Active DNA demethylation promotes cell fate specification and the DNA damage response

Abstract

Neurons harbor high levels of single-strand DNA breaks (SSBs) that are targeted to neuronal enhancers, but the source of this endogenous damage remains unclear. Using two systems of postmitotic lineage specification—induced pluripotent stem cell–derived neurons and transdifferentiated macrophages—we show that thymidine DNA glycosylase (TDG)–driven excision of methylcytosines oxidized with ten-eleven translocation enzymes (TET) is a source of SSBs. Although macrophage differentiation favors short-patch base excision repair to fill in single-nucleotide gaps, neurons also frequently use the long-patch subpathway. Disrupting this gap-filling process using anti-neoplastic cytosine analogs triggers a DNA damage response and neuronal cell death, which is dependent on TDG. Thus, TET-mediated active DNA demethylation promotes endogenous DNA damage, a process that normally safeguards cell identity but can also provoke neurotoxicity after anticancer treatments.

Active DNA demethylation in neurons

The DNA of neurons is continually damaged due to lifelong, high-level metabolic and transcriptional activity. Recent studies have also demonstrated extensive “programmed” DNA damage in differentiating postmitotic neurons. Wang et al. identified endogenous lesions as single-strand-break intermediates of thymine DNA glycosylase (TDG)–mediated removal of oxidized methylcytosines during active DNA demethylation (see the Perspective by López-Moyado and Rao). Interrupting active DNA demethylation using antineoplastic cytosine analogs triggered TDG-dependent neuronal cell death. This work suggests that the well-known neurotoxic side effects of certain chemotherapies, also called “chemobrain,” could be linked to DNA repair processes intrinsic to normal neuronal differentiation. —DJ

Defects in DNA repair result in cancer predisposition as well as neurological diseases. Although all cell types incur DNA damage and mutations, neurons are exceptionally vulnerable to defects in single-strand break (SSB) repair (1). SSBs are detected by XRCC1 and PARP (poly-ADP ribose polymerase), which recruit factors involved in DNA end modification (PNKP, polynucleotide kinase 3′-phosphatase; APTX, aprataxin), DNA gap filling (POLβ, DNA polymerase β family), and ligation (DNA ligase 1 or 3) (2). These proteins are essential for most SSB repair events, usually comprising “short-patch” reactions in which only a single missing nucleotide is replaced. More rarely, a long-patch subpathway is employed, which uses extended DNA synthesis before ligation. However, long-patch SSB repair cannot completely compensate for loss of short-patch SSB repair, as evidenced by the association of neurological diseases with mutations in genes required for short-patch repair, like XRCC1, APTX and PNKP (2).

Recently, we and others demonstrated that highly active long-patch, synthesis-associated repair (SAR) of SSBs in neurons can be detected by incorporation of 5-ethynyl-2′-deoxyuridine (EdU) (SAR-seq) (3, 4). DNA repair synthesis and SSBs were localized to neuronal enhancers, corresponding to 2% of the genome (3, 4). Why neurons concentrate the repair machinery to these hotspots is an unsolved question; moreover, the source of endogenous SSBs and their physiological relevance remain unclear.

Active DNA demethylation generates SSBs and ADP-ribosylation in neurons

SAR-seq peaks in neurons correlate with oxidized forms of 5-methylcytosine, suggesting the potential involvement of active DNA demethylation via ten-eleven translocation (TET) enzymes (4) (fig. S1A). In induced pluripotent stem cell (iPSC)–derived neurons (iNs), all TET family enzymes (TET1, TET2, and TET3) are expressed (fig. S1B). During passive demethylation, TET-mediated oxidized methylcytidines are lost during successive rounds of replication. During active DNA demethylation, the thymidine DNA glycosylase (TDG) excises TET-mediated oxidized methylcytidines 5fC and 5caC to produce SSBs (5, 6). SSBs can be detected by S1-END-seq using chain-terminating dideoxynucleosides (ddN) (Fig. 1A) (4). Given the observed accumulation of SSBs near oxidized methylcytosines (4), we set out to test whether active DNA demethylation is a source of endogenous DNA damage in postmitotic neurons.

