SUMOylation coordinates BERosome assembly in active DNA demethylation during cell differentiation

Abstract

During active DNA demethylation, 5‐methylcytosine (5mC) is oxidized by TET proteins to 5‐formyl‐/5‐carboxylcytosine (5fC/5caC) for replacement by unmethylated C by TDG‐initiated DNA base excision repair (BER). Base excision generates fragile abasic sites (AP‐sites) in DNA and has to be coordinated with subsequent repair steps to limit accumulation of genome destabilizing secondary DNA lesions. Here, we show that 5fC/5caC is generated at a high rate in genomes of differentiating mouse embryonic stem cells and that SUMOylation and the BER protein XRCC1 play critical roles in orchestrating TDG‐initiated BER of these lesions. SUMOylation of XRCC1 facilitates physical interaction with TDG and promotes the assembly of a TDG‐BER core complex. Within this TDG‐BERosome, SUMO is transferred from XRCC1 and coupled to the SUMO acceptor lysine in TDG, promoting its dissociation while assuring the engagement of the BER machinery to complete demethylation. Although well‐studied, the biological importance of TDG SUMOylation has remained obscure. Here, we demonstrate that SUMOylation of TDG suppresses DNA strand‐break accumulation and toxicity to PARP inhibition in differentiating mESCs and is essential for neural lineage commitment.

Introduction

A well‐established pathway of DNA demethylation operates through oxidation of 5‐methylcytosine (5mC) to 5‐formylcytosine (5fC) or 5‐carboxylcytosine (5caC) by ten‐eleven translocation (TET) proteins and the replacement of these bases with unmodified cytosine by thymine DNA glycosylase (TDG)‐initiated DNA base excision repair (TDG‐BER; Wu & Zhang, 2014). Base excision by TDG generates abasic sites (AP‐sites) that are processed by AP endonuclease (e.g., APE1) to DNA single‐strand breaks (SSBs; Waters & Swann, 1998; Jacobs & Schar, 2012). These SSBs attract and activate poly(ADP‐ribose) polymerase 1 (PARP1), providing chromatin accessibility for association of X‐ray repair cross‐complementing protein 1 (XRCC1), which, through its scaffold function, coordinates DNA end editing, gap‐filling and ligation involving DNA polymerase β (POLβ) and DNA ligase III (LIGIII; Caldecott, 2014). Mouse genetic work demonstrated that TDG (Cortazar et al, 2011; Cortellino et al, 2011), but no other DNA glycosylase, and the core BER proteins (APE1, XRCC1, POLβ, LIGIII) are essential for embryonic development, indicating that active DNA demethylation engaged BER plays an important role in cell differentiation (Menisser‐de Murcia et al, 2001; Wallace et al, 2012).

Excessive TDG activity was shown to cause cell death by generation of cytotoxic DNA strand‐breaks (Kunz et al, 2009). Quantitation of modified cytosines by liquid chromatography–tandem mass spectrometry (LC‐MS/MS) showed that 5fC and 5caC are generated at a steady state of 5.3 × 10⁻⁶ nucleotides in mouse embryonic stem cells (mESCs; Appendix Table S1). This translates into 60,000 DNA base derivatives triggering TDG‐BER events in every cell at any one time, and this number increases above 95,000 upon induction of mESC differentiation. Since 5mC oxidation to 5fC and 5caC occurs primarily at gene regulatory regions (Raiber et al, 2012), i.e., in a small fraction of the genome, the ensuing TDG‐BER activity will generate spatiotemporally clustered DNA breaks (Weber et al, 2016), posing a considerable cytotoxic and mutagenic threat. The maintenance of genome integrity under conditions of widespread active DNA demethylation thus generates a specific need for mechanisms that couple the excision of 5fC/5caC by TDG to the repair of the resulting AP‐sites.

Biochemical work showed that the modification of TDG with small ubiquitin‐like modifiers (SUMOs) can regulate its dissociation from AP‐sites, and it was proposed that the timing of this event may provide an opportunity to coordinate AP‐site release with downstream BER (Hardeland et al, 2002). SUMO modification has been functionally associated with DNA repair and DNA methylation (Hendriks & Vertegaal, 2016), where it was shown to control protein stability and action and/or to promote the formation of DNA repair factories in response to DNA damage (Psakhye & Jentsch, 2012; Jackson & Durocher, 2013). Notably, SUMO‐conjugation was also associated with stem cell renewal and differentiation and reported to be essential for embryonic development (Nacerddine et al, 2005; Liu et al, 2014). While these observations point to an important role of SUMO modifications in developmentally controlled DNA transactions, the underlying molecular events and pathways have not been identified. While physical and functional interactions of TDG with SUMO are well‐described (Hardeland et al, 2002; Baba et al, 2005; Steinacher & Schar, 2005; Mohan et al, 2007; Smet‐Nocca et al, 2011; Coey et al, 2014), neither their biological role nor their precise molecular function in the context of BER‐mediated active DNA demethylation has been resolved. We therefore investigated whether and how SUMO is involved in TDG‐BER‐dependent processing of TET generated DNA demethylation intermediates in vitro and during mESC differentiation. Our results establish a central role for SUMO modifications and interactions in orchestrating TDG‐BER and describe a molecular pathway for SUMO‐mediated coupling of DNA base excision with AP‐site repair that is critical during neural differentiation.

Results

TDG and XRCC1 are SUMO targets in mESCs

To explore the role of SUMO in TDG‐mediated active DNA demethylation, we first assessed the SUMO modification (SUMOylation) of relevant BER proteins in differentiating mESCs, making use of Tdg ^−/− mESCs complemented with Tdg mini‐gene variants expressing endogenous levels of either wild‐type (TDGwt), catalytic inactive TDG (TDGN151A, TDGcat), SUMOylation‐deficient TDG (TDGK341R, TDGsnm), or the vector only (TDGnull; Fig EV1A; Cortazar et al, 2011). Immunoblotting of whole‐cell extracts for TDG, XRCC1, APE1, LIGIII and POLβ revealed SUMO modification of TDG and the BER scaffold XRCC1 (Fig 1). TDG‐SUMO conjugates were detectable in pluripotent TDGwt mESCs, cultured in 2i medium, but not in TDGsnm cells (Figs 1A and EV1A). Upon passage of cells to differentiation‐inducing medium containing all‐trans retinoic acid (RA, 5 μM), TDG‐SUMO conjugate levels first decreased and then increased again (Fig 1A). These changes are relatively small, and their significance in the context of ESC differentiation remains uncertain. Yet, TDG‐SUMO conjugates also accumulated following pulse treatment of mESCs with H₂O₂ (Fig 1B), supporting that oxidative stress can trigger SUMOylation of TDG. SUMO modification causes TDG to dissociate from DNA (Hardeland et al, 2002; Steinacher & Schar, 2005) and to change its subnuclear localization (Mohan et al, 2007). Nuclear fractionation showed that unmodified TDGwt and TDGsnm located preferentially in the chromatin fractions (Chr1, Chr2), whereas SUMOylated TDG was exclusively in the chromatin‐unbound fraction (Fig 1C). Quantitation of the immunoblot signals of TDG confirmed that the fraction of TDG associated with chromatin (Chr1 and Chr2 fractions) is higher for TDGsnm (64%) than for TDGwt (43%; Fig EV1B), demonstrating that TDGsnm enriches in chromatin. These data are consistent with an earlier observation in Xenopus, showing that non‐SUMOylatable TDG accumulates on chromatin (Slenn et al, 2014).

Figure EV1. TDG and XRCC1 are SUMO‐modified at conserved SUMO acceptor lysines.

1. Immunoblot of mESC Tdg^−/− cells stably complemented with TDGwt, TDGcat or TDGsnm minigenes. Expression levels relative to a β‐actin control were similar for TDGwt, TDGcat and TDGsnm. TDGwt and TDGcat were SUMOylated. No SUMOylation of TDGsnm was detected.
2. Quantitation of all chromatin‐bound (Chr1 + 2) or soluble (Sol) TDGwt and TDGsnm.
3. SUMO consensus and non‐consensus sequences of human and mouse XRCC1. Ψ: aliphatic amino acid, K: conserved SUMO acceptor lysine. Immunoblotting revealed efficient SUMOylation of the XRCC1wt but was strongly reduced in the XRCC1K176R mutant.
4. Ponceau and immunoblot analysis of total protein and TDG/TDG‐SUMO in mESCs complemented with empty vector, GfpTdgwt (GFP‐TDGwt) GfpTdgpip (GFP‐TDGpip) and Tdgwt (TDGwt) minigenes.
5. Immunoblot analysis of TDG/TDG‐SUMO in GFP‐TDGwt and GFP‐TDGpip mESCs.
6. Immunoblot analysis XRCC1/XRCC1‐SUMO and Ponceau staining of total proteins in GFP‐TDGwt and GFP‐TDGpip mESCs. Quantitation of immunoblot was performed from exposed film using Fiji image processing package.
7. Quantification of XRCC1 total protein in GFP‐TDGwt and GFP‐TDGpip mESCs. n = 3.

Data information: Means with standard deviation (SD). *P < 0.05. Nonparametric, Mann–Whitney test.Source data are available online for this figure.

Figure 1. TDG and XRCC1 are SUMO‐modified in mESCs during differentiation and oxidative stress.

Immunoblot analyses of TDG, XRCC1 and TDG‐SUMO, XRCC1‐SUMO, and TET1 isoforms in whole‐cell extracts of mESCs and MEFs.

1. TDG and TDG‐SUMO conjugates before (2i, 0 h) and during differentiation of TDG wild‐type (TDGwt) and TDG‐deficient (TDGnull) mESC and MEFs (5 μM all‐trans retinoic acid (RA, 24 h, 48 h). Percentages of TDG‐SUMO relative to total TDG signal are indicated.
2. TDG and TDG‐SUMO before and after H₂O₂ treatment (10 mM, 15 min).
3. Chromatin fractionation from mESCs expressing wild‐type (ESC‐TDGwt) or SUMOylation‐deficient TDG (ESC‐TDGsnm). Sol, chromatin‐unbound proteins; Chr1, chromatin‐associated proteins; Chr2, chromatin‐bound fraction; Histone H2B, stably chromatin‐bound marker.
4. XRCC1 and XRCC1‐SUMO conjugates before (2i, 0 h) and during RA‐induced differentiation (24 h, 48 h) in mESCs and MEFs. Percentages of XRCC1‐SUMO relative to total XRCC1 signal are indicated.
5. XRCC1 SUMOylation in native whole‐cell extracts before and after incubation with purified SUMO protease (SENP2_364–489).
6. XRCC1 SUMOylation before and after exposure to H₂O₂ (5 mM, 15 min).
7. TET1 and TET1s isoform protein levels in mESCs during RA‐induced differentiation and in MEFs.
8. Fluorescence microscopy of GFP‐TDGwt GFP‐TDGpip in mESCs. Scale bar: 100 μM.
9. Quantitation of fluorescence signals of GFP‐TDGwt and GFP‐TDGpip in mESCs. Median with interquartile range. > 470 cells were analysed of each GFP‐TDGwt and GFP‐TDGpip ESCs. ***P < 0.0001. Nonparametric, Mann–Whitney test.
10. Immunoblot analysis of XRCC1 and XRCC1‐SUMO in GFP‐TDGwt and GFP‐TDGpip mESCs.

