The type of DNA damage response after decitabine treatment depends on the level of DNMT activity

mESCs as a model system to study the mode of action of AzadC and AzaC

Before we started to investigate the effects of AzadC and AzaC using our anticipated model system, we checked whether wt and the DNMT-TKO cells provided similar cellular features, for example, distinct morphological changes upon priming, despite the absence of all three DNMT enzymes in the DNMT-TKO. Upon priming, a comparable number of proteins showed significant expression-level changes from the naïve to the primed state for both genotypes (Fig 1B). The differently expressed proteins between the naïve and the primed state differed, however, substantially between the two genotypes. Nevertheless, both genotypes underwent similar morphological changes. When cultured in 2iL, both wt and DNMT-TKO cells formed densely packed colonies with distinct borders. Within the colonies, individual cells could not be spotted (Fig 1C, naïve). Upon priming, the colonies dissected and individual cells with distinct morphology were increasingly observed for both genotypes (Fig 1C, primed). Moreover, in the wt, Dnmt1, Dnmt3a, and Dnmt3b were all significantly more expressed in the 96-h primed state compared with the naïve state (Fig 1D), which consequently resulted in higher DNMT activity as indicated by increasing levels of 5mdC (Fig 1E). Because new 5mdC patterns are established during priming, wt mESCs allow to monitor closely the effects of AzadC and AzaC when DNMT activity is highly dynamic in cells with the same genetic background. Altogether, these results indicated that the wt and the DNMT-TKO mESCs undergo fundamental cellular changes from the naïve to the 96-h primed state. Although the individual changes on the proteome level were different between the two genotypes, the extent and the resulting morphological changes were similar, and therefore, we decided that the mESCs provided a suitable model system to directly compare the effects of AzadC and AzaC in the presence and absence of DNMT enzymes.

Presence of genomic AzadC substantially contributes to the anti-proliferative and cytotoxic effect without formation of DNMT-DNA crosslinks when overall DNMT activity is low

To investigate how much the presence of genomic AzadC contributes to its toxicity by base mismatch (MoA 3) or formation of abasic sites by spontaneous hydrolysis of the base (MoA 4), independent from DNA hypomethylation (MoA 1) and DNA damage by DNMT crosslinking (MoA 2), we first compared the effects of AzadC on wt and DNMT-TKO cells under naïve conditions (2iL) when DNMT expression and activity are low in the wt (Fig 1D and E). First, we confirmed by our previously reported QQQ-MS method for exact quantification of nucleosides (34) that AzadC had the anticipated hypomethylating effect in the wt even under low DNMT activity (Fig 2A). We detected a significant decrease in the amount of mdC after 48-h treatment with 1.25 or 2.5 μM of AzadC with no difference between the concentrations, indicating that the hypomethylating effect was already at the maximum at the concentrations applied. Because the mdC levels of the DNMT-TKO were under the limit of detection even without treatment (Fig S1D), a DNA hypomethylating effect could not be observed upon treatment. Even though DNA demethylation was already at the maximum at 1.25 μM AzadC in the wt, inspection of the phenotypic changes of the cells by brightfield microscopy revealed that the anti-proliferative and cytotoxic effect of AzadC was strictly concentration-dependent with only mild effects at lower concentrations (Fig 2B). This result was in line with previous reports that AzadC-induced DNA hypomethylation already occurs at low concentrations before inducing cytotoxic effects (9, 35). Interestingly, the DNMT-TKO showed a very similar cell viability pattern in the brightfield microscopy images, albeit starting at higher concentrations (Fig 2B). We confirmed this observation in an independent flow cytometry–based apoptosis assay, where we observed a concentration-dependent increase in apoptotic events for both genotypes (Fig 2C). Compared with the respective untreated controls, the increase in apoptotic events was significant from 2.5 μM of AzadC on in the wt and from 5 μM of AzadC on in the DNMT-TKO (Fig 2D). Next, we compared the proliferation rate of the wt and the DNMT-TKO. Untreated controls showed an identical proliferation rate with a doubling time of 17–19 h. Treatment with AzadC slowed down proliferation in both genotypes with a higher impact on the wt (Fig S2A). However, proliferation rates after treatment were overall within the same range for both genotypes, and after treatment with 2.5 μM AzadC, proliferation was slowed down in both genotypes to the same extent with a doubling time of ∼26 h (Fig 2E).