Fig. 1. Single-strand breaks at neuronal enhancers are TDG dependent.

(A) Schematic overview of SSB generation during SP- and LP-BER and SSB detection by ddC S1-END-seq. (B) Heatmaps of S1-END-seq peaks in iN upon overnight treatment with different dideoxynucleoside (ddA, ddC, ddT, or ddG), plotted 1kb on either side of SAR-seq peak summits, ordered by SAR-seq intensity. (C) Genome browser screenshot showing single-strand breaks (ddC S1-END-seq) detected in TDG^degron iN treated with or without dTAG. (D) Heatmaps of ddC S1-END-seq in TDG^degron iN treated with or without dTAG. ddC S1-END-seq is ordered by SAR-seq intensity (4) and plotted 1 kb on either side of SAR-seq peak summits in iN. (E) Representative images (left) and quantification (right) of iN with immunofluorescence staining for DAPI (blue) and PARylation (green) in TDG^degron iN treated with or without dTAG. Scale bar, 10 μm. Statistical significance was determined using Mann-Whitney test (****p < 0.0001).

Initially, we used CRISPR interference (CRISPRi) to deplete TDG in human iPSCs (fig. S1C), after which they were differentiated into neurons (iNs) (7, 8). TDG-deficient, but not wild-type (WT) iNs, accumulated high levels of 5fC/caC (fig. S1D). We previously found that the mixture of ddNs led to robust accumulation of SSBs detected by S1-END-seq in iNs(4). By treating cells with each ddN separately, we found that only ddC, but not ddA, ddT, or ddG, produced single-strand DNA gaps (Fig. 1B). We therefore tested whether the formation of SSBs was dependent on TDG and found that, although SSBs accumulated at enhancers in WT iNs, TDG-depleted iNs harbored far fewer lesions (fig. S1, E and F).

Because TET-mediated active DNA demethylation may be important for transcriptional changes during neuronal differentiation, we inactivated TDG after cells had already differentiated into iNs. To do this, we used the degradation tag (dTAG) system, wherein a FKBP12 tag was knocked into the N terminus of endogenous TDG in iPSCs (fig. S1G). After 6 days of differentiation, TDG was acutely degraded in neurons upon addition of dTAG (fig. S1, H and I). On day 7, we performed ddC S1-END-seq. Consistent with the CRISPRi results, we found that TDG was required for SSB formation (Fig. 1, C and D).

A high level of PARP activity was found in iNs (Fig. 1E) (4). To test whether this is related to active DNA demethylation, we measured PARP activity when TDG was acutely degraded with dTAG. Indeed, ADP-ribosylation decreased on average by twofold in neurons lacking TDG (Fig. 1E). Thus, TET-mediated active DNA demethylation is the source of SSBs at neuronal enhancers, although other lesions could also contribute to ADP-ribosylation.

TET2 is required for pre-B to macrophage differentiation

In contrast to neurons, recurrent DNA repair peaks were not observed in other postmitotic cells, including G₀-arrested pre-B cells and iPSC-derived skeletal muscle cells (4). To understand whether overall DNA damage and repair is exclusive to neurons, we examined postmitotic macrophages derived by transcription factor–induced transdifferentiation from pre-B cells (Fig. 2A) (9). C/EBPα turns on the myeloid program in pre-B cells efficiently (>95%) and rapidly (48 hours) (Fig. 2B) (9, 10). Cell conversion requires up-regulation of the C/EBPα target PU.1, which drives macrophage-specific gene expression and down-regulation of the B cell-specific transcription factor PAX5 to shut down the B cell transcriptional program (11). Previous studies demonstrated that TET2 knockdown partially delayed C/EBPα-induced macrophage (iM) differentiation because of impaired up-regulation of myeloid genes (12). To confirm the involvement of TET2 in pre-B to macrophage cell differentiation, we knocked out TET2 in pre-B cells by CRISPR-Cas9 targeting (fig. S2A). Consistent with previous findings, the macrophage cell surface marker MAC-1 was induced in more than 95% of control cells 2 days after C/EBPα stimulation, whereas only 4% of TET2-deficient cells expressed MAC-1 (fig. S2B). Thus, TET2 is required for pre-B to iM conversion.

