Decitabine (DAC) and all-trans retinoic acid (ATRA) disclose cooperative antileukemic activity, inducing transcriptome changes in AML cell lines

Among six AML cell lines (3 with mutated, 3 with wildtype TP53) treated with DAC and ATRA, single-agent activity upon cell growth inhibition was variable (with notable ATRA sensitivity of the three cell lines with MLL rearrangements), without antagonistic effects (Fig. 1A, B). U937 and MOLM-13 were selected for further study due to the strongest drug cooperation, with U937 disclosing a TP53 mutation, MOLM-13 (wildtype TP53) a translocation (9;11) with MLL-AF9 fusion. Combining DAC with ATRA resulted in higher time- dependent growth inhibition and stronger reduction in viability (particularly at 120 h) than either drug alone (Fig. 1C). The observed cytotoxicity induced by the dual treatment was indicative of apoptosis, which was confirmed by significant activation of caspases 3 and 7 at 72 and 120 h (Fig. 1D).

Fig. 1: Decitabine and all-trans retinoic acid exhibit cooperative effects on inhibition of proliferation and reduction of viability in AML cell lines.

A Treatment scheme. B Baseline-corrected proliferation of OCI-AML3, MOLM-13, MV4-11, U937, THP1, and HL-60 cells treated with DAC (100 nM) and ATRA (1 µM) after 120 h. Experiments were conducted in technical triplicates. C Proliferation and viability of U937 and MOLM-13 cells were measured 72 and 120 h after the first dose of Decitabine, untreated or treated with DMSO, DAC (U937: 200 nM; MOLM-13: 100 nM), ATRA (1 µM) or DAC + ATRA in combination. Three independent experiments, each with three independent technical replicates, were conducted. Standard deviation is shown as error bars. D Caspase-3/7 activity in U937 (left panel) and MOLM-13 cells (right panel) displayed as fold-change relative to a DMSO-treated control at 72 and 120 h. Six (72 h) and nine (120 h) independent experiments, each with three independent technical replicates, were conducted, respectively. Standard deviation is shown as error bars. Statistical significance was tested for by unpaired t-test with a threshold of p < 0.05.

To investigate early transcriptome changes induced by these treatments, i.e., before the onset of secondary effects, U937 and MOLM-13 were analyzed by RNA-sequencing at the 72 h time point. U937 revealed DAC sensitivity, with the majority of transcripts being upregulated, whereas the effect of ATRA was limited; the combination resulted in an additive effect. MOLM-13 cells disclosed a limited effect of DAC, but were quite sensitive to ATRA, also with an additive effect of the combination (Fig. 2A). Notably, approximately one-third of the transcripts regulated by DAC + ATRA in each cell line was shared, i.e., 1251 commonly regulated genes (Fig. 2B), with 902/1251 transcripts (72.1%) in the same direction. Co-regulated transcripts are depicted by unsupervised clustering (Fig. 2C). As expected, more genes were induced in U937 compared to MOLM-13 (and vice versa); GO analysis unveiled a marked enrichment for upregulated transcripts involved in leukocyte differentiation and immune response, and downregulated transcripts associated with translational initiation (Fig. 2D).

Fig. 2: Decitabine and all-trans retinoic acid cooperate in inducing transcriptome changes in AML cells.

A Global expression changes in U937 and MOLM-13 were measured by RNA-seq (FDR < 0.01). Cells were treated with DAC, ATRA or both in combination, and harvested 72 h after the first dose of Decitabine. Bar and Venn diagrams depict the numbers of altered transcripts in comparison to untreated; upregulated transcripts are shown in red, downregulated in blue. B Comparison of significantly regulated genes in U937 and MOLM-13 treated with DAC + ATRA vs. untreated (determined by RNA-seq). C Heatmaps (unsupervised clustering) of the commonly regulated genes in U937 or MOLM-13, comparing untreated (untr.) to DAC + ATRA-treated cells. D GO analysis for the commonly up- or downregulated genes in U937 and MOLM-13 by DAC + ATRA.

Synergistic transcriptional induction of retinoic-acid responsive genes by DAC and ATRA is associated with increased chromatin accessibility

We hypothesized that single-agent DAC, ATRA or the combination of both may alter chromatin accessibility [14], and thus performed ATAC-seq on U937 cells. As shown in Fig. 3A, globally, DAC alone had a much stronger effect on chromatin accessibility than ATRA alone.