The similarity of phenotypic (Fig 2B–D) and proliferative (Figs 2E and S2A) changes in the wt and DNMT-TKO after 48-h treatment with AzadC was unexpected because AzadC cannot induce deleterious DNMT-DNA crosslinks in the DNMT-TKO and should therefore have a substantially lower impact in the DNMT-TKO on cellular well-being. There were two possible options to explain those results. Either AzadC was metabolized differently in the wt and the DNMT-TKO, with higher uptake and genomic incorporation in the DNMT-TKO, or AzadC itself without formation of DNMT-DNA crosslinks contributed substantially to the anti-proliferative and cytotoxic effects at the concentrations applied. To test which option explained our results, we checked AzadC uptake and genomic incorporation in the wt and the DNMT-TKO by treating both genotypes with 2.5 μM of AzadC for 24 and 48 h and quantifying the global, sequence context-independent, genomic incorporation levels of AzadC by QQQ-MS (Fig 2F). For the 24-h timepoint, we added 2.5 μM of AzadC once at the start (0 h) and harvested the cells after 24 h (treatment a). For the 48-h timepoint, we treated the cells for 48 h before harvest, but tested three different treatment regimens—addition of 2.5 μM of AzadC at 0 h and neither medium change nor second compound addition (treatment b); addition of 2.5 μM of AzadC at 0 h and medium change after 24 h but no second compound addition (treatment c); and addition of 2.5 μM of AzadC at 0 h and medium change after 24 h with second compound addition of 2.5 μM of AzadC (treatment d). The untreated control served as a background control, and as expected, an AzadC signal was not detectable neither in the wt nor in the DNMT-TKO (Fig S2B). For all timepoints and treatment regimens tested, we did not observe any difference between the wt and the DNMT-TKO (Fig 2F), indicating that there was no difference between the two genotypes regarding AzadC uptake and metabolization. However, we detected substantial differences between the different treatment regimens. After 24 h, the amount of genomically incorporated AzadC, normalized to the amount of dG, was about two times higher compared with the 48-h timepoint when no additional compound was added. In contrast, when AzadC was added twice (after 0 h and in addition after 24 h), we detected an additional increase in genomic AzadC after 48 h compared with the 24-h timepoint. These results suggested that 24 h after addition, all available AzadC had been either incorporated or inactivated by hydrolysis or removal from the genome. Taken into account that the proliferation rates of the wt and the DNMT-TKO cells were close to 24 h when exposed to 2.5 μM of AzadC, these results implied that removal of genomically incorporated AzadC mostly depended on the global scale on passive dilution by cell division (Fig 2G), whereas removal by repair mechanisms or spontaneous decay only played a minor role. In consequence, to avoid passive dilution and to increase the amount of genomically incorporated AzadC, the compound has to be added continuously (Fig 2H).

To also investigate the formation of DNA damage in the wt and the DNMT-TKO after AzadC treatment in a dose-dependent manner, we performed an immunoblot analysis against γH2AX (Figs 2I and S2C). H2AX is a histone variant, which is placed as a mark at sites of DNA double-strand breaks (DSBs) and is subsequently phosphorylated at Ser-139 (γH2AX) to recruit the repair machinery. Because one γH2AX is placed per DSB, it is a very sensitive and quantitative marker for DSB formation (36). As expected, we detected only a very faint γH2AX signal in the untreated controls of both genotypes, whereas AzadC treatment with all concentrations tested (1.25–5 μM) resulted in comparable formation of DSBs in the wt. In the DNMT-TKO, we observed strict dose-dependent formation of DSBs with no detectable signal after 1.25 μM AzadC treatment, but a clearly visible signal at the higher concentrations. In parallel, we detected the levels of Parp1, which serves not only as a marker for DNA damage but also induction of apoptosis. Parp1 can be cleaved in a 24-kD fragment, which remains at the lesion, and an 89-kD fragment that is released in the cytosol to induce apoptosis (37). The total Parp1 levels did not differ between the wt and the DNMT-TKO, but with slightly lower levels in the untreated wt (Figs 2I and S2C). In contrast, the 89-kD cleaved fragment showed only a strong signal in the wt, already at a concentration of 1.25 μM of AzadC, but not in the DNMT-TKO at any concentrations tested.

In summary, our results suggest that genomically incorporated AzadC is mainly removed by passive dilution during cell division unless a new substance is provided at least every 24 h. AzadC already exhibits an intrinsic anti-proliferative and apoptotic effect independent from DNMT crosslinking, but dependent on active DNA replication. After 24 h, when AzadC concentration was at a maximum (Fig 2D), the cells had undergone one replication cycle, and already a significant reduction of mdC (12), but apoptosis was only induced to a very low extent (Fig S2D). This result was in accordance with previous studies showing that the cytotoxic potential of AzadC unravels during the second DNA replication cycle after incorporation (26).