Data information: In (A and D), membranes were sequentially probed with anti‐TDG, anti‐XRCC1 and anti‐ β‐actin antibodies. Quantitation of immunoblots in (A, D and J) was performed from exposed film using Fiji image processing package.Source data are available online for this figure.

XRCC1 SUMOylation was noticeable as a < 150‐kDa protein in mESC extracts cross‐reacting with the XRCC1 antibody in immunoblots (Fig 1D). Due to the branched structure of SUMO‐conjugated proteins, modification with a single SUMO can give rise to a migration shift in gel electrophoresis higher than the 10 kDa expected from the molecular weight of SUMO peptides. Consistent with the < 150‐kDa protein representing SUMO‐conjugated XRCC1, the signal disappeared upon incubation of the extracts with the SUMO protease SENP2 (SENP2_364–489; Reverter & Lima, 2006; Fig 1E), and in vitro SUMOylation of recombinant XRCC1 (Steinacher & Schar, 2005) yielded a protein conjugate migrating at the same position (Fig EV1C; Weber et al, 2014). Exposure of mESC to H₂O₂ induced a significant enrichment of < 150‐kDa XRCC1 conjugates (Fig 1F), suggesting that XRCC1, like TDG, can be modified by DNA lesions generated by general oxidative stress. A main source of oxidative DNA attack in unperturbed mESCs is the action of the highly expressed TET1 and TET2 proteins (Koh et al, 2011; Neri et al, 2015; Zhang et al, 2016), as indicated by relatively high levels of the oxidative 5mC derivatives, 5‐hydroxymethyl‐C (5hmC), 5fC and 5caC in such cells (Fig 4D). High levels of full‐length TET1 protein and its shorter isoform TET1s (Fig 1G) in ESCs also correlated with higher steady‐state levels of XRCC1‐SUMO in comparison with mouse embryonic fibroblasts (MEFs; Fig 1D and G), indicating that TET‐dependent base oxidation may promote XRCC1 SUMOylation. XRCC1‐SUMO conjugates also increased (7–13%) upon transfer of the TDGwt mESCs from 2i medium to differentiating conditions (Fig 1D), along with increased formation of 5fC and 5caC, best seen in TDG‐deficient cells (Fig 4D). To investigate whether TDG activity may trigger XRCC1 SUMOylation, we complemented Tdg ^−/− mESCs with GfpTdg mini‐gene variants expressing from the authentic Tdg promoter either wild‐type TDG (GFP‐TDGwt) or a variant with a mutated PCNA‐interacting peptide motif (PIP; GFP‐TDGpip). GFP‐TDGpip escapes PCNA and CRL4Cdt2 E3 ubiquitin ligase‐dependent degradation in S‐phase and therefore accumulates to slightly higher levels in cells (Fig 1H; Hardeland et al, 2007; Shibata et al, 2014; Slenn et al, 2014). mESC expressing GFP‐TDGpip had moderately but significantly increased TDG levels (median, > 1.25‐fold increase) when compared to GFP‐TDGwt or endogenous TDG‐expressing cells, all cultivated in 2i (Figs 1I, and EV1C and D). Increased GFP‐TDGpip levels correlated with elevated (> 1.6‐fold) XRCC1 protein (Figs 1J, and EV1F and G) and significantly increased XRCC1 SUMOylation in the comparison of GFP‐TDGpip‐ and GFP‐TDGwt‐expressing mESCs (Figs 1J and EV1F). From these results, we conclude that XRCC1 SUMOylation is comparably high in mESCs, correlates with high TDG activity and is inducible by TDG, indicating that XRCC1 SUMOylation may be triggered as a consequence of TET‐TDG‐initiated BER.

SUMOylation of XRCC1 promotes the formation of a TDG‐BERosome

Next, we examined the effect of SUMO modification on physical interactions of XRCC1 with TDG and the downstream acting BER enzymes APE1, POLβ, LIGIII (Fig EV2A). “Far‐Western” blotting confirmed known interactions of XRCC1 with POLβ and LIGIII but did not detect an interaction with TDG (Fig 2A; Caldecott et al, 1996, 1994; Dianova et al, 2004; Kubota et al, 1996). As TDG possesses a SUMO interaction motif (SIM; Song et al, 2004) that binds free SUMO and SUMO‐conjugated proteins (Hardeland et al, 2002), we investigated whether SUMO‐modified XRCC1 can mediate interactions between TDG and the core BER machinery. “Far‐Western” analyses with purified proteins (Fig EV2A) indicated that SUMO modification enables XRCC1 to interact with TDG, while it does not affect its interactions with POLβ and LIGIII (Fig 2A). Neither unmodified nor SUMOylated XRCC1 showed an interaction with APE1 under the DNA‐free conditions of the assay. We then investigated the interaction of endogenous TDG with XRCC1 or XRCC1‐SUMO. As chromatin‐associated TDG (Fig 1C), unlike the chromatin‐unbound fraction, is notoriously inaccessible to immunoprecipitation, and as the fraction of SUMOylated endogenous XRCC1 in unperturbed mESCs is too small (< 10%) to enrich detectable amounts of TDG, we followed an affinity purification approach. We immobilized purified, in vitro SUMOylated 6hisXRCC1‐SUMO (> 80% SUMOylated) and unmodified 6hisXRCC1 on Ni‐NTA beads and incubated them with mESC extract. Notably, XRCC1 SUMOylation in vitro can produce small SUMO chains of two or three molecules, all attached to the major conjugation site K176 (Fig EV1C), resulting in accordingly higher migrating protein bands on immunoblots (Fig 2B). Ni‐NTA beads coated with 6hisXRCC1‐SUMO enriched detectable amounts of unmodified endogenous TDG, whereas 6hisXRCC1‐coupled beads failed to do so (Fig 2B), supporting a specific interaction of endogenous TDG with XRCC1‐SUMO. Next, to test if SUMO modification can mediate the formation of a BER complex on a DNA demethylation substrate, we assayed the concerted binding of TDG, XRCC1, POLβ and LIGIII to biotinylated DNA duplexes containing a 5caC base. In this experimental setup, TDG will bind and immediately excise 5caC and then remain bound to the AP‐site (Hardeland et al, 2002). Accordingly, TDG binding was obvious but neither XRCC1, nor POLβ or LIGIII was detectable in the streptavidin precipitate of DNA substrate (Fig 2C). Performing the experiment under exact same conditions with in vitro SUMOylated XRCC1 (50% XRCC1‐SUMO), however, resulted in an efficient and specific binding of all, XRCC1‐SUMO, POLβ and DNA LIGIII to the TDG‐bound substrate (Fig 2D). Notably, despite the apparent SUMO‐XRCC1 dependency, we were able to detect unmodified XRCC1 to enrich on TDG‐bound 5caC substrate (Fig 2D). This, however, was observed only when unmodified XRCC1 was added in a mixture with SUMOylated XRCC1 (as produced upon in vitro SUMOylation) and can be explained by the tendency of XRCC1 to homodimerize (Moor et al, 2015), i.e., to form XRCC1‐SUMO/XRCC1 dimers in mixtures of modified and unmodified protein. We thus conclude that TDG interacts specifically with XRCC1‐SUMO, which may also interact with, and carry along unmodified XRCC1 (Moor et al, 2015).

Figure EV2. SUMOylation facilitates XRCC1 TDG interaction.

1. Purified BER proteins were separated on a denaturing 4–15% SDS–PAGE. Coomassie staining revealed homogeneous protein fractions of TDG‐SUMO, TDG, POLβ, APE1, LIGIII, XRCC1 and XRCC1‐SUMO.
2. “Far‐Western” blot analysis. 10 μg of recombinant purified human TDG, TDG‐SUMO or 1 μg of POLβ, LIGIII BER proteins was dot blotted onto a nitrocellulose membrane and hybridized with purified human XRCC1‐SUMO or XRCC1 (20 μg). XRCC1 and XRCC1‐SUMO bound to BER proteins were analysed by immunoblotting against XRCC1. XRCC1‐SUMO interacted with TDG, POLβ, LIGIII but not with TDG‐SUMO. XRCC1 interacted with POLβ, LIGIII but not with TDG and TDG‐SUMO.
3. Protein interaction analysis. TDG and TDG‐SUMO were incubated with SUMO or BSA coated Affi‐Gel10 beads. I, Input; LW, Last Wash; E, Elution.
4. Schematic of full‐size TDG. Catalytic core (blue), SUMO interaction motif (SIM, orange), SUMO acceptor site (K330, red). Interaction study of purified recombinant wild‐type TDG (TDGwt) or TDG‐SUMO‐binding mutant (TDGE310Q) with 6hisSUMO. 6hisSUMO was coupled to Ni‐NTA beads and incubated with TDGwt or TDGE301E. Shown is an immunoblot against TDG and SUMO. Input (I), Flow (FL). Last wash (LW), Elution (E).

Source data are available online for this figure.

Figure 2. SUMOylation of XRCC1 facilitates TDG‐BERosome assembly.

