Hypomethylating agents (HMA) have become the backbone of nonintensive acute myeloid leukemia/myelodysplastic syndrome (AML/MDS) treatment, also by virtue of their activity in patients with adverse genetics, for example, monosomal karyotypes, often with losses on chromosome 7, 5, or 17. No comparable activity is observed with cytarabine, a cytidine analogue without DNA-hypomethylating properties. As evidence exists for compounding hypermethylation and gene silencing of hemizygous tumor suppressor genes (TSG), we thus hypothesized that this effect may preferentially be reversed by the HMAs decitabine and azacitidine. An unbiased RNA-sequencing approach was developed to interrogate decitabine-induced transcriptome changes in AML cell lines with or without a deletion of chromosomes 7q, 5q or 17p. HMA treatment preferentially upregulated several hemizygous TSG in this genomic region, significantly derepressing endogenous retrovirus (ERV)3–1, with promoter demethylation, enhanced chromatin accessibility, and increased H3K4me3 levels. Decitabine globally reactivated multiple transposable elements, with activation of the dsRNA sensor RIG-I and interferon regulatory factor (IRF)7. Induction of ERV3–1 and RIG-I mRNA was also observed during decitabine treatment in vivo in serially sorted peripheral blood AML blasts. In patient-derived monosomal karyotype AML murine xenografts, decitabine treatment resulted in superior survival rates compared with cytarabine. Collectively, these data demonstrate preferential gene derepression and ERV reactivation in AML with chromosomal deletions, providing a mechanistic explanation that supports the clinical observation of superiority of HMA over cytarabine in this difficult-to-treat patient group.
These findings unravel the molecular mechanism underlying the intriguing clinical activity of HMAs in AML/MDS patients with chromosome 7 deletions and other monosomal karyotypes.
See related commentary by O'Hagan et al., p. 813
DNA hypomethylating agents (HMA) have become the mainstay of nonintensive acute myeloid leukemia/myelodysplastic syndrome (AML/MDS) treatment, increasingly superseding low-dose cytarabine as the backbone of first-line treatment when standard induction chemotherapy is not an option. This is by virtue of their limited nonhematologic toxicity, allowing treatment also of elderly patients with reduced performance status and comorbidities. Furthermore, they display a surprising activity in patients with adverse genetics, particularly monosomal karyotypes and TP53 mutations, which is not observed with cytarabine. This, at first counterintuitive, clinical observation was seen in MDS studies using 5-aza-2′deoxycytidine (decitabine; refs. 1–3), and later also described for 5-azacytidine (azacitidine; refs. 4–6). This important characteristic sharply differentiates the DNMT-inhibiting azanucleosides decitabine and azacitidine from the structurally related cytidine analogue cytarabine (lacking DNMT-inhibitory activity); the underlying molecular mechanism, however, is not yet well understood (7, 8).
In AML and MDS, loss of chromosomal material (most frequently involving chromosomes 7, 5, and 17) often results in hemizygous instead of normal, biallelic expression of tumor suppressor genes (9, 10). Such a gene dosage reduction may be compounded by further epigenetic repression under selection pressure. Zhou and colleagues described that with loss of one copy of chromosome 7 in MDS, enhanced gene silencing by DNA hypermethylation of the remaining allele occurred for the tumor suppressor gene DOCK4 (located on 7q31.1; ref. 11). This cooperating silencing mechanism of gene loss and hypermethylation was described in AML/high-risk MDS for the tumor suppressor CNNTA1 on the long arm of chromosome 5, and has also been demonstrated in primary osteosarcoma and breast cancer cells (12–14).
We aimed to unravel the molecular mechanism underlying the intriguing clinical activity of HMAs in AML/MDS patients with deletions of the long arm of chromosome 7 [del(7q)] and other monosomal karyotypes (MK). We hypothesized that these drugs may be employed to identify genes that are cooperatively repressed on hemizygous chromosomal regions (compared with intact biallelic regions). Because gene-regulatory effects of decitabine and azacitidine can also occur in the absence of promoter methylation changes, these derepressive events were primarily approached by transcriptome analyses (15).
The recurrently deleted regions on the long arm of chromosomes 7 and 5 (7q and 5q) and the short arm of chromosome 17 (17p) harbor many genes relevant to hematopoietic cancers, often with putative or bona fide tumor suppressor activity (16–18). Therefore, we selected these chromosomal regions for expression mapping. As an in vitro model to study HMA-induced gene derepression on 7q, we employed the UCSD-AML1 (AML1) cell line, which displays a monosomy of the entire chromosome 7, but carries only very few additional chromosomal gains or losses. As a comparator cell line, and as model for a del(5q) and del(17p) AML, the cell line ELF-153 (ELF) was chosen, because it has two intact alleles of 7q, but large deletions on 5q and 17p. Using RNA-sequencing (RNA-seq) to identify transcripts that are differentially regulated by decitabine, we identified several genes on 7q, as well as 5q and 17p, which displayed derepression preferentially when in a hemizygous state. Validation experiments were conducted by profiling in vivo transcriptional changes in sorted primary blood blasts serially isolated from patients receiving decitabine (19). Finally, we could demonstrate in AML patient-derived xenograft (PDX) models (established from patients with either MKs or normal cytogenetics) that the antileukemic activity of decitabine was superior to cytarabine, also in AML models with multiple chromosomal losses.