Fig. 2. TDG-dependent SSBs at macrophage enhancers.

(A) Schematic overview of pre-B to iM transdifferentiation. (B) FACS plot (left) and quantification (right) of MAC-1 and CD19 expression in C/EBPα-ER–infected wild type and TDG^−/− pre–B cells before induction and 2 days after induction with β-estradiol. Statistical significance was determined using unpaired t test (mean ± SD). (C) Genome browser screenshots illustrating SAR-seq and ddC S1-END-seq in iM. For SAR-seq, cells were either not treated (NT) or treated with PARPi on day 1 and harvested on day 2. (D) Bar graph comparing fold increase of SAR-seq peak numbers upon PARPi versus nontreated iM or nontreated iN. Also plotted is the fold-increase in iN expressing POLB-targeted (sgPOLB) or XRCC1-targeted (sgXRCC1) CRISPRi plasmids versus nontargeted control. (E) Heatmaps illustrating the correlation in iM between SAR-seq, ddC S1-END-seq, enhancer markers (ChIP-seq for H3K4me1, H3K27ac, and PU.1), and 5hmC (5hmC Seal). (F) Genome browser screenshots showing SAR-seq (with PARPi), ddC S1-ENDseq, transcription factor binding (C/EBPα and PU.1), and enhancer markers (ChIP-seq for H3K4me1, H3K27ac) in iM at the Klf4 de novo enhancer. “B” represents pre-B cells and “iMΨ” represents induced iM. (G) Genome browser screenshot showing SAR-seq (with PARPi) and ddC S1-END-seq in WT and TDG knockout iM. (H) Heatmaps showing SAR-seq (with PARPi) and ddC S1-END-seq in WT and TDG-deficient iM.

Active DNA demethylation triggers SP-BER during macrophage trans-differentiation

To test whether active DNA demethylation operates during macrophage cell conversion, we knocked out TDG in pre-B cells (fig. S2C). In contrast to the severe differentiation block in TET2^−/− cells (fig. S2B), most TDG^−/− cells down-regulated B-cell surface marker CD19 and expressed MAC-1 by 2 days post-C/EBPα induction (Fig. 2B). Thus, distinct from TET2, TDG is not required for the generation of macrophage-like cells. Nevertheless, we detected an accumulation of 5fC/caC in TDG^−/− but not WT iM (fig. S2D). Thus, TDG excises oxidized methylcytosines during the transdifferentiation of pre-B to macrophage cells.

To localize active DNA demethylation and associated repair sites, cells were treated with aphidicolin (APH) for 4 hours after 1 day of C/EBPα induction to inhibit any residual cell division without affecting differentiation (fig. S2E). Cells were then labeled with EdU for 20 hours and processed for SAR-seq (Fig. 2C) (4). Distinct from iNs, which exhibited ~55,000 hotspots of DNA repair (4), almost no peaks were detectable in iMs (Fig. 2C), reminiscent of our findings in other postmitotic cells (4). One potential explanation is that replication-independent active DNA demethylation generates only few SSBs in iMs, making a minor contribution to differentiation, as has been shown in other cell types (13). Alternatively, TDG-mediated excision of 5fC/caC might occur at a high frequency, but these bases would be replaced by unmodified cytosine via short patch (SP) base excision repair (BER). SP-BER events triggered by TDG excision of oxidized methylcytosines are undetectable by SAR-seq because this assay measures incorporated EdU, a thymidine analog.