• A
  “Far‐Western” blot analyses of BER protein interactions. Recombinant purified human TDG (10 μg), POLβ (1 μg), APE1 (1 μg), and LIGIII (1 μg) probed with XRCC1 and XRCC1‐SUMO (20 μg) and immune‐stained for XRCC1.
• B
  Immunoblot analysis of interactions of XRCC1 and XRCC1‐SUMO with endogenous TDG. Purified 6hisXRCC1 or in vitro‐modified and in vitro‐purified 6hisXRCC1‐SUMO was coupled to Ni‐NTA agarose and incubated with mESC extracts. Input (I), last wash (LW) and elution (E) fractions.
• C–H
  (C–F and H) Scheme of pulldown experiments and immunoblot analysis of purified 6hisTDG, 6hisXRCC1/XRCC1‐SUMO and 6hisBER proteins bound to streptavidin (St) biotin‐5caC DNA substrate. (C) 0.5 μM TDGwt was incubated with biotinylated 5caC substrate (G•5caC DNA, 1 μM) and homoduplex competitor DNA (G•C DNA, 10 μM) and equimolar amounts of XRCC1, POLβ and LIGIII proteins. (D) Schematic of biotin‐5caC DNA substrate pulldown experiment. 0.5 μM TDGwt or TDGcat (TDG(N140A)) incubated with biotinylated 5caC substrate (G•5caC DNA, 1 μM) and homoduplex competitor DNA (G•C DNA, 10 μM) as indicated and equimolar amounts of XRCC1‐SUMO (> 50% SUMOylated), POLβ and LIGIII. (E) 1 μM biotinylated 5caC substrate (G•5caC DNA) and 10 μM homoduplex competitor DNA (G•C DNA) were incubated with XRCC1‐SUMO (50% SUMOylated) and BER proteins as indicated. (F) 0.5 μM human TDGwt protein were incubated with 1 μM biotinylated 5caC substrate (G•5caC DNA), 0.5 μM SUMO and 10 μM homoduplex competitor DNA (G•C DNA) as indicated. Shown is an immunoblot analysis of TDG and free SUMO. (G) “Far‐Western” blot analyses of TDGwt and TDGE310Q with 6hisXRCC1 and 6hisXRCC1‐SUMO. TDGwt or TDGE310Q was dot blotted on nitrocellulose membranes and probed with XRCC1 or XRCC1‐SUMO (probe:XRCC1 or XRCC1SUMO) where indicated. TDG and XRCC1 probes were detected by immunoblotting with anti‐TDG and anti XRCC1 antibodies as indicated. (H) Schematic of biotin‐5caC DNA substrate pulldown experiment. Immunoblot analysis of purified TDGwt or TDG‐SUMO‐binding mutant (TDGE310Q), 6hisXRCC1/XRCC1‐SUMO and 6hisBER proteins bound to streptavidin (St) biotin‐5caC DNA substrate. 0.5 μM TDGwt or TDG‐SUMO‐binding mutant (TDGE310Q) was incubated with biotinylated 5caC substrate (G•5caC DNA, 1 μM) and homoduplex competitor DNA (G•C DNA, 10 μM) as indicated and equimolar amounts of XRCC1, POLβ, APE1 and LIGIII proteins. Grey ovals, BERosome.

Source data are available online for this figure.

The assembly of the core BER machinery on substrate DNA was independent of the catalytic activity of TDG but fully dependent on its structural presence; TDGcat was as capable of assembling the BER complex as TDGwt (Fig 2D) but neither XRCC1‐SUMO, nor POLβ or LIGIII bound on their own to the 5caC substrate (Fig 2E). In line with previous observations (Mohan et al, 2007), free SUMO peptide failed to interact with substrate‐bound TDG (Fig 2F), corroborating that the formation of the TDG‐dependent BER complex is not mediated by free SUMO but depends on the conjugation of SUMO to XRCC1. Also, free SUMO did not bind SUMOylated TDG (Fig EV2C), suggesting that the conjugated SUMO occupies the SUMO interaction site(s) of TDG (Baba et al, 2005). TDG possesses two SUMO interaction motifs (SIMs), one near the active site and another at the carboxy‐terminus. Both are involved in the binding of free SUMO, in SUMO modification and in the interaction with other SUMOylated proteins (Baba et al, 2005; Mohan et al, 2007). Mutating the SIM1 (D133A) generates a misfolded TDG that aggregates, whereas mutation of SIM2 (E301Q) causes loss of SUMO binding without structurally disrupting TDG (Mohan et al, 2007; Smet‐Nocca et al, 2011). We therefore used SIM2‐mutated TDG to interrogate the role of non‐covalent SUMO interaction in the assembly of the BER complex at a TDG‐bound DNA substrate. Consistent with earlier findings, we found that mutating SIM2 (E310Q) completely abolished TDG's ability to bind free SUMO (Fig EV2D; Mohan et al, 2007; Smet‐Nocca et al, 2011). Notably, TDGE301Q also failed to interact with XRCC1‐SUMO (Fig 2G) in a “Far‐Western” blotting setup, demonstrating that the SIM2 in TDG is required for the interaction with XRCC1‐SUMO. Moreover, while TDGE310Q and TDGwt bound comparably well to biotinylated DNA duplexes containing a 5caC base (Fig 2H, lane E), only TDGwt was able to enrich XRCC1‐SUMO upon DNA pulldown (Fig 2H, lane E).

APE1 was shown before to stimulate enzymatic turnover of TDG when present in molar excess (Waters et al, 1999). Since we were not able to detect an interaction of APE1 with TDG, XRCC1 or XRCC1‐SUMO in the absence of DNA (Fig 2A), nor to measure a notable effect on TDG turnover when APE1 was added in equimolar amounts (Hardeland et al, 2002), we wanted to assess the effect of APE1 on the formation and integrity of the TDG‐dependent BER complex. We found that at equimolar concentrations, APE1 does not assemble with the XRCC1‐SUMO‐dependent BER complex at DNA‐bound TDG, nor does APE1 notably affect the stability of the assembly, i.e., stimulate the displacement of the BER complex or parts of it from the DNA substrate (Fig 2H).

These results establish that SUMO‐mediated interactions involving TDG and SUMOylated XRCC1 can nucleate the assembly of a TDG‐“BERosome” on a DNA demethylation substrate. Importantly, as substrate binding by the glycosylase is sufficient, i.e., base excision not required, TDG‐BERosome formation appears to be an early event in BER, which does not include APE1 at this stage. Together with the observation that APE1 can stimulate enzymatic turnover of SUMOylated TDG at equimolar concentrations (Hardeland et al, 2002), these results suggest a mechanism whereby a TDG‐XRCC1‐SUMO‐BERosome is first formed on a DNA substrate and then, following SUMOylation of TDG, APE1 will be able to compete for and cleave the AP‐site, effecting TDG release and facilitating completion of BER. For this sophisticated mechanism to work, it would be required that XRCC1‐SUMO‐dependent TDG‐BERosome formation and TDG SUMOylation are linked processes.

SUMOylation within the TDG‐BERosome coordinates in vitro DNA demethylation

SUMO modification is a dynamic process controlled by SUMO‐activating (E1), SUMO‐conjugating (E2) and, in some cases, ligating enzymes (E3) that attach SUMOs to target proteins, and SUMO proteases that activate SUMO precursors and de‐conjugate modified proteins (Gill, 2004). To address the function of dynamic SUMOylation in TDG‐associated BER, we reconstituted a complete SUMO‐conjugation/de‐conjugation system with purified recombinant SUMO‐E1 and SUMO‐E2 enzymes, SAE1/SAE2 and UBC9, respectively, and the SUMO protease SENP2_364–489 (Reverter & Lima, 2006; Steinacher & Schar, 2005; Fig EV3A). We adjusted the system to achieve a robust steady state of detectable TDG‐SUMO modification with stoichiometric amounts of TDG and SUMO in the absence of DNA (Fig 3A). Biochemical work established that TDG (with a fully intact N‐terminus) excises substrate bases, including 5caC and 5fC, very rapidly upon DNA binding but then remains associated with the product AP‐site (Waters & Swann, 1998; Hardeland et al, 2000, 2002; Maiti & Drohat, 2011). SUMO modification reduces TDG's affinity for the AP‐site and facilitates its dissociation and, hence, the completion of BER (Hardeland et al, 2002). Therefore, the DNA intermediate relevant to study SUMO‐dependent coordination of TDG‐dependent BER is a TDG‐bound AP‐site. Addition of a fivefold molar excess of G•AP‐site DNA to the dynamic TDG SUMOylation system dramatically reduced the level of detectable TDG‐SUMO conjugates (Fig 3A), consistent with the inability of DNA‐bound TDG to interact with free SUMO (Fig 2F) and previous observations on the effect of DNA on TDG SUMOylation (Coey et al, 2014). This inhibitory effect of DNA was overcome by increasing the concentration of SUMO to a 20‐fold molar excess over TDG (Fig 3A), strongly indicating that DNA binding reduces TDGs affinity for free SUMO and, thereby, limits TDG SUMOylation, although an effect of an excess of free SUMO on SENP activity cannot be strictly excluded (Shen et al, 2006a,b). Addition of the BER proteins APE1, POLβ and LIGIII affected SUMOylation of DNA‐free TDG positively but failed to promote modification of DNA‐bound TDG (Fig 3A). These results show that TDG bound to G•AP‐site DNA is not SUMOylated in the presence of free SUMO peptides and suggest that additional affinity promoting factors are required.

Figure EV3. XRCC1 promotes TDG SUMOylation in ESCs.

1. Purified SUMO‐conjugation proteins. SAE1/SAE2, UBC9, SUMO and SUMO protease SENP2_364–489 were separated on a 4–15% SDS–PAGE. Proteins were visualized by Coomassie blue staining.
2. CRISPR/Cas9 XRCC1 deletion. Shown is the nucleotide sequence of the first coding exon (bold) and intron (italic) of XRCC1 with translation start (red) indicated and the gRNA sequences (5′ CCTccgccacgtcgtgtcctgca3′ and 5′ actcggtgaggggctgacgtGGG 3′) to target the nickases in close proximity. Amino acid sequence on top. Both alleles with CRISPR‐/Cas9‐induced nucleotide deletions (XRCC1null1: ∆15 bp— and ∆54 bp—; XRCC1null2 ∆55 bp— and ∆81 bp—), spanning the ATG start codon, are shown for XRCC1null1 and XRCC1null2. (Lower left) Immunoblot analysis of wild‐type (WT) and XRCC1 null (XRCC1null1, null2) mESC extracts and β‐actin loading control. (Lower right) Quantification of total XRCC1 protein and β‐actin (immunoblot analysis) in TDGwt and XRCC1 null (XRCC1null1, null2) mESC extracts. n ≥ 3. **P < 0.01, ****P ≤ 0.00001, one‐way ANOVA, Dunnett's multiple comparisons test.
3. Colony formation efficiency of XRCC1wt and XRCC1null1 and XRCC1null2 after irradiation with 1, 2, 5 Gy, relative to untreated controls (0 Gy). n = 3. Shown are means with standard deviation. **P < 0.01, ***P ≤ 0.0001, one‐way ANOVA, Dunnett's multiple comparisons test.
4. Immunoblot analysis of XRCC1, TDG/TDG‐SUMO in TDGwt, TDGsnm ESCs and XRCC1null mESCs (XRCC1null). Immunoblot identical to Fig 3D, but with two additional lanes showing analysis of XRCC1 and TDG/TDG‐SUMO in TDGwt ESCs.
5. Quantitation of TDG‐SUMO in % of total TDG protein in XRCC1wt and XRCC1null mESCs, either unchallenged (−), or H₂O₂ treated (+) as indicated. n ≥ 3. Shown are means with standard deviation (SD). *P ≤ 0.05, nonparametric, two‐tailed Mann–Whitney test.