Materials and Methods
Cell culture and drug treatment
Four AML cell lines were chosen for decitabine treatment: UCSD-AML1 (AML1), ELF-153 (ELF), KG-1, and U937. 100 nmol/L decitabine for AML1, ELF, and KG-1, 200 nmol/L decitabine for U937, as well as 1,000 nmol/L azacitidine for AML1 and ELF, were selected as optimal for studying transcriptional effects of demethylating treatment because of the highest reduction of proliferation without induction of cell death >20%. To ensure adequate nucleoside incorporation, cells were treated for either 96 hours (AML1, ELF; Supplementary Fig. S1A), or, given the higher doubling rate, for 72 hours (KG-1, U937). Proliferation and viability were determined by acridine orange/propidium iodide staining and subsequent detection by Luna Automated Cell Counter (Logos Biosystems). All cell lines were purchased from DSMZ (Leibniz Institute, DSMZ-German Collection of Microorganisms and Cell Cultures GmbH) and routinely tested for Mycoplasma (all cell lines were Mycoplasma free).
Characterization of cell lines
SNP arrays on genomic DNA (isolation as described previously; ref. 20) were performed using CytoScan750K (AML1, ELF, KG-1) and Genome-Wide SNP 6.0 (U937) arrays (Affymetrix, Thermo Fisher Scientific). Specific chromosomal aberrations were validated by interphase FISH using probes against 7q22 (SpectrumGreen; Abbott Laboratories), 7q31 (SpectrumOrange), and the centromere of chromosome 7 (CEP X SpectrumAqua); chromatin was stained with DAPI. Mutational screening for AML1 and ELF was performed using the TruSight Myeloid Sequencing Panel (Illumina) and sequenced on an Illumina Miseq sequencer. Reads were analyzed using the SeqPilot 4.3.0 Software (JSI Medical Systems).
Methylation analyses by MassARRAY
After bisulfite conversion, methylation of genomic DNA was evaluated by matrix‐assisted time‐of‐flight mass spectrometry (MassARRAY, Agena Bioscience) as described previously (21, 22). Amplicons in the ERV3–1 promoter region were selected on the basis of GC richness (although this region does not meet the formal criteria for a CpG island) and lack of repetitive elements (Supplementary Fig. S1B). Primer sequences are listed in Supplementary Table S1.
Total RNA was isolated, transcribed to cDNA, and selected target genes were validated by qRT-PCR as described previously (20). Results were analyzed by second derivative maximum quantification using ACTB, RNF20, and TGDS as reference genes (LightCycler480 Software, Roche). Primer sequences are shown in Supplementary Table S1, ACTB and RNF20 reference primers were purchased from Primerdesign Ltd.
RNA-seq and data analysis
Total RNA was isolated as described previously and depleted of ribosomal RNA (Ribo-Zero Gold rRNA Removal Kit, Illumina; ref. 20). cDNA libraries (strand-specific, 50 bp) were sequenced on an Illumina HiSeq2500 sequencer with 35 million paired-end reads/sample. Read quality was checked with FastQC (≥95% >Q30), alignment to the reference genome GRCh37/hg19 was performed with TopHat2, read-counting with HTseq-count, differential expression testing with DESeq2 (FDR < 0.01; protein-coding transcripts only; 23–25). GO enrichment analysis was performed with Metascape (26). For analysis of transposable elements, reads were aligned to the modified RepeatMasker annotation as published by Jin and colleagues, which was condensed compared with the original annotation by filtering out low complexity and simple repeats, rRNA, scRNA, snRNA, srpRNA, and tRNA (27). Read-counting was performed with enabled counting of nonuniquely mapped reads. All analysis steps were calculated on the European Galaxy instance (https://usegalaxy.eu/; ref. 28).