Depletion or inhibition of factors involved in SP-BER, including XRCC1, PARP1, and POLβ, leads to an approximately twofold increase in the number of SAR-seq peaks in iNs (Fig. 2D), suggesting that long patch (LP)–BER is the primary source of the SAR-seq signal in these cells (4). Because iMs lack robust SAR-seq signals, we asked whether SP-BER participates in active cytosine demethylation. To investigate this question, we first inactivated SP-BER via PARP inhibition (PARPi) and found a 60-fold increase in the number of SAR-seq peaks in iMs from 1951 to 119,397 (Fig. 2, C and D). To further evaluate the function of SP-BER in iMs, we treated cells with chain-terminating ddN followed by S1-END-seq. As in iNs, SSBs were detectable only when iMs were treated with ddC alone, but not with ddA, ddT, or ddG (Fig. 2C and fig. S2F). Overall, 24,000 SSB peaks were detectable in iM, which is similar to the number of S1-END-seq peaks found in iN (28,000). Thus, SSBs accumulate in both cell types, but whereas LP-BER contributes to about half of SSB repair in iNs, SP-BER is overwhelmingly dominant in iMs.

SSBs form predominantly at de novo enhancers and are TDG dependent

Like iNs (3, 4), LP-BER sites in iMs, measured after PARPi treatment, colocalized and correlated with enhancers marked with H3K4me1 and H3K27ac (Fig. 2E). Analysis of transcription factor binding motifs revealed an overrepresentation of PU.1 binding sites centered at SAR-seq summits (fig. S2G). ChIP-seq analysis of PU.1 binding confirmed colocalization with sites of SSBs, DNA repair synthesis, and 5hmC (Fig. 2E). Notably, PU.1 and C/EBPα interact with TET2, PU.1- and C/EBPα-targeted enhancers undergo TET2-mediated demethylation, and PU.1 is required for myeloid differentiation (14–16).

Myeloid enhancers have been described as “preexisting” or “de novo” depending on the order of recruitment of PU.1 and C/EBPα (16). Preexisting enhancers are prebound by PU.1 or C/EBPα and already active in pre-B cells, whereas de novo enhancers are first bound by C/EBPα before PU.1 and gradually become active during macrophage transdifferentiation (16). Although SAR-seq and S1-END-seq peaks were found at both types of enhancers, sites of DNA repair were more highly enriched for de novo macrophage enhancers (figs. S2, H to J). Examples of de novo enhancers that become demethylated and activated upon C/EBPα binding include the enhancers of Klf4 and Lefty2 (Fig. 2F and fig. S2K) (15). These de novo enhancers harbored peaks of SSBs and DNA repair synthesis that colocalized with C/EBPα and PU.1 (Fig. 2F and fig. S2K). We conclude that PU.1 guides TET2 to de novo macrophage enhancers, which promotes active DNA demethylation, SSB formation, and SP-BER.

To determine whether SSB formation and repair at macrophage enhancers is TDG dependent, we performed ddC S1-END-seq and SAR-seq in WT versus TDG^−/− cells in the presence of PARPi. These assays revealed that TDG was essential for both SSB formation and DNA synthesis–associated repair in iM (Fig. 2, G and H). Thus, TET-mediated active DNA demethylation is the source of SSBs at both neuronal and macrophage enhancers.

Active DNA demethylation contributes to cell identity

Absence of TDG-mediated excision of 5fC/caC did not prevent the generation of macrophage-like cells (Fig. 2B). Yet, thousands of SSBs were generated (Fig. 2, C and E), and thousands of genes are up- and down-regulated during normal differentiation (12). This raises the question of the physiological role of active DNA demethylation. To test the functionality of TDG^−/− macrophages, we measured the phagocytic activity of C/EBPα-induced cells by incubating cells with prelabeled dsRed fluorescent protein–expressing Escherichia coli prior to analysis by image cytometry (Fig. 3A). Whereas 90% of control MAC-1⁺ cells ingested E. coli after 48 hours following incubation, the phagocytic capacity of MAC-1⁺TDG^−/− cells was reduced to 36% (Fig. 3B). Moreover, there was a 23-fold reduction in the amount of total E. coli ingested per mutant cell (Fig. 3C). Thus, TDG loss compromises the ability to phagocytose bacteria in transdifferentiated cells.