Source data are available online for this figure.

Figure 3. SUMO transfer from XRCC1‐SUMO to TDG within the TDG‐BERosome.

1. Reconstitution SUMO‐conjugation/de‐conjugation reaction with purified SAE1/2, UBC9, SUMO protease SENP2_364–489), free SUMO in equimolar amounts and 20‐fold molar excess over TDG. TDG SUMOylation was performed in the presence or absence of AP‐site containing synthetic DNA double‐strands as indicated (G•AP). TDG and TDG‐SUMO were detected by immunoblotting. SUMOylation of DNA (G•AP) bound TDG is strongly reduced but a 20‐fold molar excess of free SUMO compensates the inhibitory effect of DNA.
2. SUMOylation assay as in (A) with SUMO provided as XRCC1‐SUMO instead of free SUMO.
3. SUMO transfer as depicted in (B) with additional APE1, POLβ, LIGIII and TDG.
4. Immunoblot analysis of XRCC1, TDG, TDG‐SUMO in TDGwt (wt), TDGsnm (snm) and XRCC1null mESCs (null).
5. Immunoblot analysis of TDG and TDG‐SUMO in TDGwt (wt) and XRCC1null mESCs (null) exposed to H₂O₂ as indicated.
6. Top: Schematic of the SUMO‐mediated TDG‐BERosome (grey oval) formation and DNA processing. Bottom: Reconstitution of active DNA demethylation on a 5caC containing synthetic DNA double‐strands in presence of dynamic SUMO‐conjugation/de‐conjugation. Left panels lanes 3–14: BER reconstitution with XRCC1 or XRCC1‐SUMO as indicated. Right panels lanes 15–21: BER reconstitution with free SUMO or XRCC1‐SUMO as SUMO donor as indicated. ds59merDNA substrates: lane 1–2 CG/CG HpaII methylation‐sensitive restriction site, cleavable by HpaII; lane 3–6, 8–14, 15–17, 21 HpaII methylation‐sensitive restriction site restriction site with a caC (5caCG/CG), not cleavable by HpaII; lane 7 and 20 repaired product (CC/CG) that is cleavable by HpaII after excision/repair of caC/CG to CG/CG by BER. Percentage of total signal of fully repaired product is indicated. ss59merDNA, single‐stranded DNA.

Source data are available online for this figure.

Given that XRCC1‐SUMO does interact with DNA‐bound TDG (Fig 2D and H), we tested whether SUMO might be transferred from XRCC1 to the DNA‐bound TDG. We reconstituted dynamic SUMO modification with XRCC1‐SUMO as a SUMO donor instead of free SUMO. De‐conjugation of XRCC1 and SUMO modification of TDG occurred efficiently in the presence of SAE1/2 and UBC9 and was dependent on the SUMO protease SENP2_364–489; XRCC1‐SUMO was fully de‐conjugated and the SUMO transferred to TDG at equilibrium of the reaction (Fig 3B). This showed that SUMO can be transferred from XRCC1 to TDG in a directional process in the presence of SUMO protease and SUMO‐E1 and SUMO‐E2 enzymes and in the absence of DNA. Addition of DNA did not affect de‐conjugation of XRCC1 but inhibited SUMOylation of TDG (Fig 3B), as observed in reactions with free SUMO peptide (Fig 3A). Thus, XRCC1‐SUMO and SENP, effectively liberating the SUMO moiety, are not sufficient to support SUMOylation of DNA‐bound TDG. Importantly, adding APE1, POLβ and DNA LIGIII in stoichiometric amounts with XRCC1‐SUMO significantly stimulated the transfer of SUMO from XRCC1 to TDG in the presence of DNA (Fig 3C); SUMO was cleaved from XRCC1 and transferred to TDG in a unidirectional manner with XRCC1 losing all detectable modification. Upon SUMO modification, TDG no longer interacted with XRCC1‐SUMO (Fig EV2B), nor did TDG‐SUMO bind free SUMO (Fig EV2C), suggesting that SUMO transfer from XRCC1 to TDG renders the glycosylase inert for interaction with the BER scaffold and, thereby, promotes its eviction from the BERosome. Altogether, these results establish that SUMO modification of DNA‐bound TDG cannot be achieved by free or XRCC1‐coupled SUMO and the SUMO‐conjugation/de‐conjugation system alone, but requires the proper assembly of a BERosome consisting of XRCC1‐SUMO/APE1/POLβ/LIGIII.

To corroborate the role of XRCC1 in promoting TDG SUMOylation in vivo, we generated XRCC1‐deficient mESCs, using the CRISPR/Cas9 technology (Fig EV3B). Two independently isolated XRCC1 clones showed greatly reduced protein levels (> 95%; Fig EV3B). Both XRCC1null (XRCCnull1/null2) mESC clones showed no overt phenotype under unperturbed culture conditions (Fig EV3B) but a modest hypersensitivity to γ‐irradiation (Fig EV3C). Immunoblot analyses of mESC whole‐cell extracts demonstrated that TDG SUMOylation (Fig 3D) was significantly reduced (3.8‐fold, Fig EV3D and E) in XRCC1null ESCs compared to TDGwt mESCs and not detectable in TDGsnm ESCs (Fig 3D). Treatment with H₂O₂ strongly induced SUMOylation of TDG in XRCC1wt (1.9‐fold) but also in XRCC1null cells (13‐fold; Fig 3E). This shows that a significant fraction (> 22%) of steady‐state SUMOylation of TDG in unchallenged mESCs depends on XRCC1, whereas TDG SUMOylation in response to H₂O₂ stress is independent of XRCC1. The effect of H₂O₂ exposure on SUMO modification of TDG and XRCC1 is much more pronounced than the changes observed during RA‐induced differentiation of the same cells (Figs 1A, B, D, F, and 3E). As H₂O₂ is well‐known to cause global oxidative damage to DNA and elsewhere and to induce DNA glycosylase action (e.g., MUTYH, OGG1) and BER (Xie et al, 2008), the inducibility of XRCC1 SUMOylation by H₂O₂ suggests that the role of SUMO modification in BER is not limited to oxidative DNA demethylation and may extend to the repair of general base damage.

To demonstrate the importance of SUMO transfer in TDG‐mediated DNA demethylation in vitro, we reconstituted the entire BER‐side of the process, using a DNA oligonucleotide substrate with a HpaII recognition site (CCGG) in a hemi‐carboxymethylated configuration (CcaCGG). While methylation‐sensitive HpaII does not cleave the CcaCGG sequence (Fig 3F, lane 4, 10, 16), it will do so after complete replacement of the 5caC with a C by BER. Reconstituting BER in a dynamic SUMO modification context with purified recombinant TDGwt, APE1, POLβ, XRCC1‐SUMO/LIGIII, SUMO‐E1 and SUMO‐E2 enzymes, SAE1/SAE2, UBC9 and the SUMO protease SENP2_364–489, yielded efficient 5caC repair, producing an appreciable amount of HpaII‐cleavable CCGG sites (demethylated and repaired sites; Fig 3F, lane 7, 20). TDG SUMOylation was critical in this process; when TDGwt was replaced with TDGsnm in the reaction, no HpaII‐cleavable sites were generated (Fig 3F, lane 5). Corroborating the requirement for the SUMO moiety being provided as XRCC1 conjugate for efficient transfer to TDG, SENP2 activity and XRCC1‐SUMO were likewise essential for 5caC repair to HpaII cleavable CCGG sites (Fig 3F, lane 8, lane 13). Reconstitution of TDG‐mediated demethylation with free SUMO instead of XRCC1‐SUMO and non‐SUMOylatable XRCC1snm (to avoid in situ XRCC1 modification) demonstrated that free SUMO on its own is not sufficient to promote TDG‐BER‐mediated replacement of 5caC with C (Fig 3F, lane 17). Altogether, this line of evidence establishes a mechanism whereby SUMO‐mediated interactions of XRCC1 with DNA‐bound TDG and SUMO transfer from XRCC1 to TDG are capable to orchestrate TDG‐BERosome dynamics that facilitate the coupling of base excision with AP‐site repair.