Assay for transposase accessible chromatin-sequencing and data analysis
For assay for transposase accessible chromatin-sequencing (ATAC-seq), 50,000 fresh cells were processed in triplicates as described in the Omni-ATAC protocol by Corces and colleagues (29). Size selection and purification was performed using Agencourt AMPure XP beads (Beckman Coulter). Validation of enriched open chromatin sites was determined by qPCR for constitutively unoccupied TATA-box binding protein binding sites (TBP), a specific heterochromatic region on chromosome 18 (chr18) and mitochondrial DNA (mtDNA; ref. 30). Primer sequences are shown in Supplementary Table S1. Paired-end libraries were sequenced with 40 million reads/sample. Read quality was checked with FastQC (≥95% >Q35), alignment to the GRCh38/hg38 assembly was performed with Bowtie2, PCR duplicates were removed with MarkDuplicates and narrow peaks were called using Genrich (FDR < 0.05). Heatmaps of enriched regions overlapping with POLR2A-binding sites and visualization of gene-specific enrichment sites were generated using deepTools2 and HiCexplorer, respectively (31, 32). All analysis steps were calculated on the European Galaxy instance (https://usegalaxy.eu/) (28).
Fixed cells (1% formaldehyde, 10 minutes) were lysed and chromatin was sheared with the Covaris M220 Focused-Ultrasonicator (Covaris) using the Covaris TruChIP Shearing Kit according to manufacturer's protocol. Immunoprecipitation using the Diagenode αH3K4me3-antibody (Diagenode) was performed with the iDeal ChIP-qPCR kit. Quantitative PCR was performed as described above using 36B4 as a single-copy genomic reference gene to correct for copy-number differences and three primer sets localized in the ERV3–1 promoter region. The amplicons were selected on the basis of an overlap with H3K4me3 signals from publicly available chromatin immunoprecipitation (ChIP)-seq data (33) and, again, the lack of repetitive elements (Supplementary Fig. S1B). Primer sequences are listed in Supplementary Table S1.
Patient samples for expression analyses
Peripheral blood mononuclear cells (PBMC) were collected from 16 patients with AML [8 with MK and 8 normal karyotype (CN); see Supplementary Tables S2 and S3 for patients' characteristics] treated with decitabine (20 mg/m2 i.v. over 1 hour for 5 days) within the DECIDER trial (NCT00867672; ref. 19). Among the MK patients, 7 had a deletion of 7q [del(7q)], 5 had a deletion of 5q (del(5q)), and 7 patients a deletion of 17p (del(17p)). Leukemic blasts before treatment and at day 8 were isolated using automatic magnetic sorting of cells labeled with anti-human-CD34 and -CD117 microbeads (Miltenyi Biotec). RNA isolation, qRT-PCR, and expression profiling with Human Gene 2.0 arrays were performed as described previously (20). Patients provided written informed consent for the research use of the clinical data and biomaterial in accordance to the Declaration of Helsinki.
PDX models and in vivo treatment
T-cell–depleted peripheral blood or bone marrow cells from 6 patients with AML (Supplementary Tables S2 and S3) were injected into NSG mice (NOD/Shi-scid/IL2Rγnull; 3 × 10e6 cells/mouse) and stable PDX models were generated as described previously (34). Mice were treated with a low-toxic schedule of decitabine (1 mg/kg/day, i.p.), cytarabine (15 mg/kg/day, i.v.) or vehicle (PBS) for 5 consecutive days (one cycle). Drug doses were titrated to be equitoxic but as high as possible without causing side effects; the treatment duration was chosen according to the clinical standard of 5-day intravenous administration of decitabine. Treatment commenced upon disease onset, evaluated by positivity for human CD33, increased white blood counts in peripheral blood and weight loss. Overall survival served as the read-out. The study was carried out in accordance with the recommendations by the Society of Laboratory Animal Science in an AAALAC-accredited animal facility. The animal experiments were approved by the regional council (Regierungspräsidium Freiburg; ref. 35; permit no. G-12/86).
Raw data access
SNP array, RNA-seq, and expression array data are available at GEO under accession numbers GSE138438, GSE140347, and GSE138696, respectively.
Hypomethylating agent–induced transcriptome changes in AML cell lines with or without monosomy 7
Among a large panel of AML cell lines with different cytogenetic aberrations, two diploid AML cell lines were selected. One of them, UCSD-AML1 (AML1) bears a monosomy 7; the other, ELF-153 (ELF), has two intact copies of 7q (7p being deleted).
By SNP array analyses, both AML1 and ELF showed only a low number of aberrations compared with most AML cell lines available. In AML1, loss of one entire chromosome 7 was the only numerical aberration (Fig. 1A). In ELF, two intact copies of 7q and several deletions (and gains) common in AML/MDS were present, including del(5q) and del(17p) (Fig. 1B). Copy numbers of chromosome 7 were validated by interphase FISH (Supplementary Fig. S1C). Mutational analyses of AML1 and ELF revealed characteristic AML mutations in both cell lines, such as KIT and PTPN11 (AML1), DNMT3A, GATA2, NRAS, and TP53 (ELF; Supplementary Fig. S1D; ref. 18). Two additional control AML cell lines, KG-1 and U937, represented cytogenetically highly perturbed cell lines without monosomy 7.