Fig. 3. Active DNA demethylation contributes to iM cell identity.

(A) Imaging flow cytometric analysis of phagocytosis in iM. Scale bar, 7 μm. BF (bright field), green fluorescent protein (GFP) (C/EBPα expression), MAC-1 (MAC-1 expression, red), E. coli (dsRed-E. coli, yellow). (B) Bar graph of percentages of MAC-1-positive WT and TDG^−/− iMs that internalized dsRed-E. coli. Statistical significance was determined using unpaired t test (****p < 0.0001). (C) Bar graph and median intensity of internalized dsRed-E. coli in MAC-1-positive cells. Statistical significance was determined using unpaired t test (****p < 0.0001). (D) Linear plots showing expression changes of up- or down-regulated genes on average in WT and TDG^−/− cells. (E) Left, genome browser screenshot showing SAR-seq, ddC S1-END-seq, and enhancer activity (H3K27ac ChIP-seq) in WT and TDG^−/− iM at the Csf1r enhancer. Right, expression of Csf1r in WT and TDG^−/− iM measured by RNA-seq. (F) Heatmap of gene expression associated with macrophage activation in WT and TDG^−/− cells before and during transdifferentiation. Tlr (toll-like receptor) genes are indicated.

To determine the role of active cytosine demethylation on enhancer activity and gene expression, we performed RNA sequencing (RNA-seq) on WT versus TDG^−/− cells before (0 hours) and during (24, 48, and 72 hours) macrophage differentiation (Fig. 3D). In WT cells, most down-regulated genes were associated with biological processes involving cell cycle and DNA repair, likely caused by the rapid cell cycle arrest induced by C/EBPα (fig. S3A, left panel) (10). Gene Ontology analysis revealed up-regulation of genes relating to macrophage function (fig. S3A, right panel). Although similar gene sets were down-regulated in WT and TDG^−/− cells, TDG^−/− cells failed to effectively up-regulate many genes related to macrophage differentiation (Fig. 3D).

We then sorted up-regulated genes into those that were either close to or far from DNA repair (SAR) loci. Because the majority (70%) of up-regulated genes are localized within 100 kb of stable C/EBPα-binding sites (15, 16), we used 100 kb as the cut-off within which genes are considered neighboring enhancers. We observed that TDG^−/− cells exhibited a defect in the up-regulation of genes that were located close to enhancers active for DNA repair (fig. S3B). Genes that were up-regulated in a TDG-dependent manner included those that are critical for macrophage identity and function (fig. S3C) such as Csf1r, a marker of cells of mononuclear phagocyte lineage. The highly conserved super-enhancer (fms-intronic regulatory element, FIRE) that controls Csf1r expression (17) showed robust SAR-seq and ddC S1-END-seq peaks (Fig. 3E). DNA repair, SSB formation, and enhancer activity, as well as Csf1r gene expression, were impaired in the absence of TDG (Fig. 3E).

Because MAC-1⁺TDG^−/− cells showed a defect in phagocytosis, we focused on genes known to be critical for macrophage activation (Fig. 3F). Most of these genes were up-regulated in differentiating WT cells but failed to increase in TDG^−/− cells (Fig. 3F). Notable among them were TLR4 (Fig. 3F) and CD14 (fig S3C), which encode for the innate immune receptor complex that recognizes the lipopolysaccharide (LPS) cell wall component of Gram-negative bacteria including E. coli. Thus, active DNA demethylation of lineage-specific enhancers is required for proper up-regulation of genes critical for terminal differentiation and activation of macrophages.