When DNMT activity is high, DNA-DNMT crosslinking dominates the efficacy profile of AzadC

Upon priming, wt mESCs undergo reprogramming, including changing and overall increasing DNA methylation patterns in the wt (30, 31) as a result of higher expression of all three DNMT enzymes and higher DNMT activity (Fig 1D and E). Therefore, we expected increased sensitivity of the wt towards AzadC treatment. Whereas AzadC-induced DNA lesions that do not originate from DNA-DNMT crosslinking are strictly dose-dependent as they only depend on the amount of incorporated AzadC, DNA lesions from DNA-DNMT crosslinking depend on the amount of incorporated AzadC but even more on the activity of DNMT enzymes. With increasing amounts of these crosslinks, genome instability is multiplied. Consequently, when DNMT activity is high, a lower amount of genomically incorporated AzadC can nevertheless lead to more devastating effects than a substantially higher amount of AzadC under conditions of low DNMT activity. In line with these preliminary considerations, we did not observe any proliferation of AzadC-treated wt cells under Lif conditions (Fig S3A). Furthermore, brightfield images indicated that under Lif conditions, massive cell death already occurred in the wt when the cells were treated with as little as 0.1 μM of AzadC (Fig 3A). The drastic increase in dead cells after low-dose AzadC treatment was in addition confirmed by the flow cytometry–based apoptosis and cell death assay (Fig 3B). In contrast, the DNMT-TKO did not respond equally sensitive (Figs 3C and S3A). However, primed DNMT-TKO mESCs were also more sensitive towards low concentrations of AzadC (Fig 3D) than the ones under naïve conditions, which did not respond to an AzadC concentration below 2.5 μM (Fig 2B). Immunofluorescence analysis of γH2AX in AzadC-treated DNMT-TKO cells under naïve and primed conditions suggested that DNA damage was already present in the primed state at lower concentrations of AzadC (Fig S3B), but the underlying reason was not apparent. Nevertheless, apoptotic events in the wt compared with DNMT-TKO after AzadC treatment were disproportionately higher in the primed state compared with the naïve one (Fig 3E). Overall, these results suggested that the DNMT-dependent effects are dominant for the toxicity profile of AzadC when DNMT activity is high and DNMT-independent DNA lesions only play a subordinate role.

The activation of specific DNA repair pathways as a response to AzadC-induced DNA lesions depends on DNMT activity

To investigate the DDR towards AzadC in a systematic and holistic way, we treated wt and DNMT-TKO cells with AzadC under Lif and 2iL conditions. Afterwards, we divided the cells into two portions—one to enrich the chromatin fraction to check for recruitment of DNA repair–associated proteins to the site of action, and one to isolate the whole proteome to check for their expression-level changes. Subsequently, the isolated proteins were subjected to an established SP3 workflow and analysed by LC-MS/MS (Fig 3F). To validate our workflow, we first analysed the detected proteins in the chromatin fraction against the whole proteome and vice versa. GO-term analysis revealed that in the chromatin fraction, GO terms associated with regulation of (DNA-templated) transcription (e.g., GO:0006355, GO:0006357), DNA binding (e.g., GO:0003677, GO:0043565, GO:0000976), and DNA-binding transcription factor activities (e.g., GO:0003700, GO:0000981, GO:0001228, GO:0001227) were highly overrepresented compared with the whole proteome (Fig S3C, Supplemental Data 2). On the contrary, the whole proteome showed massive enrichment for all cellular compartments except the nucleus when compared to the proteins of the chromatin fraction (Fig 3G). The comparison in both directions confirmed that the prepared chromatin-associated proteome was as anticipated as an overrepresentation of nuclear- and chromatin-bound proteins in comparison with the total cellular proteome.

Next, we investigated the differences of the chromatin-associated proteome between the untreated and the AzadC-treated wt cells under Lif conditions with a focus on enrichment changes of proteins that are directly involved in DNA repair as assigned by the Reactome pathway knowledge base (R-HSA-73894.5) (38). Upon treatment, we observed an overall massive change of the chromatin-associated proteins. Among the significantly deregulated chromatin-associated proteins, 23 DNA repair–associated proteins were enriched after AzadC treatment and 11 DNA repair–associated proteins were depleted compared with the chromatin fraction of the untreated control (Fig 3H, Supplemental Data 3, Table S1). A highly enriched DNA repair–associated protein was Parp1, which was already reported to be involved in the repair of AzadC-induced DNA lesions via base excision repair (BER) (27) and now confirmed by our complementary approach. Another top enriched hit was the flap endonuclease 1 (Fen1) that belongs to the structure-specific endonucleases (39). Fen1 is important for the maturation of Okazaki fragments during DNA replication, but it is also required to deal with DNA replication stress caused by a stalled replication fork and involved in long-patch BER (40) and (microhomology-mediated) alternative end joining (a-EJ) (41, 42). A-EJ is considered as an alternative NHEJ pathway that results in deletions. Therefore, this error-prone repair pathway is a mutagenic driver, but it is activated when NHEJ and homologous recombination (HR) repair pathways are compromised (43).