To generate transcriptomes reflecting the effects of noncytotoxic decitabine concentrations, AML1 and ELF cells were both treated with 100 nmol/L decitabine, which reduced proliferation by 70% and 60% after 96 hours of treatment, respectively, whereas viability remained above 90% (Fig. 1C and D; similar effects also in KG-1 and U937; Supplementary Fig. S2A and S2B). Treatment with azacitidine resulted in comparable antiproliferative, noncytotoxic effects in AML1 and ELF (Supplementary Fig. S2C and S2D). ATAC- and RNA-seq of decitabine-treated cells disclosed a massive increase in accessible chromatin sites, leading to several thousands of significantly differentially expressed protein-coding transcripts (DETs; FDR < 0.01; Fig. 1E; Supplementary Fig. S2E) in AML1 and ELF, respectively (upregulated: 1,704/1,205; downregulated: 1,879/661 after 96 hours of decitabine). Accordingly, LINE-1 methylation, representative of global DNA methylation levels (22), was significantly reduced in both cell lines (Fig. 1F). Gene Ontology (GO) analysis of genome-wide upregulated transcripts showed a marked enrichment for immune-response and differentiation pathways whereas downregulated transcripts were enriched for replication and cell division terms, implying nonrandom HMA effects on the transcriptome in AML1 and ELF (Supplementary Fig. S3A and S3B).
Decitabine results in reexpression of hemizygous genes on chromosomes 7, 5, and 17
We hypothesized that hemizygous and additionally silenced genes may respond more strongly to HMA-induced derepression than genes that are present in two copies. We therefore integrated copy number and transcriptome data: In AML1, decitabine regulated a total of 121 transcripts on lost gene regions (all but one on chromosome 7), compared with only 19 transcripts on gained regions (Fig. 2A). A comparable number of hemizygous chromosome 7 transcripts was up- and downregulated (57 vs. 63), notably the mean induction was 4.9-fold (compared with 2.9-fold mean downregulation). Concordantly, in ELF, 176 transcripts on hemizygous regions (including 5q, 7p, and 17p) and 78 on gained regions were preferentially induced rather than repressed by decitabine (Fig. 2B).
To identify genes that were exclusively regulated by HMAs when hemizygous, we first analyzed protein-coding transcripts on 7q that were significantly regulated in AML1 but not in ELF (FDR < 0.01). Indeed, 43 genes on 7q were only altered in AML1 by decitabine, with 15 up- and 28 downregulated (Fig. 3A and B). Among the upregulated monosomy-specific genes with known tumor suppressor functions were homeodomain interacting protein kinase 2 (HIPK2, 7q34), zyxin (ZYX, 7q34), and high mobility group box transcription factor 1 (HPB1, 7q22.2). HIPK2 and ZYX were shown to be crucial for chromosomal stability and apoptosis (17, 35, 36), whereas HBP1 promotes immune response and negatively regulates DNMT1 transcription (37, 38). Validation by qRT-PCR confirmed the RNA-seq results: In AML1, untreated cells showed haploinsufficient expression compared with ELF. Decitabine significantly increased the mRNA expression levels of all three genes (HIPK2 1.8-fold, ZYX 1.5-fold, HBP1 2.6-fold). In contrast, in ELF only a 1.3-fold induction could be detected for HIPK2 (albeit statistically significant), with nonsignificant expression changes of 1.0-fold for ZYX and 1.3-fold for HBP1 (Fig. 3C).
To confirm whether these observed derepression events represent a class effect of azanucleoside DNMT inhibitors, we interrogated gene expression also in AML1 and ELF cells treated with 1,000 nmol/L AZA. Hereby we could show a comparable preferential upregulation of HIPK2, ZYX, and HBP1 mRNA by AZA in AML1 (2.0-, 1.5-, and 1.4-fold, respectively) compared with ELF (1.1-, 0.6-, and 0.7-fold; Supplementary Fig. S4).
To further validate the decitabine-induced derepression of hemizygous genes, we utilized sorted blasts from newly diagnosed patients with AML before and after a 5-day decitabine administration with del(7q) (n = 4) or normal karyotype (CN; n = 4) to analyze the induction of HIPK2 by qRT-PCR. Here, we could detect an, albeit not statistically significant, preferential induction in the del(7q) compared with the CN patient cohort (median induction 1.5- compared with 1.2-fold, respectively; Fig. 3D).
Induction of endogenous retroviral transcripts in leukemic blasts by decitabine
Next, we analyzed the RNA-seq data for genes that were significantly regulated by decitabine on 7q in both cell lines. If the starting hypothesis was correct, there would be genes that show higher levels of induction in AML1 than in ELF. In concordance with our starting hypothesis, 19 overlapping differentially regulated genes were found on 7q, 12 of which were concordantly upregulated and 7 downregulated (Fig. 3B).