In vivo long-patch tract lengths

TET activity is targeted to methylated CG dinucleotides within enhancers (5, 6). Although ddN (and ddC) S1-END-seq demonstrated the expected prevalence of cytosines at S1-END-seq summits, composite motif analysis failed to reveal CG dinucleotides at ddC S1-END-seq peak summits (4). The lack of a CG motif could result from ddC incorporation during LP-BER distal from the initiating methylated CG dinucleotide (Fig. 1A and fig. S4A).

To test this idea, we developed a genomewide base resolution assay to map 5fC and 5caC residues excised by TDG (fig. S4A). In this method, termed oxEND-seq, 5fC/caC sites are cleaved into double-strand breaks (DSBs) and then detected by END-seq. DNA is first treated with pyridine borane (PB) which converts 5fC and 5caC into dihydrouracil. USER enzyme, a mixture of uracil DNA glycosylase (which excises the uracil base) and endonuclease VIII (which breaks the phosphodiester backbone with lyase activity), generates a single-nucleotide gap. S1 nuclease will then cleave these 5fC:G- or 5caC:G-formed gaps to generate a DSB (fig. S4A). We applied this technique in TDG-knockout iMs, in which 5fC/5caC bases accumulate but SSBs do not form (fig. S2D). OxEND-seq revealed peaks that frequently overlapped with sites of SAR-seq and ddC S1-END-seq (fig. S4B). Motif analysis confirmed the expected pattern of CG dinucleotides at oxEND-seq peak summits (fig. S4C). We then calculated the relative distance of 5fC/caC to the nearest ddC S1-END-seq summit (fig. S4D). We observed a distribution of ddC S1-END-seq summits distal from the initiating CG dinucleotide. Approximately 17% of SSBs were located precisely at the CG dinucleotide, which reflect S1-END-seq detection of SP-BER events (Fig. 1A and fig. S4D). However, 67% were found within 30 residues from the TDG^−/− oxEND-seq summits, corresponding to long-patch repair (Fig. 1A and fig. S4D). Thus, tract lengths vary considerably in vivo, but most are within 30 base pairs (bp) of the initiating lesion.

Chain termination with cytosine analogs triggers LP-BER and the p53 response

Incorporation of ddC at macrophage and neuronal enhancers results in a single-strand DNA (ssDNA) gap because a phosphodiester bond cannot be formed between the missing 3′ OH group and the next nucleotide. We found that, like PARPi [which triggers LP-BER (4)], ddC treatment increased DNA synthesis–associated repair in both iN and iM (Fig. 4A). Moreover, PARPi and ddC SAR-seq peaks overlapped (fig. S5A). This suggests the possibility that chain termination provokes LP-SSB repair (fig. S5B). Consistent with this, we found that other cytosine analogs, such as arabinosylcytosine (Ara-C) and gemcitabine (GEM), similarly increased the intensity of SAR-seq peaks at enhancer sites (Fig. 4B). By contrast, adenine analogs, Ara-A and ddA, did not result in any change in SAR-seq (fig. S5, C and D). We speculate that when antimetabolites (ddC, Ara-C or GEM) are incorporated into DNA during SP-BER, unligated gaps are detected, leading to nucleoside analog excision, and processing by long-patch repair, thereby explaining the observed increase in DNA synthesis–associated repair at enhancers (fig. S5B).

Fig. 4. Anti-C metabolites enhance repair synthesis and trigger the DNA damage response.