Fig. 3: Cooperation between DAC and ATRA in modulating chromatin accessibility.

A Enrichment of accessible chromatin sites across the whole genome determined by ATAC-sequencing in U937 cells (technical triplicates) either untreated (untr.) or treated with DMSO, DAC, ATRA or DAC + ATRA at 72 h. Sequencing was performed with 40 million reads per sample (paired-ends). Mean alignment was at about 99%, with 51–63% uniquely mapped sequences. B Principal component (PC) analysis (PCA) of the assay for transposase-accessible chromatin sequencing (ATAC-seq) data of control (untreated, CNTL; n = 3), ATRA (n = 3), DAC (n = 3), and DAC + ATRA-treated (n = 3) samples. Chromatin accessibility of the 200,000 most variably accessible peaks was used for dimensionality reduction. C Heatmap of the 50,000 most variable differentially accessible regions (DARs) from the ATRA vs. untreated (CNTL), DAC vs. CNTL, and ATRA + DAC vs. CNTL comparison. Euclidean distance of the z-scaled chromatin accessibility of DARs in all samples was visualized. D Homer transcription factor motif enrichment of differentially accessible regions (adj. p-value < 0.05 and absolute log 2 fold-change >1), stratified in open and closed DARs. The top five most significantly enriched transcription factor motifs as well as all significantly enriched (adjusted p-value < 0.05) retinoic acid response (RAR) motifs are visualized. The union set of all peaks was used as a background for enrichment analyses.

However, when focusing on the 200,000 most variably accessible peaks for dimensionality reduction, a principal component analysis (Fig. 3B) revealed a distinct grouping for each treatment, clearly separating DAC- from non-DAC-treated (PC1, explaining 63% of the variability) and ATRA- from not-ATRA-treated samples (PC2; explaining 27% of the variability). These results indicate a global effect of both DAC and ATRA, as single agents, on chromatin accessibility, as confirmed in a differential accessibility analysis (Suppl. Fig. 2). Visualization of the 50,000 most variable differentially accessible regions (Fig. 3C) revealed regions of chromatin opening induced by DAC but not ATRA and, in a smaller population, by ATRA but not DAC; for a subset of peaks, both drugs cooperated in chromatin opening. Regarding chromatin closing, activity was noted not only for DAC, but also for ATRA, and here also, both drugs showed cooperative activity for a substantial number of regions.

Next, we asked whether increased chromatin accessibility either by single-agent treatment with DAC or ATRA, or by combination treatment, was associated with transcription factor (TF) and retinoic acid response (RAR) motifs. Indeed, TF motif analysis (Fig. 3D) revealed a drug-specific enrichment, with chromatin opening by ATRA, but not DAC, for IRF1, -2, -3, -8 and ISRE (middle panel), whereas with DAC, the 5 TF motifs with the highest degree of chromatin opening included GATA1, -2, -4, -6, and ELK4 (left panel). Upon dual treatment, the ATRA signature for chromatin opening remained stable (right panel). Visualizing drug effects on chromatin opening at retinoic acid response elements (RAREs), single-agent ATRA, but not DAC, resulted in chromatin opening at the RAR/RXR, RARA and RARG motifs; here also, the ATRA signature could also be discerned with the combined treatment.

Next, we selected individual genes bearing RAREs to interrogate regions of differential chromatin accessibility. Three genes with known RAREs and CpG-rich regions (HIC1, CYP26A and GBP4) were selected from the genes disclosing the highest levels of induction by DAC + ATRA. HIC1 is a well-established tumor suppressor gene [15], with ATRA-responsiveness previously demonstrated in AML [16], CYP26A1 is central in ATRA metabolism [17], and GBP4 represents an interferon-response gene recently described as an ATRA-regulated target of IRF1 [18]. In addition, we also interrogated LYZ, a DAC-sensitive gene [19] (also with an RARE, but lacking a CpG island) that others as well as ourselves described as epigenetically regulated during myelopoiesis [20], with DAC-induced chromatin opening [21].

As shown in Figs. 4 and 5, the dual treatment resulted in synergistic transcriptional upregulation of all four genes, with increased chromatin accessibility compared to untreated cells. Single-agent treatment with either DAC or ATRA resulted in variable induction as well as chromatin opening. This is exemplified by a 5-fold induction of LYZ mRNA by DAC, with concomitant broadening of the ATAC-seq peak around LYZ exon 1 (Fig. 5B). Across all four genes, the seven peaks representing significantly treatment-induced higher chromatin accessibility are marked by blue boxes beneat the ATAC-seq tracks; the majority of regions with treatment-induced increased chromatin accessibility co-localized with Pol2 peaks, RARA, RXRA and CpG islands. This is in line with ATRA acting within a transcriptional activator complex, directly involved in initiation of transcription, in these genes selected for known retinoic acid responsiveness.