The two most strongly upregulated genes on this list, also with the most pronounced difference between the two cell lines, were the endogenous retrovirus group 3 member 1 (ERV3–1) and the zinc finger protein 117 (ZNF117; Supplementary Fig. S5A). In detail, in AML1 ERV3–1 and ZNF117 were upregulated by decitabine 86.4- and 119.4-fold, respectively, compared with only 4.3- and 6.5-fold in ELF (Supplementary Fig. S5A). In line with this data, ATAC-seq showed increased accessible chromatin sites in the shared ERV3–1/ZNF117 promoter region (located at the 5′-end of ERV3–1) in AML1 but not in ELF153 when treated with decitabine (Fig. 4A). qRT-PCR of ERV3–1 and ZNF117 confirmed these derepressive events on this locus (localization of primers depicted in Supplementary Fig. S5B): in both cell lines, the two genes were highly upregulated, again to a much higher extent in AML1: induction of ERV3–1 was 83.6-fold (2.4-fold in ELF) and of ZNF117 was 191.7-fold (1.6-fold in ELF; Fig. 4B). Azacitidine treatment also resulted in a preferential, albeit weaker, upregulation of ERV3–1 and ZNF117 mRNA in AML1 (4.3- and 7.1-fold, respectively) compared with ELF (0.7- and 0.8-fold; Supplementary Fig. S5C). In two additional cell lines also diploid for 7q (KG-1, U937), decitabine-induced upregulation of ERV3–1 and ZNF117 after 72 hours was 2.1- and 1.5-fold in KG-1 (Supplementary Fig. S6A) and 4.1- and 6.5-fold, respectively, in U937 (Supplementary Fig. S6B).
Mechanistically, targeted bisulfite-MassARRAY of 17 CpGs revealed a hypermethylated ERV3–1/ZNF117 promoter region in untreated AML1 (mean methylation 75%) that could be significantly reduced by decitabine (mean methylation 63%, P = 0.0021). Especially highly (≥60%) methylated CpGs were targeted by decitabine (mean methylation 91% untreated vs. 77% decitabine) compared with CpGs with lower (<60%) methylation levels (mean methylation 45% vs. 37%). In ELF, the same region was almost completely unmethylated (mean methylation 5%) without any further demethylation by decitabine (Fig. 4C; Supplementary Fig. S7). Decitabine treatment also increased enrichment levels of the activating histone mark H3K4me3 in the ERV3–1/ZNF117 promoter region in AML1 by an average of 3.0-fold (1.8-, 2.2-, and 4.9-fold for each amplicon), whereas in ELF no change was detected (average 1.0-fold; 0.8-, 0.9-, and 1.2-fold for each amplicon; Fig. 4D). To validate these results in vivo, transcriptomes of sorted blasts from patients with newly diagnosed AML before and after a 5-day decitabine administration with del(7q) (n = 7) or CN (n = 8) were generated. When focusing the expression analysis on 7q, ZNF117 was the transcript with the strongest statistically significant induction in the del(7q) cohort, whereas in CN no expression change was detectable (median induction 1.3- compared with 1.1-fold, P = 0.0255; Fig. 4E; Supplementary Table S4). qRT-PCR results of sorted blasts from 4 patients of each group confirmed the preferential induction of ERV3–1 and ZNF117 in the del(7q) cohort (median induction 1.3- and 1.0-fold, respectively, in del(7q) compared with 1.1- and 0.8-fold in CN patients; Fig. 4F).
ERV3–1 and ZNF117 are known to be regulated as a single transcriptional unit that does not contain bona fide CpG islands (reviewed in ref. 39). By enrichment of POLR2A-binding sites in AML1 and ELF (Fig. 4A, green boxes, POLR2A tracks from UCSC), as well the activating histone acetylation marks H3K9 and H3K27 in this region in U937 cells, which was enhanced by decitabine, we provided further evidence for the presence of a single promoter in the 5′ region of ERV3–1 (Supplementary Fig. S8A, raw data from Blagitko-Dorfs and colleagues; ref. 20). This was also supported by DNase-seq and ChIP-seq data in HL60 and K562 of RNA-Polymerase II, H3K4me1–3 and, as a silencing mark, H3K9me3 (Supplementary Fig. S8B; data from ENCODE; ref. 40).
Prompted by this finding, we extended the search for decitabine-induced genes by interrogating also multi-copy transposable elements (TE) that are broadly distributed over the entire genome. As already observed for unique protein-coding transcripts, 447 TE were significantly induced in AML1 (FDR < 0.05, Supplementary Fig. S9A) compared with 104 in ELF. Sixty-three TEs were regulated in both cell lines, with long terminal repeat (LTR)-retrotransposons being strongly enriched (56/63, 88.9%, Supplementary Fig. S9B). This was also observed for the top 10 TEs that were exclusively induced only in AML1 or ELF: 10/10 and 8/10 represented LTR-retrotransposons (Supplementary Fig. S9C and S9D). In that regard, LTR12C, shown to be crucial for the expression of neoantigens after decitabine treatment (41), was induced in both AML1 (2.5-fold) and ELF (2.0-fold; Supplementary Fig. S9E).