Other important repair proteins that we found significantly enriched in the chromatin fraction of AzadC-treated wt cells under Lif conditions were the DNA-dependent protein kinase catalytic subunit Prkdc, the DNA mismatch repair proteins Msh2 and Msh6, the single-strand selective monofunctional uracil DNA glycosylase Smug1, and the DNA repair protein Rad50. Prkdc regulates phosphorylation of H2AX and is one of the key proteins in NHEJ (44, 45). Msh2 and Msh6 are central components of the post-replicative DNA mismatch repair system (MMR) by forming the heterodimer MutSα that binds to DNA mismatches to initiate repair (46). Rad50 on the contrary plays a pivotal role in DSB repair (47). Interestingly, the uracil DNA glycosylase Ung, which belongs to the same protein superfamily as Smug1, was significantly depleted from the chromatin fraction of AzadC-treated wt cells (Fig 3H). Smug1 and Ung are both part of the BER machinery to repair the presence of uracil in the DNA after deamination of cytosine. However, although Ung is important for repair of uracil in replicating DNA, Smug1 is more important to remove uracil in non-replicating chromatin (48). Among the depleted proteins on the contrary were the E3 ubiquitin-protein ligase Rad18, which is an essential component to link DNA damage signalling to activation of HR (49), and the DNA repair protein complementing XP-C cell homolog Xpc, which is an integral component of the nucleotide excision repair machinery (NER) (50).

To obtain a holistic picture of the affected cellular pathways, with a focus on the involved DNA repair pathways, we performed a pathway enrichment analysis using pathfindR (51) (Fig 3I, Supplemental Data 3, Table S2). One of the top enriched terms was “Proteasome” with more than eightfold enrichment and 23 contributing proteins being highly enriched in the chromatin fraction of the AzadC-treated wt cells compared with untreated cells (Table 2). Protein–DNA crosslinks can be repaired by proteasomal activity that digests the bulky protein adducts to the peptide level to allow the canonical repair machinery to excise the affected DNA sites in a next step (52, 53). Our chromatin proteomics data strongly indicated that the proteasome removes the covalently trapped DNMTs, whereas other proteases involved in removal of protein–DNA crosslinks like Spartan could not be detected in the chromatin fraction. As expected, pathway enrichment analysis also revealed that “DNA replication” and “cell cycle” were heavily affected after AzadC treatment as many checkpoint proteins were significantly enriched in the chromatin fraction after AzadC treatment. Among them was the serine/threonine-protein kinase Chek1, which is central for checkpoint-mediated cell-cycle arrest and the activation of the DDR when DNA damage is present (54), as well as all components of the mini-chromosome maintenance (MCM) complex (Mcm2–Mcm7) (Table 2). MCM proteins were reported to play an important role in sensing DNA damage at replication forks and subsequently recruiting the DNA repair machinery (55, 56). Regarding the activated DNA repair pathways, NHEJ, MMR, and BER showed strong enrichment in the pathway analysis in AzadC-treated wt cells compared with untreated cells under Lif conditions (Fig 3I and Table 2). This is in line with the up-regulated chromatin recruitment of key components of those pathways (Fig 3H), indicating that those repair pathways were activated.

To gain better insights into the activation of the DDR in wt cells under Lif conditions, we then analysed the global protein expression-level changes after AzadC treatment (Supplemental Data 3, Table S3) and subsequently compared the chromatin enrichment of the DNA repair–associated proteins with their global expression-level changes (Supplemental Data 3, Table S4). Pathway enrichment analysis of the whole proteome data revealed that BER also showed significant enrichment (Fig 3J) (Supplemental Data 3, Table S5). Among the BER-associated proteins, Smug1 and Fen1 showed both significantly increased expression levels and significantly changed chromatin enrichment (Fig 3K). These results suggested that BER is activated by increased protein expression levels and by active recruitment of the involved DNA repair proteins to the site of action. Strikingly, all other DNA repair–associated proteins that were found to be enriched in the chromatin fraction after AzadC treatment did not show any significant expression-level changes (Fig 3K), revealing that DDR activation in response to AzadC-induced DNA lesions is substantially based on active chromatin recruitment. On the contrary, there were several DNA repair–associated proteins that showed exclusively higher expression levels after AzadC treatment but were not significantly enriched in the chromatin fraction. The most prominent one, with a 16-fold higher protein level after AzadC treatment (−log[P-value] = 1.4, log₂FC = 4), was the key tumour suppressor p53. P53 is a nuclear transcription factor that plays, among many other functions, a pivotal role in DDR by inducing the expression of many DNA repair–associated genes and genes that trigger cell-cycle arrest and apoptosis in the presence of DNA damage (57). Under normal conditions, p53 has a very short half-life time and is in addition mostly present in an inactivated state. Our whole proteome data are in accordance with literature that upon DNA damage sensing, the p53 protein levels are substantially increased as the Tp53 mRNA is more efficiently translated and the p53 protein is stabilized, leading to its accumulation (58). The activation of p53 and the subsequent effects were also observed in the pathway enrichment analysis that showed p53 signalling among the top enriched pathways (Fig 3J, Supplemental Data 3, Table S5). The reason why we did not observe significant enrichment of p53 in the chromatin fraction of AzadC-treated cells might be that transcription factors usually bind very transiently to the DNA and can therefore often not be captured without crosslinking. One of the target genes of p53 is the DNA damage–binding protein 2 (Ddb2), a key component of NER that subsequently forms a complex at the lesion site with the ubiquitin ligase Cullin-4a (Cul4a), which then initiates the proteolytic digest of Ddb2. This targeted degradation at the lesion site in turn recruits the DNA repair protein Xpc to initiate global genome NER (GG-NER) (59, 60). Interestingly, Ddb2 showed a significant increase in global protein levels, but we failed to detect it in the chromatin fraction, and Cul4a and Xpc were even depleted from the chromatin fraction after AzadC treatment (Fig 3H, Supplemental Data 3, Table S1). Having a closer look at other components of the GG-NER, we observed that most of the depleted proteins in the chromatin fraction after AzadC treatment belong to or are at least associated with this repair pathway, including TIMELESS-interacting proteins Tipin and Claspin (Clspn), which are part of the intra-S checkpoint that is activated upon DNA replication stress, the NAD⁺-dependent protein deacetylase sirtuin-6 (Sirt6), and the DNA polymerase e (Pole) (61, 62, 63). Moreover, there were several NER components, including replication factor 1 (Rfc1), general transcription factor IIH subunits Ercc2 and Ercc4 (Xpd and Xpf), and Gtf2H4 and Cyclin-H (Ccnh) (39, 64, 65, 66) that were not enriched in the chromatin fraction after AzadC treatment (Fig 3K), but significantly up-regulated on the global protein expression level. Consequently, pathway enrichment analysis also revealed an up-regulation of NER based on the whole protein expression data (Fig 3J). These combined results suggest that when replication stress is very high after AzadC treatment, NER is initially activated on the protein expression level but then either not used or even actively excluded from the AzadC-induced DNA lesion repair as it progresses.