Fig. 4: Transcriptional induction of HIC1 and CYP26A1 by DAC and ATRA is associated with changes in chromatin accessibility.

A, B mRNA expression (normalized read-counts) of HIC1 (A) and CYP26A1 (B) of both untreated U937 cells and U937 after treatment with either DAC, ATRA, or DAC + ATRA at 72 h was determined by RNA-seq (left panel). Chromatin accessibility of these cells and treatment modes was determined by ATAC-seq (upper right panel). Blue boxes below the ATAC-seq track represent significant effects. Annotations for POLR2A binding sites (green boxes), retinoic acid receptors RARA and RXRA (brown boxes) and CpG islands (black boxes) were extracted from ChiP-seq data accessed from ENCODE and GEO (right panel). CpG methylation data of U937 cells untreated or treated with DAC were determined by Infinium HumanMethylation450 BeadChip Assay [8] (lower right panel).

Fig. 5: Transcriptional induction of GBP4 and LYZ by DAC and ATRA is associated with changes in chromatin accessibility.

A, B mRNA expression (normalized read-counts) of GBP4 (A) and LYZ (B) of both untreated U937 cells and U937 after treatment with either DAC, ATRA, or DAC + ATRA at 72 h was determined by RNA-seq (left panel). Chromatin accessibility of these cells and treatment modes was determined by ATAC-seq (upper right panel). Blue boxes below the ATAC-seq track represent significant effects. Annotations for POLR2A binding sites (green boxes), retinoic acid receptors RARA and RXRA (brown boxes) and CpG islands (black boxes) were extracted from ChiP-seq data accessed from ENCODE and GEO (right panel). CpG methylation data of U937 cells untreated or treated with DAC were determined by Infinium HumanMethylation450 BeadChip Assay [8] (lower right panel).

Three of these four genes have CpG islands of variable size, with HIC1 disclosing the by far largest island (composed of 810 CpGs, compared to 160 CpGs in the CYP26A1 promoter and 5’ coding region), GBP4 also exhibiting a CpG-dense promoter region (albeit not fulfilling CpG island criteria), LYZ lacking a bona fide CpG island [21]. We asked whether U937 cells treated with DAC disclose DNA hypomethylation at these regions, analyzing Illumina array data [8]. Indeed, demethylation was observed for all four genes (Figs. 4 and 5, lower panels), co-localizing with regions of increased chromatin accessibility.

Global DNA demethylation by decitabine is not enhanced by all-trans retinoic acid

Given evidence in the literature that ATRA may also act by inhibiting DNA methyltransferase activity, we wished to pursue this hypothesis using a methylome-wide approach. Methylation in U937 was analyzed by Infinium HumanMethylation450 BeadChip Array [22] at 72 und 120 h (Fig. 6A). The massive demethylating effect in gene bodies and promoters of the DNMT inhibitor DAC could be detected both for the single-agent treatment as well as when combined with ATRA. Correlation-based clustering revealed a subclustering, in the DAC/DAC + ATRA arm, for time point of harvest (72 and 120 h) with one outlier (DAC sample harvested at 72 h) in the 120 h group.

Fig. 6: Global DNA demethylation by DAC is not enhanced by ATRA.

A U937 cells (technical triplicates) were treated with either DAC (lilac), ATRA (green), DAC + ATRA (orange) or left untreated (untr.; pink) and analyzed after 72 and 120 h. Global DNA methylation was acquired by Infinium HumanMethylation450 BeadChip Assay [8]. Methylation data was analyzed with RnBeads as described in Materials and Methods and Suppl. Methods. The heatmap displays the methylation percentiles (%) for all 24 samples analyzed. Unsupervised, hierarchical clustering showed subclustering in the DAC/DAC + ATRA arm for the two time points with an outlier in the 120 h group. B LINE-1 methylation of three selected CpGs was determined by MassArray in U937 cells (technical triplicates) treated with DAC, ATRA or DAC + ATRA at 72 h.