Given this general induction of LTR-containing TE by decitabine, and thus the potential increase of immunogenicity in the different AML cell lines (by dsRNA and neoantigens), we next analyzed two downstream targets of the antiviral immune response cascade: RIG-I and IRF7 were upregulated 2.3-fold and 1.7-fold, respectively, in AML1, 5.1-fold and 3.5-fold in ELF, indicative of activation of an antiviral response (Fig. 4G). Comparable results were observed for KG-1 and U937 (RIG-I: 2.5- and 2.2-fold induction, IRF7: 3.5- and 3.8-fold, respectively, Supplementary Fig. S10A and S10B), and in AML patient blasts, with a median induction of RIG-I by 1.4-fold in del(7q) patients and 0.7-fold in the CN cohort (Fig. 4H).
Preferential in vitro and in vivo induction of hemizygous tumor suppressor genes on chromosomes 5 and 17
After having demonstrated the preferential upregulation of hemizygous genes by decitabine on chromosome 7 in AML1 and del(7q) patient samples, we extended this unbiased approach to chromosomes 5 and 17, both also often affected in AML/MDS (18).
Therefore, we now used ELF as a model for del(5q) and del(17p), whereas AML1 served as the control because both regions were intact in this cell line (Fig. 1A and B). By focusing our RNA-seq analyses on these two chromosomal loci, we found 39 genes on 5q and 4 genes on 17p to be significantly preferentially upregulated by decitabine in ELF compared with AML1. These included the tumor suppressor genes early growth response gene 1 (EGR1, 5q31.2) and hypermethylated in cancer 1 (HIC1, 17p13.3), with an induction of 7.3- and 15.5-fold in ELF (P = 0.0027 and 0.0019, respectively) compared with 1.8- and 4.2-fold in AML1 (P = 0.0403 and 0.0192; Fig. 5A). In line with this data, genome-wide expression analysis of sorted primary AML patient blasts showed a median induction of EGR1 in del(5q) patients (n = 5) by decitabine of 1.5-fold compared with a -2.5-fold reduction in CN patients (n = 8; P = 0.0314), and a median induction of HIC1 in del(17p) patients (n = 7) of 1.1-fold compared with a -1.1-fold reduction (P = 0.0149; Fig. 5B).
Decitabine improves survival of murine PDXs with MK or CN
Clinically, MK patients respond better to HMAs than to cytarabine (42). To further investigate this observation, we established well-characterized murine PDX models (Supplementary Fig. S11; Supplementary Tables S2 and S3) from 6 AML patients prior to first-line therapy (decitabine or induction) with MK (n = 3) with a median of 2 autosomal monosomies [range 1–6; of chromosomes 4, 6, 7 (2x), 15, 17 (2x), 18, 21], and with CN (n = 3). All patients were wild-type for FLT3 and NPM1; 2 MK and 3 CN patients were wild-type for TP53, and 1 MK patient had mutated TP53. All models were treated with either decitabine, cytarabine, or vehicle.
Across both cytogenetic cohorts, decitabine prolonged median survival by 41 days when compared with vehicle (P = 0.0004), cytarabine prolonged median survival by 18 days compared with vehicle (P = 0.0431; Fig. 6A, left). Thus, survival extension was higher with decitabine compared with cytarabine (P = 0.0383).
Regarding cytogenetic subgroups, decitabine increased median survival of MK-PDX by 40 days (P = 0.0197; range 72–105 days with decitabine, compared with 39–50 days with vehicle; Fig. 6A, middle) and of CN-PDX by 41 days (P = 0.0330, range 71–105 days with decitabine, compared with 43–65 days with vehicle; Fig. 6A, right). Notably, disease progression in two decitabine-treated PDX cohorts (one MK with 6 monosomies, one CN) was delayed such that median survival could not be determined (Fig. 6B and C). Cytarabine increased median survival of MK-PDX by 21 days (not significant, range 46–92 days; Fig. 6A, middle) and of CN-PDX by 14 days (not significant, range 53–75 days; Fig. 6A, right).
Despite the broad and increasing clinical use of HMAs in patients with AML/MDS, these drugs are not well-understood regarding their in vivo mechanism of action. Therefore, it is not surprising that, despite the robust clinical observation of recurrent complete remissions in HMA-treated AML/MDS patients with adverse genetics (i.e., MKs involving chromosomes 7, 5, and 17, with a high prevalence of TP53 mutations), recognition of this peculiar drug activity in a very unfavorable risk group has been slow in coming (3, 43, 44).