As DNMT activity and therefore the induced replication stress upon AzadC treatment are very different in the wt under Lif and 2iL conditions, we replicated the chromatin enrichment workflow in wt cells that were treated with 2.5 μM of AzadC under 2iL conditions, with the respective untreated 2iL wt cells as a control. Under 2iL conditions, we observed 33 DNA repair–associated proteins being enriched in the chromatin fraction after AzadC treatment and only two repair-associated proteins being depleted (Fig S3D, Supplemental Data 3, Table S6). 14 DNA repair–associated proteins were enriched in the chromatin fraction under 2iL and Lif conditions (Fig 3L). Among them were several components of BER and MMR, which were also significantly enriched as pathways under 2iL conditions (Fig 3M, Supplemental Data 3, Table S7). Moreover, Rad50 and Chek1 were highly enriched after AzadC treatment under both conditions, indicating that they play a central role in dealing with AzadC-induced DNA lesions. In addition, we observed after AzadC treatment under 2iL conditions, chromatin enrichment of the Ataxia-telangiectasia mutated (Atm) and Rad3-related (Atr) protein kinases, including their downstream targets Chek1 and Chek2 (Figs 3L and S3C). Atm and Atr are essential for genome integrity, the DDR and checkpoint signalling, and p53 activation (67, 68). p53 signalling was also significantly enriched in the pathway enrichment analysis of the chromatin fraction of AzadC-treated wt cells under 2iL conditions (Fig 3M). Interestingly, the FA repair pathway, including HR, that was previously reported to play an important role in repair of AzadC-induced DNA lesions (26), was under Lif conditions neither enriched as a pathway nor were the DNA repair protein Rad51 homolog 1 (Rad51, also known as Fancr) or other central components of the FA repair pathway enriched in the chromatin fraction of AzadC-treated cells. In contrast, under 2iL conditions, several important FA proteins, including Rad51 and Fanci, were significantly enriched (Figs 3L and S3D), and HR and FA, but not NHEJ, were overrepresented pathways (Fig 3M). Overall, these results indicated that BER and MMR are commonly activated DDRs to deal with AzadC-induced DNA lesions. In contrast, FA and HR are only activated when DNMT activity and therefore the resulting replication stress after DNMT crosslinking are moderate, whereas NHEJ and a-EJ, which are more error-prone but very efficient, are the repair pathways of choice when replication stress is immense after AzadC treatment because of high DNMT activity. To validate the applicability of our chromatin–proteome-centred approach for comprehensive investigation of DDR after treatment with DNA-damaging reagents in a human AML cell line, we compared the chromatin-enriched DNA repair–associated proteins of AzadC-treated mESCs under 2iL conditions with those of AzadC-treated KG-1 cells. In accordance with existing literature on comparable cancer cell lines (26), proteins of the FA pathway were highly enriched in KG-1 cells, including RAD51 (Fig S3E, Supplemental Data 3, Table S8). Of note, although proteins for HR were enriched in KG-1, several proteins involved in NHEJ were depleted from the chromatin fraction, suggesting that also cancer cells promote, if possible, FA-HR over NHEJ to repair AzadC-induced DNA double-strand breaks.

