MBD4 loss results in global reactivation of promoters and retroelements with low methylated CpG density

Primary WT or KO MEFs for Mbd4 were isolated from Mbd4^+/+ and Mbd4^−/− mice respectively, as previously described [31].

After isolation of total cellular RNA from subconfluent MEFs (two independent samples were purified from both WT or KO MEFs for Mbd4), libraries of template molecules suitable for strand specific high throughput DNA sequencing were created using “TruSeq Stranded Total RNA with Ribo-Zero Gold Prep Kit” (# RS-122–2301, Illumina). Briefly, starting with 300 ng of total RNA, the cytoplasmic and mitochondrial ribosomal RNA (rRNA) were removed using biotinylated, target-specific oligos combined with Ribo-Zero rRNA removal beads. Following purification, the RNA was fragmented into small pieces using divalent cations under elevated temperature. The cleaved RNA fragments were copied into first strand cDNA using reverse transcriptase and random primers, followed by second strand cDNA synthesis using DNA Polymerase I and RNase H. The double stranded cDNA fragments were blunted using T4 DNA polymerase, Klenow DNA polymerase and T4 PNK. A single ‘A’ nucleotide was added to the 3’ ends of the blunt DNA fragments using a Klenow fragment (3' to 5'exo minus) enzyme. The cDNA fragments were ligated to double stranded adapters using T4 DNA Ligase. The ligated products were enriched by PCR amplification (30 s at 98 °C; [10 s at 98 °C, 30 s at 60 °C, 30 s at 72 °C] × 12 cycles; 5 min at 72 °C). Then surplus PCR primers were removed by purification using AMPure XP beads (Agencourt Biosciences Corporation). Final cDNA libraries were checked for quality and quantified using 2100 Bioanalyzer (Agilent). The libraries were loaded in the flow cell at a concentration of 7 pM, and clusters were generated in the Cbot and sequenced in the Illumina Hiseq 2500 as single-end 50 base reads following Illumina's instructions. Image analysis and base calling were performed using RTA 1.17.20 and CASAVA 1.8.2. Reads were mapped onto the mm9 assembly of the mouse genome by using Tophat [42] and the bowtie aligner [43]. Quantification of gene expression was performed using HTSeq (http://​www-huber.​embl.​de/​users/​anders/​HTSeq) and gene annotations from Ensembl release 67. Read counts have been normalized across WT and KO libraries with the statistical method proposed by Anders and Huber [44] and implemented in the DESeq Bioconductor library. Resulting p-values were adjusted for multiple testing by using the Benjamini and Hochberg method [45].

Genomic DNA was isolated from subconfluent primary MEFs as previsously described [46]. Digested DNA (500 ng) was converted with EZ DNA Methylation-Gold Kit (Zymo Research Corporation). Primer design was accomplished using Methprimer. Bisulfite sequencing primers (5′-TTTTTTTATGAATAAGTAATTTAATAATAT-3′ and 5′-AATTTCCTAAAATCCCAAATCTCTC-3′ for Zic5, 5’- GTTGGAGGTGATTAGGGTTTAAAA-3’ and 5’- TCTAATCAAAAAAACTCCCTAAACC for Bzrap1) were used to amplify the corresponding promoters. PCR included an initial incubation at 95 °C for 10 min, followed by 40 cycles of 95 °C for 30 s, 52 °C for 30 s, and 72 °C for 60 s, followed by one cycle of 72 °C for 10 min. The PCR products were cloned into the pCR2.1-TOPO vector using the TOPO TA cloning kit (Invitrogen) for sequencing. A total of 10 clones from each sample were sequenced at the GATC Biotech company, and the methylation status for each CpG site was determined by assessing the presence of T (unmethylated) versus C (methylated) at each CpG site.

For RRBS, bisulfite-converted genomic DNA libraries were prepared according to the previously described methods [47]. Briefly, genomic DNA was digested with MspI (New England Biolabs), followed by end-repair and addition of 3′ A overhangs. Methylated adaptors (Illumina) with a 3′ T overhang were ligated to the A tailed DNA fragments. For reduced representation, 40 to 220 bp (pre-adaptor-ligation size) fragments were excised from 2% TAE agarose gels and bisulfite-converted with EZ DNA methylation Gold kit (Zymo Research). Bisulfite converted libraries were amplified by PCR and sequenced on an Illumina HiSeq 2000 sequencer with a single-ended, 49 bp run (Beijing Genomics Institute). FASTQ sequence files containing sequenced reads were obtained for both samples (WT or KO MEFs for Mbd4).

After removal of the adaptor sequences, the 49 bp reads from each sample were aligned to the reference genome (mm9) as well as the size-selected MspI fragments generated by our in-silico simulation. Because of the strand specificity of DNA methylation, two rounds of alignments were carried out, i.e. the bisulfite converted reads were aligned to the genome sequences termed the “T genome” with each cytosine converted to thymine, and simultaneously the reads were also aligned to the genome sequences termed the “A genome” with each guanine converted to adenosine. The alignments were carried out with BAMAP aligner allowing up to two mismatches for successful mapping. Summary of the data quantity after each step of filtration is shown in Supplementary Fig. 1A.