SUMO modification is essential for TDG function in neural cell differentiation

Having shown in vitro that XRCC1‐SUMO‐mediated TDG SUMOylation is important for BER of 5caC, we aimed to establish its biological relevance in the context of mESC differentiation. We subjected mESCs expressing either wild‐type, SUMOylation‐deficient or a catalytic inactive TDG (Fig EV1A) to embryoid body (EB) formation and RA‐induced differentiation towards the neural lineage (Cortazar et al, 2011; Fig 4A and B). TDG protein is expressed to the same levels in TDGwt, TDGcat and TDGsnm and does not change during RA‐induced differentiation ESCs (Figs 1A, and EV1A and EV4A). While TDG‐proficient mESCs differentiated into a homogeneous population of neuron‐like cells, forming dense axon grids, TDGcat cells were dramatically impaired (neuron‐like cells 78% reduced) in this process, as previously noted for TDGnull mESCs (Wheldon et al, 2014; Fig 4B). TDGsnm mESCs were equally deficient (70% reduced neuron‐like cells) in neuronal differentiation (Fig 4B); although the occasional neuron‐like cell emerged upon extended cultivation, TDGsnm cells, as TDGcat cells, mainly produced mixed populations of heterogeneous cells (Figs 4B, and EV4B and C). Hence, both catalytic activity and SUMOylation of TDG are required for efficient neuronal differentiation. Loss of TDG activity was reported to cause aberrations in DNA methylation (hyper‐ and hypomethylation) and associated changes in chromatin states and gene expression (Cortazar et al, 2011; Hassan et al, 2017), a phenotype which seems to be most pronounced in differentiating ESCs (Valinluck & Sowers, 2007). To test whether the defect in differentiation is caused by an inability of TDGcat and TDGsnm mESCs to establish lineage‐specific gene expression, we measured the transcriptional responses of key developmental genes during mESC differentiation to neural progenitor cells (NPC; Fig 4C). Downregulation of pluripotency genes (Nanog, Oct4, Rex1) was similarly efficient in TDGwt and TDGsnm but significantly less pronounced in TDGcat mESCs, indicating that the silencing of pluripotency genes requires TDG catalytic activity but not TDG SUMOylation (Fig 4C). The inefficient silencing of pluripotency genes in TDGcat ESCs can be explained by the accumulation of 5fC and 5caC in respective regulatory regions (Shen et al, 2013). This, in turn, will cause continuous replicative DNA demethylation and, thereby, interfere with programmed de novo DNA methylation and stable silencing of pluripotency genes in cell differentiation. On the other side, activation of the neurogenesis/ectoderm markers Pax6, Tubb3 and NeuroD1 was significantly less efficient in TDGsnm (Pax6: 30‐fold, Tubb3: 10‐fold, NeuroD1: 20‐fold difference) and TDGcat (Pax6: 30‐fold, Tubb3: 16‐fold, NeuroD1: 20‐fold difference) when compared to TDGwt NPCs (Fig 4C). Regulation of mesoderm (Brachyury, Pdgfra) and endoderm (Gata6, Gata4) marker genes was variably affected in TDGcat and TDGsnm cells, but differences were below statistical significance (Fig 4C). The defect of TDG‐deficient (TDGcat) mESCs in upregulating ectodermal genes can again be explained by the disturbance of DNA methylation dynamics, which at some genomic loci results in a hypermethylated state (Cortazar et al, 2007; Cortellino et al, 2011; Hassan et al, 2017). The observation that TDGsnm affects developmental up‐ but not downregulation of genes indicates that coordinated actions downstream of 5fC or 5caC excision, i.e., DNA strand‐break formation and repair, are required for gene activation but not for silencing, where 5fC/5caC excision appears sufficient. These results suggest that the catalytic activity as well as SUMOylation of TDG is required to establish proper lineage‐specific gene expression.

Figure 4. SUMOylation of TDG is required for neural differentiation.

1. Neural differentiation protocol.
2. Bright‐field captures of TDGwt, TDGcat and TDGsnm mESCs at indicated stages of differentiation. DAPI, Alexa Fluor® 680 phalloidin actin staining at the bottom. Scale bars 25, 50, 500 μM as indicated. Quantitation of neural cells with neurites of total differentiating TDGwt, TDGcat and TDGsnm mESCs. n ≥ 5. Means with standard deviation (SD). **P < 0.01. Nonparametric, Mann–Whitney test.
3. Relative changes of lineage marker expression between NPCs and undifferentiated mESCs (ESCM LIF, 0 h), quantified with GAPDH as normalizer. n ≥ 3. Means with standard deviation (SD)). *P < 0.05; **P < 0.01 by nonparametric Mann–Whitney test.
4. LC‐MS/MS analysis of global cytosine‐modification levels in DNA from ground‐state mESCs (2i) and EBs. Means with standard deviation (SD) of 5mC, 5hmC, 5fC and 5caC per 10⁶ bases). n ≥ 3. *P ≤ 0.05, **P < 0.01, ***P < 0.0001 by one‐way ANOVA, Dunnett's multiple comparisons test.
5. Alkaline comet assay with pluripotent (0 h) and differentiating (24 h 5 μM RA) mESC as indicated. Olive Tail Moments and % DNA in tail. n ≥ 3. Medians with standard deviation (SD). *P ≤ 0.05; **P ≤ 0.01, ***P ≤ 0.0001 by one‐way ANOVA, Sidak's multiple comparisons test.
6. Cell proliferation assay. Relative viability of talazoparib treated to untreated cells. n = 3. Means with standard deviation (SD), P ≤ 0.05, **P < 0.01, by one‐way ANOVA, Dunnett's multiple comparisons test.
7. ChIP–qPCR analysis of TET1, TDG and XRCC1 at gene promoters in mESCs (24 h 5 μM RA). Relative enrichments to an unbound control region (Chr2neg). n ≥ 3. Means with standard deviation (SD). *P ≤ 0.05; **P < 0.001; ***P ≤ 0.0001, by two‐tailed Mann–Whitney test.

Figure EV4. SUMOylation of TDG is required for neural differentiation.

1. Immunoblot of TDG and β‐actin before (2i, 0 h) and during differentiation of TDGsnm and TDGcat mESCs (5 μM all‐trans retinoic acid (RA), 24 h, 48 h).
2. Bright‐field microscopy captures of differentiating TDGwt, TDGcat and TDGsnm mESCs. Upper row: mESCs grown in ESCM supplemented with LIF (ESCM LIF). Lower part: Representative pictures of neurite growth after 4 days in neuronal differentiation medium (4 days B27). TDGwt cells formed neurites. TDGcat and TDGsnm mESC formed mixed populations of various cell types.
3. Fluorescence microscopy of TDGwt, TDGcat and TDGsnm neurite growth after 11 days in neuronal differentiation medium (11 days B27). DAPI (blue) and Alexa Fluor® 680 phalloidin actin (red) staining was used to visualize cell nuclei and axons. TDGwt formed dense axon grids. TDGcat and TDGsnm formed various cell types. Scale bar: 50 μm.
4. Comet assay. Shown are representative pictures of SybrGreen stained nuclei from TDGwt, TDGcat, TDGsnm grown in ESCM (0 h) and ESCM supplemented with 5 μM RA for 24 h (24‐h RA). TDGsnm showed more DNA fragmentation compared to TDGwt and TDGcat in ESCM supplemented with LIF (0 h). DNA fragmentation increased in TDGwt and TDGsnm upon RA‐induced differentiation (24‐h RA). Scale bar: 25 μm.
5. ChIP of TET1, TDG and XRCC1 from mESCs (ESCM plus 24‐h RA), quantified by qPCR for enrichment at the transcription start sites of genes indicated and at a chromosome 2 negative control (Chr2neg) region not bound by TET1 or TDG. Shown are means with SD of ≥ 4 biological replicates. *P ≤ 0.05, **P < 0.001, ***P ≤ 0.0001 by nonparametric Mann–Whitney test, two‐tailed.

To address the contribution of SUMOylation to the processing of DNA demethylation intermediates by TDG, we quantified global 5mC, 5hmC, 5fC, and 5caC levels by liquid chromatography–tandem mass spectrometry (LC‐MS/MS) in pluripotent mESCs as well as in EBs undergoing RA‐induced differentiation. 5caC levels were significantly (> 3‐fold) increased in mESCs expressing catalytic inactive TDG, whereas mESCs with SUMO‐deficient TDG had wild‐type amounts of the modification (Fig 4D). Likewise, RA‐exposed TDGcat EBs showed increased 5fC (> 4‐fold) and 5caC (> 6‐fold) levels, but EBs with SUMO‐deficient TDG were not different from wild‐type (Fig 4D). These results support a role for TDG SUMOylation downstream of 5fC and 5caC excision.

As suggested by biochemical evidence, failure of TDG SUMOylation may compromise the handover of the AP‐site from TDG to the downstream‐acting BER machinery, thereby causing an accumulation of fragile BER intermediates such as DNA single‐strand breaks (SSBs; Hardeland et al, 2002; Steinacher & Schar, 2005). Quantitation of SSBs by alkaline single‐cell electrophoresis (comet assay, Fig EV4D) revealed that RA‐induced mESC differentiation for 24 h [ESC medium (ESCM) + 5 μM RA] is associated with a significant increase in DNA breakage in TDGwt but not in TDGcat and TDGsnm cells (Fig 4E). Cell differentiation thus promotes TDG‐dependent DNA base excision, generating repair intermediates including AP‐sites and SSBs. TDGsnm mESCs produced significantly higher comet measures than TDGwt and TDGcat cells in all conditions analysed (Fig 4E), indicating a specific requirement of TDG SUMOylation for effective AP‐site repair following base excision. Together with findings that a human TDG variant (TDGG199S) with enhanced AP‐site binding also causes an accumulation of DNA SSBs (Sjolund et al, 2014), these results provide consistent genetic evidence for SUMO transactions coupling the dissociation of TDG with a coordinated downstream repair of AP‐sites, thereby limiting the accumulation of fragile BER intermediates in cells. In the context of active DNA demethylation in differentiating mESCs, TDG SUMOylation would thereby promote effective repair of AP‐sites generated by 5fC and 5caC excision.

To functionally link TDG SUMOylation with BER, we examined the impact of SUMOylation deficiency (TDGsnm) on the toxicity of PARP inhibition in differentiating mESCs, the rationale being that inhibiting PARP will impair XRCC1‐dependent repair of TDG‐generated AP‐sites (Bryant et al, 2005; Farmer et al, 2005; Horton et al, 2014). RA‐induced mESC differentiation was accompanied by a marked TDG‐dependent sensitivity against PARP inhibition; addition of talazoparib in the 24‐h time course of RA exposure caused significantly more cell death in TDGwt cells (48% survival, 10 nM) than in TDGnull cells (80% survival). TDGsnm cells responded to PARP inhibition essentially as the TDGnull cells, showing significantly higher survival than TDGwt cells (75% survival; Fig 4F). Notably, TDGsnm excises 5fC and 5caC as efficiently as TDGwt (Fig 4D) and is expressed at equal levels in mESCs (Fig EV1A), suggesting that TDG‐dependent AP‐site generation is similar in TDGsnm and TDGwt mESCs. Together, persistent AP‐site binding and chromatin accumulation of TDGsnm (Fig 1C), and increased occurrence of alkaline‐sensitive DNA damage (Fig 4E) and PARP inhibitor resistance (Fig 4F) in TDGsnm ESCs indicate that the inability to SUMOylate TDG channels TDG‐generated AP‐sites into PARP‐independent repair. We therefore conclude that SUMOylation of TDG determines downstream processing of AP‐sites.

XRCC1 is implicated in SSB repair following TDG catalysed 5fC/5caC excision in TET‐dependent DNA demethylation (Hajkova et al, 2010; Cortazar et al, 2011; Weber et al, 2016). To provide evidence for co‐operation of these demethylation factors in living cells, we examined by chromatin immunoprecipitation whether TET1, TDG and XRCC1 can be detected at the transcription start sites of the Pax6, Nestin, NeuroD1 and Tubb3 genes. All three proteins showed a specific but differential enrichment at these loci relative to a control locus where TDG does not bind (Figs 4G and EV4E); TET1 was significantly enriched at all promoters, TDG at Pax6, Nestin and NeuroD1 and XRCC1 at Nestin and NeuroD1 promoters. These results support a co‐operation TET1, TDG and XRCC1 in the developmental activation of these genes. As in differentiating but otherwise unperturbed mESC the main source of TDG substrates is TET catalysed 5mC oxidation (> 6 × 10⁴ lesions per genome), the results also establish a critical role of TDG SUMOylation in this context.