In the last step, we investigated which repair pathways are involved to deal with genomically incorporated AzadC when DNMT-DNA crosslinking cannot take place. To this end, we treated DNMT-TKO cells under Lif conditions with an equal dose of AzadC compared with wt cells (200 nM over 48 h) and enriched the chromatin protein fraction. The total amount of DNA repair–associated proteins that we could detect in the chromatin fraction of DNMT-TKO cells did not differ from the wt (Supplemental Data 3, Table S9). However, in contrast to wt cells that showed a massive change of the chromatin-bound proteins, hardly any changes of chromatin-associated proteins in general could be observed in the DNMT-TKO cells compared with the untreated control (Fig 3N). Among the few significantly enriched proteins were only three repair-associated proteins: the endonuclease 8-like 3 DNA glycosylase Neil3, the crossover junction endonuclease MUS81, and the WD repeat–containing protein 48 (Wdr48), which is part of the FA repair pathway. To rule out that only the low concentration of AzadC failed to activate DDR in DNMT-TKO cells, we treated the DNMT-TKO cells under 2iL conditions with 2.5 μM of AzadC over 48 h and checked chromatin enrichment. However, although we observed a loss of cellular fitness of DNMT-TKO cells (Fig 2B) when exposed to this concentration, the chromatin-bound protein fraction was hardly affected and no concerted DDR was observed either (Fig S3F, Supplemental Data 4, Table S10). These results indicated that without DNMT crosslinking, AzadC does not invoke a strong and concerted DDR that can be monitored on the bulk chromatin level.

The cytotoxic effect of AzaC depends on its DNA incorporation and DNMT activity but not on crosslinking of the RNA methyltransferase Trdmt1

To elucidate the RNA-dependent effect of AzaC in the absence of DNMTs, we compared treatment of wt and DNMT-TKO cells under 2iL and Lif conditions with increasing concentrations of AzaC, because an RNA effect should manifest in both genotypes and would be potentially dominant in the AzaC-treated DNMT-TKO mESCs (Fig 1G, Table 1). First, we checked the incorporation levels of AzaC into DNA as AzadC (Fig 4A) and into RNA as AzaC (Fig 4B) after 2.5 μM AzaC feeding for 48 h under 2iL conditions. For the wt, we observed the expected lower incorporation rate into DNA with ∼12.5% AzadC/G intensity compared to feeding with AzadC after 24 h (Fig 4A, wt treatment a). This result is in perfect accordance with literature where it was reported that 10–20% of AzaC are incorporated into DNA (12). The pattern that the DNA-incorporated levels dropped by half 48 h after treatment when no additional compound was added, but could be increased by additional 80% when a new compound was added after 24 h, was also observed for AzaC treatment in the wt. However, although we did not detect any difference regarding the uptake and metabolization of AzadC between wt and DNMT-TKO cells (Fig 2F), AzaC incorporation into DNA was significantly lower in the DNMT-TKO compared with the wt (Fig 4A). On the contrary, the incorporation rates of AzaC into RNA were comparable between the wt and the DNMT-TKO cells (Fig 4B). Therefore, different intercellular metabolization kinetics and not a difference in the uptake of AzaC were probably the reason for the lower DNA incorporation in the DNMT-TKO. Moreover, the level of AzaC in RNA in wt and DNMT-TKO mESCs could only be maintained but not further increased when an additional compound was added after 24 h (Fig 4B). This observation indicated that RNA compared with DNA has a higher turnover and AzaC could therefore not accumulate. Next, we investigated the phenotypic changes after AzaC treatment (2.5 μM for 48 h, treatment b) in wt and DNMT-TKO cells under 2iL conditions. Whereas the proliferation rate initially slowed down in the first 24 h after AzaC treatment in the DNMT-TKO cells, proliferation quickly resumed in those cells and the cell number after 48 h was the same as for the untreated control (Fig 4C). In contrast, the wt showed a persistent slower proliferation rate after AzaC treatment with doubling times comparable to AzadC treatment. Intriguingly, the slower proliferation rate in the wt was not reflected in higher cytotoxicity as indicated by brightfield microscopy (Fig 4D) and the flow cytometry–based apoptosis assay (Fig 4E). In the DNMT-TKO, AzaC did not have any cytotoxic activity in the applied concentration either (Fig 4D and E). When we had a closer look at the DNA hypomethylating effect of AzaC compared with AzadC in the wt cells under 2iL conditions, we observed a substantial decrease in mdC for both compounds with no difference between AzadC and AzaC at 1.25 and 2.5 μM (Fig 4F). In contrast, γH2AX levels in the wt were much higher after AzadC compared with AzaC treatment as indicated by the intensity of the γH2AX signal in fluorescence microscopy (Fig 4G). In leukaemic cells, it was reported that AzaC incorporation into RNA destabilizes the mRNA of RRM2, which results in lower RRM2 protein expression levels. As the ribonucleotide reductase is essential to produce DNA nucleotides, depletion of one of its core subunits ultimately leads to a proliferation block (13). To check whether this mechanism also contributes to the inhibition of proliferation in AzaC-treated mESCs, we checked our 2iL whole proteome data of AzaC-treated wt and DNMT-TKO cells for proteins involved in nucleotide metabolism according to Reactome (R-HAS-15869.7) but found only the glutathione reductase significantly down-regulated in the wt (Supplemental Data 3, Table S11), and thymidine kinase (Tk) as well as inosine-5′-monophosphate dehydrogenase 1 (Impdh) and adenine phosphoribosyltransferase (Aprt) down-regulated in the DNMT-TKO (Supplemental Data 4, Table S12). When we checked the chromatin-enriched proteome of wt cells after AzaC treatment (Supplemental Data 4, Table S13), we observed a very high correlation compared with the AzadC-treated cells (Pearson’s r = 0.7778; Supplemental Data 4, Table S14), but only few repair-associated proteins reached the significance threshold (Fig 4H), which is in line with the observation that AzaC introduces less DNA damage than AzadC. In summary, the obtained results from the DNMT-TKO suggest that under steady conditions, the RNA incorporation of AzaC leads to an initial proliferation delay that can be, however, quickly overcome by the cells because of high RNA turnover. As a consequence, no persistent RNA-dependent effect can be achieved without additional AzaC supply. As DNA damage seemed to be marginal under 2iL conditions after AzaC treatment in the wt, which is in line with the absent toxicity, the inhibition of DNMTs and the resulting significant decrease in mdC appear to be responsible for the significantly decreased proliferation rate in the wt after AzaC treatment. Although AzaC incorporation into gDNA reaches only 12.5% compared with AzadC treatment in the wt, the changes of the chromatin-associated proteins are highly similar for AzaC and AzadC treatment but less pronounced after AzaC treatment.