Global DNA methylation can also be detected by the methylation status of transposable repetitive LINE elements. These make up 21% of the human genome and are highly methylated. These properties make them an ideal surrogate for genome-wide methylation status. Focusing on three informative CpGs within the LINE-1 promoter, we quantified methylation by MassArray [23]. At 72 h, untreated and ATRA-treated cells showed a comparable degree of methylation, whereas DAC-only- and DAC + ATRA-treated cells showed similar, marked demethylation across all three CpGs compared to untreated cells (Fig. 6B). Thus, we conclude that also by this approach, no activity of ATRA in demethylating DNA could be discerned.

Gene induction in AML peripheral blood blasts during treatment with DAC and ATRA

Observing cooperative, in part synergistic effects of DAC and ATRA on gene transcript expression, we considered that this may be observed, at least to some degree, also in vivo. Peripheral blood blasts from nine AML patients treated with DAC alone for 5 days, and six patients treated with DAC followed by ATRA initiated on day 6 (DECIDER trial [24]) were interrogated by cDNA array analyses. TP53 was mutated in three patients treated with DAC alone and none of the patients receiving DAC + ATRA (Suppl. Table 3). Therefore, comparability of TP53 mutation status was limited for this cohort.

The two time points compared for changes in transcript expression were: i. prior to and ii. at day +8 from treatment start (48 h after initiation of ATRA). As shown in Fig. 7A, 157 genes regulated by DAC + ATRA were shared by all six patients and the two AML cell lines (see Fig. 7B for GO analyses). For HIC1, GBP4 and LYZ, median mRNA expression changes at day 8 tended to be higher in patients receiving the dual treatment compared to those receiving single-agent DAC (Fig. 7C) with a statistically significant difference reached for LYZ, whereas no difference was observed for CYP26A1 (not shown).

Fig. 7: Changes in primary blasts of AML patients treated with decitabine and all-trans retinoic acid.

A Venn-diagram of transcriptionally regulated genes in U937 and MOLM-13 cells (determined by RNA-seq) and AML patients (n = 6) treated with DAC + ATRA (primary AML blasts; expression arrays performed on day 0 and day 8, compared to untreated). B GO analysis of up- or downregulated genes, in U937 and MOLM-13, and primary AML blasts. C Scatter plots of HIC1, GBP4 and LYZ expression changes in patient samples (day 8 vs. day 0) having received DAC only (n = 9) or DAC + ATRA (n = 6). Fold-changes are displayed. Statistical significance was tested for by unpaired t-test with a threshold of p < 0,05 (ns, not significant). D Scatter plots displaying expression changes in CTSG, ELANE, IL7R and CXCR1 as described in Fig. 7C. E Scatter plot of MYC expression changes (downregulated by DAC + ATRA compared to DAC alone) as described in Fig. 7C. F Levels of fetal hemoglobin (HbF,%) in AML patients’ peripheral blood erythrocytes before treatment (baseline) and after 2 courses (p.c.2) of treatment with either DAC only (n = 6) or DAC + ATRA (n = 5) were determined by HPLC as described in Materials and Methods.

As shown in Fig. 7D, significantly higher fold-changes (FCs) with DAC + ATRA treatment compared to DAC alone were also noted for several other genes with important roles during myeloid differentiation: cathepsin G (CTSG; mean FC DAC + ATRA 1.92 vs. mean FC DAC 1.09; Suppl. Table 4), neutrophil elastase (ELANE; mean FC DAC + ATRA 1.73 vs. mean FC DAC 1.02; Suppl. Table 4), the IL7 receptor (IL7R; mean FC DAC + ATRA 1.11 vs. mean FC DAC 1.02; Suppl. Table 4) and the interleukin-8 receptor CXCR1 (CXCR1; mean FC DAC + ATRA 1.03 vs. mean FC DAC 0.97). On the other hand, numerous genes were downregulated by DAC + ATRA (Suppl. Table 5), including MYC (mean FC DAC + ATRA 0.79 vs. mean FC DAC 1.45; Fig. 7E and Suppl. Table 5).

In vivo induction of fetal hemoglobin by continued treatment with DAC and ATRA

Since these patients had received only limited (48 h) ATRA exposure, we asked whether upregulated fetal hemoglobin (HbF), one of the few biomarkers for HMA response [25], would undergo more pronounced kinetics of gene induction than observed after 48 h of ATRA. HbF was serially measured before treatment start and after 2 treatment courses (i.e., after 46 days of ATRA intake). As shown in Fig. 7F, the median HbF protein levels measured in patients’ peripheral blood erythrocytes were not altered for six patients not receiving ATRA as an add-on to DAC (0.9 vs. 0.9%) and rose from 0.7 to 2.0% in five patients receiving DAC + ATRA (paired t-test p = 0.0632, not significant).