To unravel the mechanisms of action underlying this apparent drug selectivity for MKs, we established a system allowing for the investigation of the gene-regulatory effects of the HMAs decitabine and azacitidine on monosomic versus disomic chromosomes. Choosing low-cytotoxic decitabine concentrations, we could study chromatin accessibility and gene induction (and repression) in AML1 cells (a model for monosomy 7), and in the comparator ELF cells (with two intact copies of the long arm of chromosome 7, but with deletions on chromosomes 5 and 17). This choice of cell lines was also guided by their overall lower degree of genetic instability (Fig. 1A and B). Thus, AML1 and ELF may be more representative of primary AML blasts than other cell line models.
Not surprisingly, the overall transcriptional activity on the hemizygous long arm of chromosome 7 in untreated AML1 cells was markedly reduced compared with ELF (Fig. 3A). Decitabine or azacitidine treatment resulted in derepression of a subset of the silenced genes, with “restoration” to expression levels seen in diploid cells, thus supporting our hypothesis that hemizygous genes are more sensitive to decitabine-mediated induction (Supplementary Fig. 12). Three of these genes have known interactions and a role in control of ploidy, apoptosis (ZYX, HIPK2) and DNA methylation (HBP1), and were selected for further validation. Thereby, we could confirm their significant preferential induction when the genes were in a hemizygous state. Concordant results were also observed for the hemizygous tumor suppressor genes EGR1 and HIC1 on chromosomes 5 and 17.
Notably, in AML1 the overall strongest gene derepression on the monosomic long arm of chromosome 7 in response to decitabine was observed for the transcriptional unit of ERV3–1 and ZNF117. Both are coregulated by the ERV3–1 promoter (39). The increase in accessible chromatin sites and therefore induced expression of the massively silenced ERV3–1/ZNF117 by decitabine as well as the induction of various TE is very much in line with the discovery of HMA-reactivated ERVs in several solid tumor models. In two landmark publications, they were uncovered as key mediators of an antitumor immune response by activating RIG-I, MDA5, and MAVS, leading to the secretion of IFN regulatory factors, such as IRF7, resulting in an immune cell–mediated antitumor response (45, 46).
Several experimental strategies were taken to confirm and extend the observed ERV3–1/ZNF117 derepression. First, two additional AML cell lines (KG-1, U937, both with 2 copies of ERV3–1/ZNF117) were interrogated, showing baseline expression levels comparable with ELF, with moderate transcriptional induction following decitabine. Next, primary serially sorted AML blasts (with either del(7q) or CN) of patients receiving decitabine therapy were subjected to qRT-PCR, confirming preferential ERV3–1/ZNF117 induction in the presence of del(7q). Extending this approach to expression arrays of a larger set of matched primary blasts, ZNF117 emerged as the overall highest-induced chromosome 7 gene across the group of patients with del(7q).
Functionally, ERV induction by HMAs should result in activation of downstream targets such as RIG-I and IRF7; indeed, we could demonstrate upregulation of both genes in all 4 AML cell lines and the primary patient blasts, where RIG-I was induced (albeit nonsignificantly). Activation of this cascade is unlikely to be fully explainable by decitabine-mediated activation of a single ERV, especially because ERV3–1 is to date not known to form double-stranded RNA, thus we decided to not limit the ERV expression studies to 7q. Comparing genome-wide decitabine-mediated induction of ERVs and other TE, overall higher levels were attained in AML1 than in ELF, particularly LTR-retrotransposons. These results confirm and extend those generated in primary patient samples with hematologic malignancies and several solid tumor models, that TEs are sensitive to HMA-based induction (45–47).
To rule out that the effects observed were only specific for decitabine, we demonstrated similar cellular effects and the preferential upregulation of hemizygous TSG with the structurally closely related azacitidine. This is indicative of a common mechanism of action by which HMA induced hemizygous genes and raised the question whether the observed gene reactivation events were mediated by promoter demethylation of hypermethylated CpG islands. Indeed, we could observe vast genome-wide demethylation in both cell lines upon decitabine and a significant demethylation of the ERV3–1/ZNF117 promoter region in AML1. In line with this data, we also detected increased abundance of the activating histone mark H3K4me3 at this locus. Hence, we could demonstrate for ERV3–1/ZNF117 a direct correlation between DNA demethylation, increased levels of H3K4me3 and transcript induction.
Potential limitations of this study are the use of only two AML cell lines (plus two for validation) and the lack of a functional genetic model. Furthermore, with only a limited number of primary serially sorted blast isolates amenable to expression analyses by qRT-PCR and microarrays, the degree of induction in patients with del(7q) was not statistically significant for the 3 of the 6 genes analyzed. Nevertheless, we believe that by utilizing these patient blasts to corroborate our findings observed in the cell lines, we could demonstrate the preferential induction of hemizygous TSG by decitabine as it occurs in vivo during HMA treatment.