Next, we investigated the effect of AzaC when the transcriptome and proteome and therefore cellular identity of both genotypes drastically change by switching the mESC culture conditions from naïve to primed. In stark contrast to 2iL conditions, AzaC treatment of wt cells under Lif conditions had a very strong effect on the cellular viability as indicated by brightfield microscopy (Fig 4I) and the flow cytometry–based apoptosis assay (Fig 4J). Similar to AzadC treatment, cellular proliferation of the wt completely stopped using an AzaC concentration as low as 0.2 μM (Fig S4A). The effect on the viability of DNMT-TKO cells, however, was minimal at this concentration (Fig 4I and J), and the proliferation rate was not affected either (Fig S4A). To further investigate the RNA-dependent features of AzaC, we quantified cell death in wt, in DNMT-TKO, and in addition in Trdmt1-deficient (Dnmt2-deficient) mESCs (Trdmt1-KO) after exposing the cells to increasing AzaC concentrations for 48 h under Lif conditions. In the Trdmt1-KO, no Trdmt1 crosslinking to RNA by AzaC can take place. For the wt, we observed a steady increase in toxicity with increasing concentrations of AzaC, whereas for the DNMT-TKO, AzaC was not toxic at the applied concentrations (Fig 4K). Remarkably, the Trdmt1-KO was significantly the most sensitive genotype towards AzaC treatment. Based on these data, we could not explain why the Trdmt1-KO reacted so sensitively towards AzaC treatment, but this observation contradicts the idea that Trdmt1-RNA crosslinking as a result of RNA incorporation of AzaC has a negative impact on cellular well-being, at least in the chosen mESC model system.

Last, we wanted to check whether DNA damage or decreased mdC levels were responsible for the substantially increased cell death events in the wt under Lif conditions. First, we quantified mdC levels in the untreated control, AzadC-, and AzaC-treated cells and observed that although mdC levels were significantly decreased after AzaC treatment compared with the untreated control, the levels remained at significantly higher levels compared with AzadC-treated cells with no difference between the two concentrations tested (Fig 4L). Under 2iL conditions, mdC levels were equally reduced after AzaC and AzadC treatment (Fig 4F); cytotoxicity of AzaC, however, remained low. In contrast, under Lif conditions, AzaC was significantly less efficient compared with AzadC to reduce mdC (Fig 4L), but cytotoxicity was high. Therefore, we concluded that the induction of substantial DNA damage after AzaC treatment because of high DNMT activity was the underlying reason for the dramatically increased sensitivity of the primed wt mESCs towards AzaC compared with the naïve wt mESCs. To substantiate this hypothesis, we in addition tested the non-nucleoside DNMT inhibitor RG-108 (32) on wt cells under Lif conditions, and although we reached a reduction in mdC comparable to the reduction after AzaC treatment (Fig S4B), we failed to detect increased cell death (Fig S4C), showing that the reduced mdC levels did not impair cellular viability under these conditions.

In summary, the effects of AzaC appear to primarily rely on its incorporation into DNA, and although DNMT inhibition and thereby reduced mdC levels significantly slow down the proliferation rate, the induction of cell death depends on the severity of the induced DNA damage. At moderate DNMT activity, the level of severe DNA lesions and therefore cell death rates remain low after AzaC treatment, whereas under high DNMT activity, AzaC has substantial cytotoxicity comparable to AzadC.