XRCC1 deficiency phenocopies the TDGsnm defect in neural differentiation

XRCC1 is essential for mouse embryogenesis (Tebbs et al, 1999), its haploinsufficiency correlates with defects in endoderm and mesoderm lineage formation (McNeill et al, 2011), and it has been functionally implicated in neurogenesis (Lee et al, 2009). Genetic inactivation of TDG but no other DNA glycosylase generates similar developmental defects, suggesting that XRCC1 and TET‐TDG activity co‐operate in cell programming‐associated active DNA demethylation. While this seems obvious from current models of BER, it has never been shown. Examining the capacity of XRCC1‐deficient mESC clones to commit to the neuronal lineage by in vitro differentiation (Bibel et al, 2007; Cortazar et al, 2011; Fig 4A), we found that both clones were capable of forming EBs similar to wild‐type mESCs (Fig 5A). Unlike wild‐type cells, however, both clones were severely disturbed in neurite formation (80% reduced neuron‐like cells) following EB dissociation and extended growth in N2 medium (Figs 5A and EV5). Pluripotency markers (Nanog, Oct4) were as efficiently downregulated in XRCC1‐deficient mESCs as in wild‐type cells, but the induction of neuronal genes (Pax6, Nestin, Tubb3, NeuroD1; Fig 5B) was significantly reduced, as observed with mESC harbouring a catalytic‐ or SUMOylation‐deficient TDG (Fig 4C). This phenocopy of the TDG defect supports an engagement of XRCC1 in TET‐TDG‐mediated active DNA demethylation and epigenetic programming during cell‐lineage commitment.

Figure 5. XRCC1 is required for neural differentiation.

1. Representative images of mESC neuronal differentiation as in Fig 4B. Shown are XRCC1wt and XRCC1null (XRCC1null1, XRCC1null2) mESCs in 2i medium supplemented with LIF (2i LIF) and in ESCM supplemented with LIF (ESCM LIF), EBs in ESCM LIF and 5 μM RA (EB 4 days ESCM RA) and dissociated neuron‐like cells in B27 (B27). Scale bars 25, 50, 500 μM as indicated. Neurite grids were observed for XRCC1wt; heterogeneous cell populations formed with the XRCC1null (XRCC1null1, null2) clones. Quantification (below) of neural cells with neurites of total differentiating XRCC1wt, XRCC1null mESCs. n = 5. Means with standard deviation (SD). **P < 0.01. Nonparametric, Mann–Whitney test.
2. Quantification of mRNA expression of pluripotency and ectodermal/neurogenesis marker genes by RT–qPCR. n ≥ 3. Mean values and standard deviations (SD) of EBs relative to their mESC (ESCM LIF, 16 h), normalized to GAPDH. Significance, *P < 0.05, **P < 0.01, ***P ≤ 0.0001, two‐tailed Mann–Whitney test.
3. SUMO coordinates BER in active DNA demethylation. SUMOylated XRCC1/BER proteins are recruited to DNA‐bound TDG to form the BERosome (grey oval) SUMO protease (SENP2) liberates SUMO from XRCC1 for attachment to TDG by the SUMO‐conjugating enzymes SAE1/2 and UBC9. SUMOylated TDG dissociates from the DNA AP‐site and the BERosome, facilitating completion of demethylation by BER, and neuronal differentiation.

Figure EV5. XRCC1 is required for neural differentiation.

Representative bright‐field images of mESC neuronal differentiation experiment. Shown are XRCC1wt and XRCC1null (XRCC1null1, XRCC1null2) mESCs in 2i medium supplemented with LIF (2i LIF) and neuron‐like cells in B27 (B27). Neurite grids were observed for XRCC1wt, whereas heterogeneous cell populations formed with the XRCC1null (XRCC1null1, null2) clones. Scale bar: 25 μm.

Discussion

DNA transactions such as replication and repair are delicate processes that demand a high level of coordination to prevent genome instability. This holds true also for TET‐TDG‐BER‐mediated active DNA demethylation in differentiating stem cells, where thousands of 5mC oxidation events occur in a spatiotemporally clustered manner, generating substrates for potentially harmful excision repair (Kunz et al, 2009; Hajkova et al, 2010; Wossidlo et al, 2010; Weber et al, 2016). Previous structural and biochemical work on individual BER proteins has provided important clues regarding mechanisms supporting a concerted operation of the BER machinery (Wilson & Kunkel, 2000; Prasad et al, 2010). The evidence presented here introduces a novel aspect by demonstrating central roles for SUMO and XRCC1 in physically coupling base excision with AP‐site repair. Our data establish that SUMO modification of TDG is essential in cell differentiation and cell‐lineage commitment and that XRCC1 is critical in this process, nucleating the SUMO‐dependent assembly of BERosomes at sites of TDG‐bound DNA and facilitating the SUMOylation of TDG within such complexes (Fig 5C). This mode of action assures that the excision of TET generated 5fC/5caC and the dissociation of TDG is linked to the repair of the AP‐site lesions generated (Fig 5C). Such high level of coordination in BER seems particularly important in the context of active DNA demethylation of closely spaced 5mCs, where random action is expected to generate deleterious DNA breakage and potentiate genome instability.

The SUMO‐orchestrated formation and function of a BERosome for 5fC/5caC repair is a straightforward and plausible and well‐supported concept. SUMOs were shown to target a range of nuclear and chromatin‐associated proteins and to be involved in DNA damage sensing and repair (Seeler & Dejean, 2003; Uchimura et al, 2006; Stielow et al, 2008; Jackson & Durocher, 2013). It is also known and shown here that SUMO modification is inducible by oxidative stress and DNA damage and that this can effect interactions with SIM containing proteins to support DNA damage responses and facilitate DNA repair (Psakhye & Jentsch, 2012; Raman et al, 2013; Sahin et al, 2014). Conversely, SUMO modification was also reported to block protein interactions (Schimmel et al, 2014) or to mediate ubiquitylation by the SUMO‐specific ubiquitin ligase RNF4 to either stabilize target proteins (Thomas et al, 2016) or mark them for proteasomal degradation (Sun et al, 2007). All considered, SUMO modification provides a range of actions suitable not only to regulate individual proteins but also to orchestrate complex multi‐protein transaction, such as the formation, action, dissociation and turnover of DNA repairosomes. While this has never been shown in a well‐defined biological process, the work presented here describes such a scenario, showing that sequential SUMO modifications and interactions facilitate the dynamic regulation of the multistep enzymatic events underlying TDG‐BER‐mediated active DNA demethylation.

Alternative modes of regulating TDG turnover have been observed and discussed. Early biochemical experiments showed for instance that APE1, when present in a molar excess (> 20‐fold), can stimulate the turnover of AP‐site‐bound TDG (Waters et al, 1999; McLaughlin et al, 2016), presumably by a passive mechanism based on competition for AP‐site binding (Hardeland et al, 2002). While this effect of APE1 (and other AP‐site‐binding proteins) on TDG turnover is a biochemical fact, it is difficult to imagine how a mechanism relying on local protein concentration differences can provide sufficiently fine‐tuned control to couple TDG dissociation with AP‐site repair. It is noteworthy, however, that the stimulatory effect of APE1 on TDG turnover is significantly enhanced when TDG is SUMO‐modified; while unmodified intact full‐length TDG requires a > 20‐fold molar excess of APE1 for a threefold stimulation of turnover, SUMOylated TDG is efficiently stimulated at already equimolar APE1 concentrations (Hardeland et al, 2002). Hence, TDG SUMOylation and APE1 interactions may co‐operate in an active mechanism linking base excision with efficient downstream BER.

Our work provides further insight into the actual biological function of 5mC oxidation‐ and excision‐mediated DNA demethylation and describes for the first time a biological function of TDG SUMOylation in this process during mESC differentiation. mESCs expressing SUMOylation‐deficient TDG or lacking XRCC1 show defects in neural differentiation and the associated transcriptional control of developmental genes, the same as observed in mESC with a catalytic‐dead TDG. This demonstrates that malfunctions in transcriptional regulation associated with TDG deficiency are not primarily caused by the inability to excise 5fC or 5caC, hence by 5fC or 5caC directly, but by an inability to effectively activate and complete BER at such sites. To what extent the defect in repair of general DNA base damage also contributes to the differentiation failure is an important and complex question. It seems clear though that the > 95,000 5fC/5caCs generated by TET proteins in differentiating mESCs must constitute a significant portion of total BER substrates in their DNA and that these bases are processed through TDG. Considering further that TDG‐knockout cells do not show a mutator phenotype in different assays (Cortazar et al, 2011; Saito et al, 2012), or increased base lesions other than 5fC/5caC in mass spectrometry DNA analyses, we like to argue that the TDG‐XRCC1 axis of BER, while highly active in mESCs, does not contribute significantly to the repair of spontaneous base damage. Hence, we propose that the main role of TDG/XRCC1 in differentiating mESCs is the coordinated removal of 5fC/5caC and subsequent repair of the AP‐site, thereby facilitating TET‐targeted, locus‐specific BER to effect transcriptional and epigenetic responses in differentiating cells. This goes very well in line with the repeated observation of DNA strand‐break formation at sites of transcription initiation (Horton et al, 2014; Madabhushi et al, 2015).

A notable and at first sight counterintuitive observation is that mESCs expressing SUMOylation‐deficient TDG showed increased levels of DNA strand‐breaks (comet measures), yet a lower sensitivity to PARP inhibition than TDGwt cells. This can be mechanistically rationalized by the reduced rate of AP‐site dissociation of TDGsnm. Slow AP‐site dissociation of this TDG variant would affect the global dynamics of generation and repair of AP‐sites in a way that less sites are generated overall (resistance to PARP inhibition), but these are shielded by bound TDG and repaired with a delay (increased comet measures), presumably involving other pathways.

