progenitor; cDC, conventional dendritic cell; pDC, plasmacytoid dendritic cell. (C) UMAP plots; cells separated by treatment group. Circled clusters correspond to bar graphs, showing average percentage of cells from a mouse that were assigned to each cluster; 3 mice per condition. (D) Heatmaps of individual gene expression displayed on UMAP plots, separated by treatment group. Intensity of red color indicates level of expression.

LSD1 inhibition also altered the distribution of cells in non-megakaryocytic clusters. Cells in cluster 10 were classified as a plasmacytoid dendritic cell progenitor based on expression of Siglech and Ly6d (Rodrigues et al., 2018). After 4 days of INCB059872 treatment, mice had fewer cells with this identity (Fig. 6C). Cluster 18, a Csf1r⁺/Cd36⁺ pre-monocytic cell population, was almost completely absent after 6 days of treatment (Fig. 6C).

FACS analysis of bone marrow from LSD1i-treated mice confirms lineage defects

To confirm the changes in lineage distribution seen in our scRNA-seq data, we repeated the experimental design but this time distributed the bone marrow between different flow cytometry antibody panels for multi-lineage analysis. We did not observe significant changes in the total numbers of erythrocyte-depleted bone marrow cells between treated and control animals (Fig. S7A). Populations of mature lymphoid cells (B220+ B cells or CD3+ T cells) were not substantially affected by LSD1i (Fig. 7A, B). Interestingly, drug treatment transiently altered the distribution of granulocyte/monocyte and erythroid cells. After 4 days of LSD1i, the Ter119+ erythroid population had significantly decreased while the Gr1+/Mac1+ population showed a compensatory increase (Fig. 7C, D). After 6 days of treatment, there were no longer differences in these populations between treated and control mice.

Figure 7.

LSD1 inhibition impairs platelet production by causing accumulation of early megakaryocyte population. (A-K) Flow cytometry analysis of bone marrow from mice treated with 10 mg/kg INCB059872 for 0, 4, or 6 days; 3 mice per group. (A-D) Percentages of mature hematopoietic lineages out of total bone marrow. (A) B cell subsets defined by levels of B220 and IgM. (B) T cells defined by CD3 expression. (C) Erythroid cells defined by Ter119 expression. (D) Mature myeloid cells defined by Mac-1/Gr-1 expression. (E) LSK population defined as percentage of Sca-1+/c-Kit+ cells out of Lin− population. (F) LMPP population defined as percentage of Flt3^high cells out of LSK population. (G) Percentage of myeloid progenitor populations within Lin−/c-Kit+/Sca-1− subset; populations defined by levels of CD34 and CD16/32. (H) Percentage of CD41+/CD200r3− megakaryocyte progenitors within Lin− subset. (I) Percentage of c-Kit+ cells within megakaryocyte progenitor population. (J) Percentage of Ly6D+/SiglecH+ plasmacytoid dendritic cell progenitors within Lin- subset. (K) Percentage of CD36+/Fcgr4+ monocyte progenitors within Lin- subset. (L-M) RNA-seq from sorted meg progenitors (based on cluster 9 from scRNA-seq). Mice were treated with 10 mg/kg INCB059872 for 0, 4, or 6 days; 2 mice per group. (L) Clustered heatmaps showing ln(RPKM +1) values with Pareto scaling and row-centering for 1000 genes with lowest FDR. (M) Ranked list gene set enrichment analysis shows “platelet activation signaling and aggregation” gene signature is downregulated at both timepoints. *p<0.05, **p<0.01, ***p<0.001.

As for classically defined progenitor populations, we did not observe any significant changes in the size of these populations after LSD1i. The Lin− Sca1+ cKit+ stem and progenitor populations (LSK), lymphoid-primed multi-potent progenitor cells (LMPP, LSK Flt3^high), common myeloid progenitor cells (CMP,Lin− Sca1− cKit+ CD34+ CD16/32^low), granulocyte/monocyte progenitor cells (GMP,Lin− Sca1− cKit+ CD34+ CD16/32^high), and megakaryocyte/erythroid progenitor cells (MEP, Lin− Sca1− cKit+ CD34+ CD16/32−) populations remained the same size after INCB059872 treatment (Fig. 7E–G). To monitor the number of cells corresponding to cluster 9 from our scRNA-seq dataset, we gated on the Lin− Itga2b+ CD200r3− population (Fig. S7B). As expected, mice from both treatment timepoints had a dramatically higher percentage of cells that fell into this population as compared to vehicle control mice (Fig. 7H). Importantly, the percentage of cells within this population that expressed Kit was also substantially increased by INCB059872 (Fig. 7I), further supporting the notion that the drug caused the expansion of a stem cell population. We also observed a significant reduction in the size of the plasmacytoid dendritic cell population six days after treatment (Fig. 7J, Supplementary Fig. S7C). Cluster 18 was defined by expression of CD36 and Fcgr4, so we used these markers to quantify this pre-monocyte population over time and confirmed that INCB059872 reduced the size of this population within 4 days of drug treatment (Fig. 7K, Supplementary Fig. S7D).

Finally, we sorted the megakaryocyte progenitor population (Lin− Itga2b+ CD200r3−) for bulk RNA-seq to measure gene expression changes after 4 or 6 days of INCB059872 treatment. This analysis indicated that most of the changes in gene expression had occurred by day 4 (Fig. 7L). While there were many transcripts affected by the drug treatment, only a few dozen met statistical significance with 61 genes up and 113 down by at least 1.5-fold after 6 days of treatment (Supplementary Fig. S7E). Importantly, gene set enrichment analysis revealed that a gene signature associated with platelet function was significantly downregulated at both time points (Fig. 7M). This signature included factors that are critical for platelet aggregation (P2ry12, Thbs1, Pf4), as well as receptors for TPO (Mpl) and VWF (Gp1ba, Gp5, Gp9). These data further support our finding that the population of cells expanded after INCB059872 treatment consisted of megakaryocyte-biased stem cells that failed to mature into efficient platelet producers.