Reagents and cell lines

Biological resources

wt J1 mESCs were described in Li et al (1992) (71) and originally provided by the Jaenisch laboratory (Whitehead Institute, USA). DNMT-TKO J1 mESCs were described in Tsumara et al (2006) (29) and originally provided by the Okano laboratory (RIKEN). Trdmt1-KO J1 mESCs were described in Okano et al (1998) (72) and originally provided by the Jaenisch lab.

Statistical analysis

For the proteomics experiments, details about the statistical analysis, including sample size, data exclusion, and significance thresholds, are given in the Proteomics—Materials and Methods section, including the respective analysis software. For all other experiments, GraphPad Prism (v 9.4.0) was used for statistical analysis and details are given in Supplemental Data 1. No statistical methods were used to pre-determine the sample size. Sample sizes were chosen based on cost, experience, and commonly used sample sizes for in vitro experiments (n ≥ 3), which provided in this study low intersample variety between samples of the same group. The number of samples for each experiment is either described in the Materials and Methods section or directly evident from the respective figure and figure legend.

Data availability

Original (unprocessed) and metadata are deposited as described in the Data Availability section.

Database references

We used the Reactome knowledge database (https://reactome.org, accessed March 2023–April 2024 (38)) to assign DNA repair–relevant proteins in our proteomics data sets. To receive optimal coverage, we mapped the mouse proteins to the respective human proteins via the gene name and did the subsequent analysis on the human pathways.

The UniProt database (https://uniprot.org) was used to download the FASTA file of Mus musculus.

GOrilla (https://cbl-gorilla.cs.technion.ac.il/; accessed February 2023 (73)) was used for Gene Ontology (GO) term analysis.

mESC handling

Microscopy

Triple-quadrupole mass spectrometry (QQQ-MS) for nucleoside quantification

For QQQ-MS experiments, mESCs were seeded in a p60 and treated with the indicated concentrations. For the AzadC and AzaC incorporation experiments, cells were treated for either 24 h (treatment a), 48 h without medium change (treatment b), 48 h with medium change after 24 h without second compound addition (treatment c) or 48 h with medium change after 24 h and second compound addition at the same concentration (treatment d). For all other QQQ-MS experiments, mESCs were treated for 48 h without medium change (treatment b). After treatment was finished, medium was aspirated, and mESCs were washed with PBS and harvested. After harvest, mESCs were washed once with PBS and afterwards directly lysed in 800 μl of GTC buffer (3.5 M guanidine thiocyanate, 25 mM trisodium dihydrate, 14.3 mM β-mercaptoethanol, pH 6.9), and either processed directly or shock-frozen in liquid N₂ and stored at −80°C. For thawing, lysed samples were quickly warmed to RT and gDNA isolation was performed.

Flow cytometry–based apoptosis assay

For flow cytometry, mESCs were seeded in a 12 well and treated as indicated. Before harvest, the medium, which includes dead floating cells, was not aspirated but collected as well and the remaining attached cells were harvested and combined with the cells from the medium. Afterwards, mESCs were washed twice with PBS and subsequently counted. 1.5 × 10⁵ cells per sample were transferred into a new tube. Apoptosis and necrosis were determined using FITC–Annexin V Apoptosis Detection Kit and SYTOX Red Dead Cell Stain. To this end, cells were resuspended in 150 μl of Annexin V binding buffer supplemented with 0.75 μl of FITC-conjugated Annexin V and 0.15 μl of SYTOX Red Dead Cell Stain, gently vortexed, and incubated at RT for 15 min in the dark. Afterwards, samples were put on ice and the cell suspension was filtered through a 35-μm strainer before measurement. For the analysis, BD FACSCanto (recording of 10,000 events per sample) and FlowJo Single Cell Analysis Software (v10.8.0) were used. Gates (FSC [A] – SSC [A] to remove cell debris, FSC [A] – FSC [H] to gate for single cells, and last FITC/APC to distinguish between live, dead, and early apoptotic cells) were set once for the control sample and then applied to all other samples.

Immunoblot analysis

Liquid chromatography–tandem mass spectrometry (LC-MS/MS)—proteomics

For proteomics experiments, mESCs were seeded in a p150 and treated for 48 h. For each sample type (specific combination of genotype, culturing conditions, treatment), four biologically independent replicates were initially generated. After treatment, mESCs were harvested and washed with PBS. 10% of the cells were used for whole proteome isolation and the rest for chromatin enrichment to generate whole proteome and chromatin-enriched proteome samples from the same batch of cells.

KG-1 handling

KG-1 cells were cultivated at 37°C in water-saturated, CO₂-enriched (5%) atmosphere in RPMI-1640 medium, supplemented with 10% FBS and 2 mM L-alanyl-L-glutamine, and maintained at a concentration of 0.5 × 10⁶ cells/ml. For AzadC treatment, 1.0 × 10⁶ cells/ml were plated in 5 ml and treated for 72 h with 1.25 μM AzadC. Untreated cells served as a control. Four biologically independent replicates per condition were used for subsequent nuclear enrichment and MS analysis.