(A) Heatmaps illustrating increased SAR-seq upon ddC treatment in iN and iM. Top, aggregate plots of SAR-seq intensity. (B) Heatmaps illustrating increased SAR-seq upon Ara-C and gemcitabine treatment of iN. Top, aggregate plots of SAR-seq intensity. (C) Gene Ontology analysis showing the enrichment of p53 target genes differentially expressed (|fold change| > 2) upon ddC, Ara-C, and Ara-A treatment in iN. The x axis represents the enrichment value as the logarithm of false discovery rate (FDR). (D) Survival of iN treated with Ara-C, dTAG alone, or Ara-C upon TDG depletion (Ara-C+dTAG). Data represent viability relative to nontreated cells. (E) Representative images (left) and quantification (right) of γH2AX foci induced upon ddC or Ara-C treatment in TDG^degron iNs treated with or without dTAG on day 6. 4’,6-diamidino-2-phenylindole (DAPI) (blue); γH2AX (red). Nontreated (NT). Scale bar, 10 μm. Statistical significance was determined using Mann-Whitney test (****p < 0.0001). (F) Maximum intensity projection of a confocal z-stack image of γH2AX (red) and 53BP1 (green) foci induced with ddC in TDG^degron iNs treated with or without dTAG on day 30. DAPI (blue) shows nuclei and overlaid differential interference contrast image (gray) shows neurites projected from the cell body. Scale bar, 10 μm.

Antimetabolites inhibit replication in mitotic cells and are thereby frequently used to treat cancer. Cancer treatments are also commonly associated with neurotoxicity through unknown mechanisms (18). Ara-C–induced cell death is dependent on p53 (19, 20), suggesting that Ara-C may kill neurons by a DNA damage–activated p53-dependent pathway. Consistently, we observed a robust p53 transcriptional response when neurons were treated with ddC and Ara-C, but not when treated with Ara-A (Fig. 4C). Moreover, 6 days after Ara-C treatment, almost all neurons in the culture had died (Fig. 4D). Acute degradation of TDG prevented Ara-C–induced cell death (Fig. 4D). This suggests that antimetabolite-induced neurotoxicity is linked to TET-initiated active DNA demethylation.

Incorporation of nucleoside analogs triggers the DNA damage response

A recent study demonstrated that Ara-C triggers histone H2AX phosphorylation in primary hippocampal neurons (21). Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) did not reveal DSBs in the nuclei of γ-H2AX–positive neurons (21). We also observed γ-H2AX and 53BP1 foci formation in almost 100% of iNs after Ara-C or ddC treatment (Fig. 4E and fig. S5E). ddC treatment led to only a small increase in DSBs revealed by END-seq signal, but a marked increase in S1-END-seq (fig. S5F), indicating a greater number of SSBs than DSBs. In line with ssDNA damage, γ-H2AX formation was largely ATR dependent (fig. S5G). Moreover, ddC-induced S1-END-seq and γ-H2AX formation required POLβ (fig S5H), likely because POLβ is highly selective for ddN (22). Finally, we observed that γ-H2AX/53BP1 formation was entirely dependent on TDG (Fig. 4E and fig. S5E). The low frequency of DSBs relative to SSBs is consistent with the finding that DSBs rarely arise from symmetrically methylated CGs because of the highly coordinated and sequential action of TET-TDG–mediated base excision repair (23).

On average, six and eight γ-H2AX and 53BP1 foci were detectable alter overnight treatment with ddC and Ara-C, respectively (Fig. 4E). Foci appeared within 2 hours of ddC and Ara-C treatment and were detectable for at least 16 hours after drug withdrawal (fig. S5I). This suggests that DNA damage either persists or is continually generated. If DNA lesions are continually produced by TET-mediated active DNA demethylation but subsequently repaired, acute degradation of TDG after drug withdrawal would lead to the disappearance of ddC and Ara-C–induced foci. Indeed, DNA damage no longer persisted when TDG was eliminated after the drugs were withdrawn (fig. S5J).

To directly monitor TET activity and DNA repair in living cells, we expressed an mCherry-53BP1 reporter (24) in neurons. Upon ddC treatment, we observed that 53BP1 foci appeared, dissolved, and then reappeared throughout the 15-hour time course (movie S1). By tracking individual foci from when they first appeared to when they disappeared, we estimate that most ddC-induced DNA damage events are resolved within 1 to 2 hours (fig. S5K).