4. Discussion

Inhibiting LSD1 is a therapeutic strategy for myeloid malignancies, as this was thought to increase H3K4 methylation to increase gene expression of LSD1-regulated genes to promote myeloid differentiation. Because LSD1 was originally identified as a histone demethylase, it was assumed that inhibiting this enzymatic activity would be necessary for an LSD1-based approach to be effective. However, INCB059872 and OG86 (ref. 31) appear to act by disrupting the interaction between LSD1 and the SNAG domain-containing proteins GFI1 and GFI1B, which is sufficient to induce differentiation of AML cells without causing dramatic changes in histone methylation (Ishikawa et al., 2017). GFI1 and GFI1B recruit the CoREST complex to repress target genes via histone deacetylase activity (Saleque et al., 2007). In fact, accumulation of H3K27 acetylation appeared to be the primary effect of INCB059872 and OG86. In our observations with INCB059872 in THP-1 cells, H3K4 dimethylation had just begun to increase 48hr after INCB059872 treatment while hundreds of enhancers had increased transcription after only 6hr, and shRNAs directed to CoREST induced similar effects. Thus, the timing of transcriptional changes was better aligned with loss of recruitment of CoREST at GFI1/1b regulated genes and enhanced histone acetylation, the latter of which changes at a faster rate than histone methylation (Barth and Imhof, 2010). As suggested by LSD1 inhibition in murine AML models (Cusan et al., 2018; Barth et al., 2019), the loss of repressor activity can allow lineage-defining transcription factors such as PU.1 or CEBPα to activate transcription of genes that had been repressed by GFI1.

We have taken advantage of PRO-seq analysis to identify genes and enhancers that are impacted by INCB059872 treatment. This method provides high resolution maps of actively-transcribing polymerases throughout the genome, giving the ability to detect changes in gene regulation before steady-state mRNA levels are affected, while allowing us to assess the transcription of all genes (including miRNAs) and enhancers. These data are also useful for overlaying with genomic maps of histone modifications, as we were able to see that transcriptional changes more closely match H3K27 acetylation than H3K4 methylation. In our analysis, we found hundreds of transcriptional changes at enhancers that preceded changes in gene body transcription, which is consistent with the interpretation that INCB059872 released CoREST from GFI1/GFI1b at enhancers, but did not greatly impact H3K4me to affect promoter-proximal pausing and polymerase elongation. As chromatin-associated LSD1 is primarily located at intergenic/intronic regions, it makes sense that the earliest, likely direct effects of LSD1 inhibition would be at enhancers.

Single-cell RNA-seq is a powerful tool for deep phenotyping of clinical samples. We were able to assess the expression of between 1500 and 5500 genes per cell, which allowed us to pinpoint a small fraction of cells within mouse bone marrow that were affected by LSD1 inhibition. Our data provide further evidence of a VWF-high megakaryocyte-biased stem cell population (Sanjuan-Pla et al., 2013), as these cells expressed Nmyc, Pbx1, Meis1 and other genes that are hallmarks of self-renewing stem cells. Within INCB059872-treated mice, this subset of cells constituted a higher percentage of the bone marrow and downregulated expression of key genes involved in platelet function, such as Mpl, Pf4, Ppbp, and Thbs1. Given that thrombocytopenia is an on-target side effect of INCB059872 treatment, the accumulation of these cells (labeled as MkP in Fig. 6B), suggests that while disruption of LSD1:CoREST activity caused myeloid differentiation, within the megakaryocyte lineage, INCB059872 either pushed more cells into this lineage or impaired the efficient maturation into platelet producing cells (Fig. 6, 7).

Single-cell RNA-seq allowed the segregation of AML blasts (control) into nine different clusters (Fig. 5A, 5B). The small clusters representing lymphoid cells (clusters 6, 8, 9, and 10) did not significantly change upon drug treatment. In contrast, the clusters containing myeloid markers (clusters 1, 2, 3, 5, 7) were dramatically altered by the 48hr incubation with INCB059872. Gene expression was altered in much the same way in these primary leukemic blasts as in THP-1 cells treated with INCB059872, which confirms that THP-1 is a useful model system for studying the transcriptional effects of INCB059872. Moreover, new clusters of cells expressed megakaryocytic markers, including GFI1 and GFI1b (Fig. 5C). It is notable that a 48hr treatment with AZA had only modest effects on gene expression, which suggests that its primary mode of action in the first 48hr is not to affect gene expression. This study indicates the power of scRNA-seq for pathophysiological analysis of human AML and of the effects of epigenetic treatments. As the extreme costs of this analysis drop over time to allow more such studies, biomarkers of response to these therapies will follow, which may allow preselection of patient populations for clinical trials.

Abbreviations list

LSD1

lysine demethylase 1

KDM1A

lysine demethylase 1A

PR-DUB

polycomb repressive-deubiquitinase

AML

acute myeloid leukemia

scRNAseq

single cell RNA sequencing

NuRD

nucleosome remodeling and deacetylase complex

GFI1 and GFI1B

Growth Factor Independent 1/1B

PROseq

precision nuclear run-on transcription sequencing

ChIPseq

chromatin immunoprecipitation

KEGG

Kyoto Encyclopedia of Genes and Genomes