As to the clinical implications of decitabine activity in cells with large chromosomal deletions, we and others have demonstrated that patients with AML/MDS with adverse genetics can clinically benefit to a degree that is not seen with low-dose cytarabine chemotherapy (42, 46). To address this important difference between these two cytidine analogues, we next compared their antileukemic activity in 3 PDX models representing AML with monosomal karyotype [including del(7q)] and 3 with normal karyotype. It is important to note that the PDXs were generated in immune-deficient mice and therefore could only reproduce immune response-independent antileukemic effects. In this comparison, survival in both groups was longer with decitabine than with cytarabine with both cytogenetic groups benefiting similarly. These results are well in line with randomized clinical trials of decitabine or azacitidine and the comparator treatment with cytarabine (48, 49). However, there is still a clear need for improvement of response rate and duration with HMA therapy in AML. Combination of azacitidine with the BCL-2 inhibitor venetoclax has resulted in impressive response rates also in patients with AML with adverse genetics/TP53 mutations in a randomized trial (50). Also the combination of decitabine with all-trans retinoic acid (ATRA) has resulted in a higher response rate, and has extended survival in a randomized AML trial when compared with single-agent decitabine (19). Notably, here also patients with adverse cytogenetics derived more benefit when decitabine was combined with ATRA, implying an in vivo synergism between both drugs also in the MK subgroup.
In conclusion, we developed and applied an unbiased RNA-seq approach of AML cell lines either mono- or biallelic for 7q, 5q, or 17p, demonstrating that decitabine and azacitidine preferentially upregulate several hemizygous TSG and massively activate the ERV3–1 gene. Induction of ERV3–1 and the dsRNA sensor RIG-I was validated in vivo, and preferentially observed in purified primary AML blasts with del(7q). The in vivo confirmation of the therapeutic superiority of decitabine over cytarabine in PDX murine models of AML with MK should further pave the way for treatment of adverse-cytogenetics AML patients with an HMA-based regimen.
N. Blagitko-Dorfs reports grants from DFG SPP1463 during the conduct of the study. M. Lübbert reports non-financial support from Janssen-Cilag and non-financial support from TEVA during the conduct of the study. No disclosures were reported by the other authors.
G. Greve: Conceptualization, data curation, supervision, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. J. Schüler: Conceptualization, resources, data curation, formal analysis, funding acquisition, validation, methodology, writing-original draft. B.A. Grüning: Resources, data curation, software, supervision, funding acquisition. B. Berberich: Supervision, validation, investigation, methodology. J. Stomper: Formal analysis, validation, investigation, methodology. D. Zimmer: Conceptualization, resources, funding acquisition, validation, investigation, methodology. L. Gutenkunst: Conceptualization, resources, formal analysis, funding acquisition, validation, investigation, visualization, methodology. U. Bönisch: Conceptualization, resources, formal analysis, funding acquisition, validation, investigation, visualization, methodology. R. Meier: Resources, data curation, formal analysis, validation, investigation, visualization. N. Blagitko-Dorfs: Conceptualization, resources, data curation, formal analysis, investigation, methodology. O. Grishina: Conceptualization, resources, data curation, supervision, writing-original draft, project administration, writing-review and editing. D. Pfeifer: Resources, data curation, formal analysis, investigation, methodology. D. Weichenhan: Conceptualization, validation, investigation, methodology. C. Plass: Conceptualization, resources, supervision, funding acquisition, investigation, methodology. M. Lübbert: Conceptualization, resources, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.
We thank Heiko Becker, Thomas Jenuwein, Monika Lachner, Rolf Backofen, Thomas Manke, Pascal Schlosser, Daniel Lipka, Pierre Cauchy, Neftali Ramirez, and Tzu-Ting Wei for continuous helpful discussions and our clinical collaborators for kindly providing patients' blood samples. We thank Milena Pantic, Sina Bengel, Tobias Ma, Ruhtraut Ziegler, Carmen Strittmatter, Eva Oswald, Angela Garding, and Marion Bähr for the technical support provided. This research was funded by the Deutsche Forschungsgemeinschaft - Project ID 192904750 - CRC 992 Medical Epigenetics, SPP1463 LU 429/8-2, FOR2674 A05/A09, the German Cancer Consortium (DKTK, FR01-SOB-AML-LUB), and the German Federal Ministry of Education and Research (DECIDER trial). The Galaxy server that was used for calculations is, in part, funded by Collaborative Research Centre 992 Medical Epigenetics (DFG grant SFB 992/1 2012) and German Federal Ministry of Education and Research [BMBF grants 031 A538A/A538C RBC, 031L0101B/031L0101C de.NBI-epi, 031L0106 de.STAIR (de.NBI)].