Active DNA demethylation in fully differentiated iNs

Induced degradation of TDG on day 7 after iN differentiation led to the disappearance of SSBs, demonstrating the existence of ongoing TET-mediated oxidation. Within 14 to 28 days of maturation, iNs display differentiated neuron markers, action potential firing, and spontaneous synaptic currents, suggesting that they are functional excitatory glutamine-releasing neurons (8). Consistent with these findings, the analysis of gene expression profiles during iN differentiation revealed that the transcriptome continued to change beyond day 7, but fewer changes were detected between days 17 and 30 (fig. S6A). Reminiscent of our finding in iMs, genes were similarly down-regulated in WT and TDG^−/− iN throughout their maturation (fig. S6B). However, TDG knockouts partially impaired the up-regulation of genes that were induced after day 7 (fig. S6B), including those regulating presynaptic signaling (fig. S6 C). Even on day 30, we observed robust generation of ddC-induced γ-H2AX and 53BP1 formation in the vast majority (>95%) of iNs, which was abolished by acute degradation of TDG (Fig. 4F). Thus, TDG-mediated SSBs are generated in both differentiating and fully differentiated iNs.

Discussion

TET is highly active in postmitotic neurons (25, 26) and to a lesser degree in other somatic cell types (6). An unsolved question is the extent to which replication-independent demethylation contributes to cell differentiation and function. We have shown that TDG-dependent SSB intermediates accumulate at high levels at lineage-specifying enhancers in iPSC-derived neurons and transdifferentiated macrophages. Moreover, TDG contributes to transcriptome reprogramming in both differentiation systems. There are other examples of DNA demethylation in postmitotic cells in which gene expression is sensitive to loss of TDG. Axonal injury in retinal ganglion neurons (27) and dorsal root ganglion neurons (28) increases TET- and TDG-dependent active DNA demethylation to induce regeneration-associated gene expression. However, in other cases, TDG does not seem to contribute to transcriptional changes downstream of TET-mediated oxidation. For example, LPS triggers cell cycle exit and 5hmC accumulation in bone-marrow derived macrophages (BMDM) (13). The induced transcriptional program is independent of TDG even though 5fC/5caC accumulates at the top latent enhancers (Batf, Mdfic, Il1b and Il6) that acquire H3K27ac after stimulation (13). Consistent with this, we observed that LPS induces SSBs at these latent enhancers in BMDM (fig S7). TDG is also dispensable for the active DNA demethylation of 5mC in mouse zygotes (29). Removal of oxidized methylcytosines could potentially be mediated by other BER glycosylases. Alternatively, pathways that do not generate DNA breaks might mediate active DNA demethylation, including direct dehydroxymethylation of 5hmC (30), decarboxylation of 5caC (31), or deformylation of 5fC (32). By using high-resolution techniques to trace SSB intermediates and SP- versus LP-BER pathways, it should be possible to clarify the physiological role of active DNA demethylation during cell differentiation, activation, and injury.

Anticancer drugs frequently produce acute and sometimes persistent neurological and neuropsychiatric symptoms referred to as “chemobrain” (18). To date, there are no effective treatments or preventive measures for chemobrain, and the underlying mechanisms are poorly understood. We speculate that perturbed gap-filling synthesis at regulatory elements during active DNA demethylation could be one potential mechanism contributing to chemobrain. Notably, Ara-C is the most common chemotherapeutic agent that induces cerebellar dysfunction (33), and some patients have permanent impairment due to Purkinje cell loss in the cerebellum (34). Moreover, recent studies provide evidence that DNA demethylation is highly active in Purkinje neurons (26). Considering our findings, it would be interesting to determine whether inhibitors of TDG, or inhibitors of additional pathway components that trigger the DNA damage response, could be promising candidates to alleviate some of the neurological complications associated with anticancer drug therapies. Note added in proof: A recent study demonstrated that SAR sites are enriched for somatic mutations detected by ultradeep genome sequencing of individual neurons from normal individuals, suggesting that active DNA demethylation may contribute to mutation (35).

Data and materials availability:

The deep sequencing datasets have been deposited in the Gene Expression Omnibus (GEO) database (GSE210317).