A general uncertainty remains with regard to the mutual roles of PARP and XRCC1 in coordination of BER. Unlike XRCC1, PARP is not required for the core BER process itself but it seems important to facilitate BER in a chromatin context (Strom et al, 2011). A plausible, yet hypothetical concept for the integration of two coordinative functions thus is that PARP associates with early steps in BER to set the stage for BERosome recruitment, while XRCC1 is organizing the BERosome itself and its interaction with the DNA‐bound glycosylase. The observation that differentiating mESC expressing wild‐type TDG is significantly more sensitive to PARP inhibition than TDGcat or TDGsnm expressing mESCs links PARP activation to TDG action and turnover, where, once activated by TDG‐dependent base excision, it may PARylate proteins in the vicinity of the lesion generated.

Finally, the question how the TET‐TDG demethylation system (Muller et al, 2014; Weber et al, 2016) is initially targeted to specific CpGs in the genome still remains open. It is worth noting, however, that TDG was shown to interact with transcription factors, including retinoid receptors (RARα, RARβ, RARγ) and oestrogen receptors (ERα, ERβ), all of which are themselves targets of SUMO (Um et al, 1998; Chen et al, 2003; Wu et al, 2004; Sentis et al, 2005; Picard et al, 2012; Leger et al, 2014; Liu et al, 2016) and known to modulate DNA methylation patterns (Kashyap & Gudas, 2010; Ung et al, 2014). In the light of the work presented, the concept of a SUMO modification and interaction controlled assembly, function and disassembly of dynamic protein complexes to target and coordinate epigenetic remodelling activities is appealing and warrants further investigation.

Materials and Methods

Chromatin fractionation

2 × 10⁷ cells were washed 3× with ice‐cold PBS pH 7.4 and harvested with a cell scraper and transferred into a 1.5‐ml Eppendorf tube. Cells were 2× pelleted and washed using centrifugation (425 rcf, 5 min, 4°C). Cell pellets were resuspended in 200 μl Buffer A [10 mM HEPES, pH 7.9, 10 mM KCl, 1.5 mM MgCl₂, 0.34 M sucrose, 10% glycerol, 1 mM DTT, 1× cOmplete Protease Inhibitor Cocktail (Roche)]. Triton X‐100 (Sigma) was added to a final concentration of 0.1% and cells incubated on ice for 6–8 min. Nuclei were collected by centrifugation (1,300 rcf, 5 min, 4°C), washed with Buffer A and lysed in Buffer B [3 mM EDTA, 0.2 mM EGTA, 1 mM DTT, 1× cOmplete Protease Inhibitor Cocktail (Roche)]. After centrifugation (1,700 rcf, 5 min, 4°C), supernatant (soluble nuclear proteins) and pellet (chromatin‐bound proteins) were collected. The pellet was resuspended in Buffer C [3 mM EDTA, 0.2 mM EGTA, 1 mM DTT, 1× Complete protease inhibitor cocktail (Roche), 50 mM Na‐phosphate pH 8, 0.5% NP‐40 (Calbiochem)] and incubated on ice for 1 h. After sonication (5 min: 30 s pulse, 15 s off, 4°C; Bioruptor, Diagenode) and centrifugation (20,000 rcf, 15 min, 4°C), solubilized chromatin‐associated proteins were harvested in supernatant (Chr1). The pellet with chromatin‐bound proteins (Chr2) was washed 1× with Buffer C, resuspended in SDS buffer and incubated at 10 min at 99°C. Proteins were separated on 10% SDS–PAGE and subjected to immunoblot analysis.

Protein interaction and “Far‐Western” analysis

For “Far‐Western” analysis, purified human TDG (10, 5, 1, 0.1 μg), TDGE301Q (5, 1, 0.1 μg), TDG‐SUMO (10 μg), POLβ (1 μg), APE1 (1 μg), LIGIII (1 μg), and BSA (10 μg) were pipetted on nitrocellulose membrane (Protran, Amersham) and incubated with purified human XRCC1 or XRCC1‐SUMO (20 μg) in 500 μl incubation buffer (50 mM Tris pH 8.0, 10% glycerol, 150 mM NaCl, 0.5 mg/ml BSA, 0.1% NP‐40, 1 mM DTT) at 4°C for 4 h. After removal of unbound protein, blots were incubated with rabbit polyclonal anti‐XRCC1 antibody (Sigma‐Aldrich, X0629, 1:1,000 dilution) in non‐fat dry milk TBS (100 mM Tris–HCl pH 8, 150 mM NaCl) 0.1% Tween‐20 (Sigma) and analysed by chemiluminescence detection (WesternBright ECL, Advansta) on film (FujiFilm) or PXi imaging system, Syngene. To monitor interaction of endogenous TDG with XRCC1‐SUMO protein, 200 μg of native mESC extract [50 mM Tris pH 8, 0.8% NP‐40, 150 mM NaCl, 10% glycerol, N‐ethylmaleimide (NEM) 1 mM] was diluted in five volumes of incubation buffer (50 mM Tris pH 8.0, 10% glycerol, 150 mM NaCl, 0.5 mg/ml BSA, 1 mM DTT) and incubated with 500 ng in vitro‐modified 6hisXRCC1‐SUMO or non‐modified 6hisXRCC1 and Ni‐NTA resin at 4°C for 4 h. After washing with incubation buffer, the bound proteins were separated on a 10% SDS–PAGE and subjected to immunoblot analysis using rabbit polyclonal anti‐XRCC1 antibody (Sigma‐Aldrich, X0629, 1:1,000 dilution) and rabbit polyclonal anti‐TDG antibody (laboratory stock, 1:10,000 dilution) in non‐fat dry milk TBS (100 mM Tris–HCl pH 8, 150 mM NaCl) 0.1% Tween‐20 (Sigma) and analysed by chemiluminescence detection (WesternBright ECL, Advansta) using the PXi imaging system, Syngene.

DNA binding assays

DNA binding of human TDG, TDGE310Q and the human TDG‐XRCC1‐SUMO complex was tested using a biotinylated 5caC double‐stranded DNA oligonucleotide substrate, generated as described before (Hardeland et al, 2002). 1 μM of the substrate DNA [and where indicated 10 μM of homoduplex competitor DNA (G•C DNA)] was incubated with 0.5 μM human TDG, TDGE301Q protein for 5 min at 30°C in reaction buffer (50 mM Tris–HCl pH 8.0, 10% glycerol, 80 mM NaCl, 0.5 mg/ml BSA, 4% NP‐40, 1 mM DTT), subsequently human POLβ, LIGIII, SUMO, XRCC1‐SUMO or XRCC1 in reaction buffer were added to a concentration of 0.5 μM each as indicated and incubated for 60 min at 4°C. After washing three times with 125 μl reaction buffer at 4°C, 10 μl of 2× SDS loading dye was added, the samples incubated at 99°C, released proteins separated on 10 and 20% SDS–polyacrylamide gels and transferred to nitrocellulose membranes (Protran, Amersham) by electroblotting. Blots were then incubated with rabbit polyclonal anti‐TDG 141 antibody (raised against recombinant full‐length hTDG), dilution 1:20,000, rabbit polyclonal anti‐XRCC1 antibody (Sigma‐Aldrich, X0629), dilution 1:1,000, rabbit polyclonal anti‐POLβ antibody (Acris, AM00275PU‐N), dilution 1:1,000, rabbit polyclonal anti‐SUMO antibody (Sigma‐Aldrich, S8070), dilution 1:1,000, dilution, rabbit polyclonal anti‐LIGIII antibody (Sigma, HPA006723), dilution 1:1,000. All antibodies were diluted in non‐fat dry milk TBS (100 mM Tris–HCl pH 8, 150 mM NaCl), 0.1% Tween‐20 (Sigma). Analysis was done by chemiluminescence detection (WesternBright ECL, Advansta) on film (FujiFilm) or PXi imaging system, Syngene.

SUMOylation and SUMO transfer reactions

GST‐SUMO‐1, GST‐SAE2‐SAE1, and GST‐UBC9 were produced and purified in Escherichia coli (Uchimura et al, 2004) and SUMOylation reactions performed as described previously (Steinacher & Schar, 2005; Tatham et al, 2001; section protein purification). An annealed 60‐mer homoduplex G•C‐ or G•5caC‐containing oligonucleotide and TDG and were pre‐incubated for 5 min at 30°C in reaction buffer (20 mM Tris–HCl pH 8.0, 80 mM NaCl, 7 mM MgCl₂, 0.5 mM EDTA, 2.5% glycerol, 0.5 mM DTT, 10 mM ATP). The reaction volume was adjusted with reaction buffer to obtain final oligonucleotide and protein concentrations: 10 μM 60‐mer homoduplex G•C‐ or G•5caC‐containing oligonucleotide, 2 μM TDG, 5 nM SENP2, 0.4 μM SAE1/2, 2 μM UBC9, and 2 μM/40 μM SUMO were added to the reactions for 4 h at 30°C as indicated. TransSUMOylation reactions were performed using 2 μM TDG and 2 μM XRCC1/XRCC1‐SUMO (> 50% SUMOylated) that were incubated for 5 min at 30°C with the G•5caC substrate (10 μM) in reaction buffer. Then, 5 nM SENP2, 0.4 μM SAE1/2, 2 μM UBC9, 2 μM APE1, 2 μM POLβ and 2 μM LIGIII were added to the reactions for 4 h at 30°C as indicated. The SUMOylation and transSUMOylation reactions were stopped with SDS loading buffer and incubating at 99°C for 5 min. The proteins were separated on a 10% SDS–PAGE and subjected to immunoblot analysis.