Decitabine and all-trans retinoic acid cooperate in derepressing transposable elements in AML cells, activating dsRNA sensing

The ability of HMAs to reactivate transposable elements (TEs) in cancer cells, thereby triggering antitumoral responses (“viral mimicry”), has been demonstrated in several recent publications [24, 26,27,28]. Given that repetitive elements may also possess RAREs, we hypothesized that ATRA (alone and in combination with DAC) may also reactivate TE expression [29]. Analysis of U937 and MOLM-13 for TE expression [24] revealed that the total number of differentially expressed TEs (Fig. 8A) mirrored the global effects on transcription shown in Fig. 2A, with U937 disclosing predominant sensitivity to single-agent DAC but not ATRA, MOLM-13 with predominant sensitivity to single-agent ATRA but not DAC. Combining both drugs resulted in synergistic effects in both cell lines, with a balanced ratio of up- vs. downregulated TEs. For U937, TEs also regulated in MOLM-13 amounted to approximately one-third of all sequences, and two-thirds of TEs regulated in MOLM-13 overlapped with those regulated in U937 with concordant up- or downregulation for most of them (Fig. 8B). In U937, the 10 most strongly induced TEs surpassed the 10 most strongly repressed TEs by amplitude, whereas in MOLM-13 cells, the fold-changes (overall lower than in U937) of induction vs. repression were in a similar range (Fig. 8C).

Fig. 8: Decitabine and ATRA cooperate in derepressing transposable elements in AML cells, activating dsRNA sensing.

A Global expression changes of TEs in U937 and MOLM-13 were measured by RNA-seq (FDR < 0.05). Cells were treated with DAC, ATRA or both in combination and harvested 72 h after first dose of Decitabine. Bar and Venn diagrams show the numbers of altered TEs for each treatment (DAC, ATRA, DAC + ATRA) in comparison to untreated; upregulated transcripts shown in red, downregulated in blue. B Comparison of regulated TEs in U937 and MOLM-13 treated with DAC + ATRA vs. untreated determined by RNA-seq. 102 out of 207 TE (49.3%) transcripts show concordance regarding induction vs. repression, as depicted in the lower graph. C Waterfall plots of the 10 most up- or downregulated TEs in U937 or MOLM-13 treated with DAC + ATRA. D Immunoblot results for RIG-I, MDA5 and MAVS in whole-cell lysates of U937 and MOLM-13 treated as indicated above. Beta-actin was used as loading control; the space between protein bands was cropped to conserve space. Results for the poly(I:C) sample in MOLM-13, repeated due to technical issues, were replaced.

As we have recently demonstrated co-regulated induction of the endogenous retrovirus ERV3-1 and the zinc finger protein ZNF117 in different AML cell lines including U937 by DAC [24], it was of interest to interrogate these two genes also for the effect of ATRA. We observed that ERV3-1 RNA was induced about threefold at two different time points by ATRA alone, and expression was further induced by the DAC + ATRA combination treatment. As expected, similar kinetics of induction were noted for ZNF117 (Suppl. Fig. 3A).

Since dsRNAs induced by epigenetic treatment trigger an antiviral immune response via binding to RIG-I or MDA5 and the complex then binding to the dsRNA sensor MAVS at the mitochondrium (Suppl. Fig. 3B), we investigated expression and interaction of these three components of the antiviral cellular response machinery. In U937, RIG-I mRNA (data not shown) and protein were readily induced by DAC and modestly by ATRA (Fig. 8D, left panel, poly(I:C) was used as a positive control). The drug combination led to further induction. In MOLM-13 cells, ATRA was much more effective than DAC in inducing RIG-I protein (Fig. 8D, right panel). Regarding MDA5 protein levels, induction was modest for both cell lines. MAVS protein was abundant in untreated cells; levels were unaffected in U937 cells but were clearly induced in MOLM-13 (ATRA, DAC + ATRA). The protein-protein interaction, interrogated by co-immunoprecipitation, remained stable under the different treatment conditions (not shown). Thus, we conclude that a “viral mimicry” response, a recently uncovered antineoplastic mechanism triggered by HMAs, can be enhanced by ATRA (possibly dependent on the TP53 status of the leukemic cells) [30, 31].