Comet assay

mESCs were grown for six passages in 2i supplemented with LIF on gelatin (0.1%, Sigma)‐coated 10‐cm dishes (Falcon). 3 × 10⁶ cells were seeded on gelatin‐coated 10‐cm dishes (Falcon) and grown in ESCM supplemented with LIF (Millipore) for 16 h (0 h time point). Glass slides (75 × 25 × 1 mm, Thermo Scientific) were precoated with 1% normal melting point agarose (Bio‐Rad) and low melting point agarose (SEAPlaque GTG, Lonza) and stored (4 h, 4°C). Cells were trypsinized (0.25% trypsin EDTA, Gibco) for 1 min, quenched with ESCM, washed in PBS and resuspended to a concentration of 3 × 10⁶ cells/ml in PBS on ice. 20 μl of the suspension was mixed with 180 μl 0.75% low melting point agarose (SEAPlaque GTG, Lonza) at 37°C, immediately spread by adding a coverslip (Menzel) on a precoated glass slide and incubated for 25 min at 4°C. The coverslip was removed and another agarose layer (120 μl, 0.75% low melting point agarose (SEAPlaque GTG, Lonza) added. After incubation for 25 min at 4°C, the coverslip was removed and the glass slides incubated in lysis buffer (Trevigen) for 45 min at 4°C in the dark. Slides were placed on a rack (Trevigen) and DNA strands denatured (200 mM NaOH, 1 mM EDTA, 4°C in the dark). DNA was separated (17 V/185 mA, 25 min). After neutralization (250 mM Tris–HCl pH 7.6, 90% MetOH, 2 × 10 min), washing (50 mM Tris–HCl pH 7.6, 80% EtOH, 5 min, room temperature (RT); 80% EtOH, 5 min RT) slides were dried overnight at RT. Slides were rehydrated (10 mM Tris–HCl pH 7.6, 2 × 15 min, RT), stained and mounted (SybrGreen, Qiagen, Vectashield, Vector Laboratories). Stained slides were screened and captured using a motorized Axio Imager Z2 (Carl Zeiss) supplied with the Metafer4 Software (MetaSystems, Altlussheim, Germany). The slide scanning platform “Metafer” scans automatically and captures the images. Slides of three independent experiments were analysed. 1,000 nuclei were analysed for each condition (0 and 24 h RA). The focus was automatically determined, and the exposure time was automatically adjusted to avoid saturation. Detected cells were rejected when doublet cells or strongly damaged cells were detected. Accepted comets were automatically analysed (Metafer4 software) and head and tail of the comet determined based on the intensity levels. The median of % DNA in tail and the median of Olive Tail Moments of 1,000 nuclei per slide were calculated (Metafer4 software). Medians of % DNA in tail and Olive Tail Moments of three independent experiments were analysed by nonparametric Mann–Whitney (Wilcoxon signed rank) tests.

Cell differentiation assay

mESCs were grown on feeders at 37°C for two passages in ESCM in a humidified atmosphere containing 5% CO₂. mESCs were grown without feeders for five passages in 2i medium containing LIF and 16 h in ESCM containing LIF (0 h). Embryoid body formation was achieved by seeding 4 × 10⁶ Tdg ^−/− wt, cat or snm mESCs onto non‐adherent bacterial dishes (Greiner Bio‐One) in ESCM without LIF and incubation at 37°C with a daily medium exchange. After 4 days, 5 μM all‐trans retinoic acid (RA) was added and cells were further incubated for 4 days with a daily medium exchange. After washing of the embryoid bodies twice with 1× PBS, they were dissociated with freshly prepared trypsin solution (0.05% TPCK‐treated trypsin in 0.05% EDTA/PBS) at 37°C for 1.5 min. Dissociated embryoid bodies were resuspended in 10 ml ESCM centrifuged at 500 × rcf for 2 min at room temperature. Cells were resuspended in N2 medium (DMEM‐F12 nutrient mixture 1:1, 1 × N2 supplement) and filtered through a 40‐mm nylon cell strainer (BD) and plated onto poly‐l‐lysine and laminin‐coated dishes at a density of 2 × 10⁶ cells per 60‐mm dish. N2 medium was exchanged after 4 and 24 h after plating. Then, cells were further differentiated in B27 medium.

Quantitative RT–PCR analyses

RNA was extracted by TRIzol RNA isolation reagents (life technologies) or RNeasy Kit (Qiagen) and reverse transcribed with the RevertAid Kit (Thermo Scientific) according to the manufacturer's protocol. qPCR was performed using Power SYBR Green Master Mix (Applied Biosystems) and a Rotor‐Gene 3000 thermocycler with target‐specific primers (Appendix Table S2). PCR products were normalized to a GAPDH control.

Protein extraction

Native protein extracts were prepared by incubation of 10⁸ cells in 1 ml single detergent lysis buffer (50 mM Na‐phosphate pH 8.0, 125 mM NaCl, 1% NP‐40, 0.5 mM EDTA, 1 mM DTT, 1 mM PMSF, 1× Complete™ protease inhibitors) for 30 min on ice. After centrifugation (20,000 × rcf, 15 min, 4°C), the soluble proteins were frozen in liquid nitrogen. For deSUMOylation experiments, 1 μg of SENP2 was added per 20 μg extract for 45 min at 4°C. Protein concentrations of all native extracts were measured by the Bradford method (Bio‐Rad). Total denaturing cell extracts were prepared in the presence of 5 mM NEM. 10⁷ cells grown in 9‐cm tissue culture dishes were washed once with 1× phosphate‐buffered saline (PBS), lysed directly in 200 μl of 2× SDS lysis buffer (120 mM Tris–HCl pH 6.8, 4% SDS, 20% glycerol, 200 mM DTT, 0.01% bromophenol blue) and scraped off the dishes. The proteins were incubated for 5 min at 96°C for denaturing.

Chromatin immunoprecipitation

mESCs were incubated in freshly prepared crosslinking solution (PBS pH 7.4, 1% formaldehyde) to crosslink (10 min, RT) protein‐bound DNA. The reaction was quenched (glycine final concentration 125 mM), cells 2× washed (PBS, ice cold) and collected using a cell scraper. After centrifugation (600 × rcf, 4°C, 5 min), nuclei were isolated by incubation in 200 μl of ice‐cold ChIP Buffer I (10 mM HEPES pH 6.5, 10 mM EDTA, 0.5 mM EGTA, 0.25% Triton X‐100) for 5 min on ice and 2 × 5 min on ice in 200 μl cold ChIP buffer II (10 mM HEPES pH 6.5, 1 mM EDTA, 0.5 mM EGTA, 200 mM NaCl). After lysis in 400 μl ChIP buffer III (50 mM Tris–HCl pH 8.0, 1 mM EDTA, 0.5% Triton X‐100, 1% SDS, 1 mM PMSF, 10 min on ice) and sonication (15 min 15 s on, 30 s off, Bioruptor, Diagenode), random chromatin fragments ranging from 300 to 1,000 base pairs were produced. After centrifugation (14,000 × rcf, 4°C, 10 min), the concentration of chromatin was measured (Nanodrop, Thermo Scientific, 260 nm). For ChIP of TDG, TET1 and XRCC1, 150 μg of chromatin was diluted (1:10) in ChIP dilution buffer I (50 mM Tris–HCl pH 8.0, 1 mM EDTA, 150 mM NaCl, 0.1% Triton X‐100, 1 mM PMSF). For input analysis, 1% of the chromatin was removed for analysis and then the chromatin precleared (4°C, 1 h) with 40 μl of a 50% slurry of magnetic Protein G beads (Invitrogen), blocked with 1 mg/ml BSA and 1 mg/ml tRNA. 2 μg TET1 (Millipore, catalogue number: 09‐872), 2 μg TDG (Cortazar et al, 2011) or 1 μg XRCC1 (Sigma, catalogue number: 0629) antibody was added to the precleared chromatin and incubated (overnight, 4°C) under slow rotation. Antibody‐chromatin complexes were precipitated with 40 μl of a 50% slurry of blocked Protein G beads and further incubated (4°C, 2 h). The beads were washed 1× with 500 μl ChIP wash buffer I (20 mM Tris–HCl pH 8.0, 2 mM EDTA, 150 mM NaCl, 0.1% SDS, 1% Triton X‐100), 2× with 500 μl ChIP wash buffer II (20 mM Tris–HCl pH 8.0, 2 mM EDTA, 500 mM NaCl, 0.1% SDS, 1% Triton X‐100) and 2× with 500 μl TE buffer (10 mM Tris–HCl pH 8.0, 1 mM EDTA). The bound complexes were eluted (2× with 150 μl elution buffer: 1% SDS, 0.1 M NaHCO₃, 65°C, 10 min). Crosslinks of input and eluates were revered (200 mM NaCl, 65°C, 4 h) and digested with proteinase K (ProtK: 50 μg/ml; 10 mM EDTA, 45°C, 1 h), DNA purified by phenol/chloroform extraction and Na‐acetate/ethanol precipitation and re‐suspended (10 mM Tris–HCl pH 8.0). qPCR analysis was performed using Quantitect SYBR Green (Qiagen) with a Rotor‐Gene 3000 thermocycler (Qiagen) and target‐specific primers (Appendix Table S3).

BER reconstitution

Reconstitution SUMO‐conjugation/de‐conjugation BER demethylation reactions were carried out in 20 μl reaction volumes in reaction buffer (50 mM Tris–HCl pH 8.9, 1 mM DTT, 0.1 mg/ml BSA, 50 mM NaCl, 1 mM ATP, 7 mM MgCl₂, 200 μM dCTP) with SAE1/2 (4 pmol) and UBC9 (4 pmol), SUMO protease SENP2_364–489 (5 fmol), XRCC1‐SUMO, or XRCC1, XRCC1snm, free SUMO (8 pmol), LigIII (8 pmol) TDG or TDG K330R (1 pmol), APE (200 fmol), Polβ (16 fmol), substrate (2 pmol; upper strand 5′‐TAGACATTGCCCTCGACGACCCGCCGCCGCGCXGGCCACCCGCACCTAGACGAATTCCG‐3′ where X = C. lower strand 5′‐CGGAATTCGTCTAGGTGCGGGTGGCXGGCGCGGCGGCGGGTCGTCGAGGGCAATGTCTA‐3′ where X = C, 5caC) as indicated. Reactions were incubated at 37°C for 2 h, and the DNA ethanol precipitated O/N at −20°C. DNA was resuspended and incubated with HpaII (1 U, NEB) in 1× CutSmart buffer (NEB) at 37°C for 1 h. DNA was ethanol precipitated O/N at −20°C, resuspended on glycerol loading buffer (0.5× TBE, 50% glycerol) and separated on a 8% native PAGE, and labelled DNA was detected using the blue fluorescence mode of the Typhoon 9400 (GE Healthcare) and analysed quantitatively by ImageQuant TL software (v7.0, GE Healthcare).

An Appendix Supplementary Methods section is available in the Appendix.

Author contributions

RS and PS conceived the project, organized funding, designed the experimental approach and analysed data. PB generated XRCC1‐knockout cell lines, performed clonogenic and differentiation assays and immunoblot analyses of TDG and TDG‐SUMO in XRCC1null mESCs. ZB performed cell survival assay, generated GFP‐TDGwt and GFP‐TDGpip mESCs and performed immunoblot analysis of TDG and XRCC1 in GFP‐TDGwt and GFP‐TDGpip ESCs. AK and GS performed the LC‐MS/MS analyses. RS and PS wrote the manuscript. All authors contributed to the manuscript writing.

Conflict of interest

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Source Data for Expanded View

Review Process File

Source Data for Figure 1

Source Data for Figure 2

Source Data for Figure 3

The EMBO Journal (2019) 38: e99242