The mouse CGI database was retrieved from the UCSC Genome Bioinformatics site using Table Browser program (genome: mouse; assembly: NCBI37/mm9; group: Expression and Regulation; track CpG islands). The mouse promoter database was produced using promoter sequences consisting of the—2 kb to 0 genomic interval relative putative TSSs, using the start of the RefSeq. The percentage of CG methylation was determined by obtaining the number of methylated CGs divided by the total number of CpG dinucleotides per region. The calculation of CG methylation percentages was limited to CGIs and promoters with a minimum 80% coverage of CpG sites. For each CpG site, the methylation level of 5mCG was calculated as the ratio of methylated reads to the number of total reads covered that CpG. We only considered regions with an average coverage of at least 5 reads per methylated CpG to calculate the average methylation level of 5mCpG in CGIs and promoters.

DNA immunoprecipitation assays were done as previously described [48]. Briefly, 10 µg of DNA was used as input, and 2 µl of 5mC antibody (Active Motif, 39,649) or 4 µl of 5hmC antibody (Active Motif, 39,791) was used to immunoprecipitate modified DNA. DNA and antibodies were incubated at 4 °C overnight in a final volume of 500 µl of DIP buffer (10 mM sodium phosphate pH 7.0, 140 mM NaCl, 0.05% Triton X-100). The bound material was recovered after incubation with 30 µl of blocked protein G Dynabeads (beads washed three times with 1 ml of DIP buffer and incubated for 4 h minimum with BSA 1 mg ml⁻¹ and yeast tRNA 0.5 mg ml⁻¹). The beads were washed three times with 1 ml of DIP buffer, then treated overnight with RNase at 65 °C in presence of 300 mM NaCl and then treated 4 h with proteinase K at 55 °C. Immunoprecipitated DNA was purified by phenol–chloroform extraction followed by ethanol precipitation. Four independent DNA immunoprecipitations were pooled for each condition before sequencing analysis. Sequencing was performed on an Illumina Hiseq 2500 as single-end 50 base reads following Illumina’s instructions. Image analysis and base calling were performed using RTA 1.17.20 and CASAVA 1.8.2. Reads were mapped to the mouse genome (mm9) using bowtie [43] with the following arguments “-m 1 –strata –best -y -S -l 40 -p 2”. We extract the tag density in a 2 kb window surrounding the gene bodies running seqMINER (http://​bips.​u-strasbg.​fr/​seqminer/​), using datasets normalized to the total number of uniquely mapped reads. Tag densities were calculated in 100 bp sliding windows spanning ± 2 kb of gene bodies. For average gene profiles, genes were divided in 40 bins of length relative to the gene length, and the adjacent 2 kb sequences in 10 bins.

Repeat analyses of RNA-seq and DIP-seq datasets were performed as follows. Reads were aligned to repetitive elements in two passes. In the first pass, reads were aligned to the non-masked mouse reference genome (NCBI37/mm9) using BWA [49] v0.6.2. Positions of the reads uniquely mapped to the mouse genome were cross-compared with the positions of the repeats extracted from UCSC (RMSK table in UCSC database for mouse genome mm9) and reads overlapping a repeat sequence were annotated with the repeat family. In the second pass, reads not mapped or multi-mapped to the mouse genome in the previous pass were aligned to RepBase [50] v18.07 repeat sequences for rodent. Reads mapped to a unique repeat family were annotated with their corresponding family name. Finally, we summed up the read counts per repeat family of the two annotation steps. Data were normalized based upon library size. Difference of repeat read counts between samples was expressed as the log₂-ratio KO/WT. The statistical significance of the difference between samples was assessed using the Bioconductor package DESeq. Processed datasets presented in this paper were restricted to retrotransposon families (SINEs, LINEs and LTRs) with more than 100 mapped reads per DIP/RNA sample to avoid over- or underestimating fold enrichments due to low sequence representation.

Full-length human cDNA clones of MBD4 (IMAGE 3534047), MLH1 (IMAGE 3451538) and PMS2 (IMAGE 7939766) were purchased from Source BioScience. The coding sequence of MBD4 was mutated using megaprimer PCR to produce MBD4 R97G and MBD4 D554A mutant proteins.

MBD4 full-length cDNA was subcloned into the XhoI-NotI sites of the pOZ-N retroviral vector to produce MBD4 protein fused with N-terminal Flag- and HA-epitope tags (e-MBD4). e-MBD4 was stably expressed in HeLa cells or in MEFs by retroviral transduction [51]. e-MBD4 nuclear complex (e-MBD4.com) was purified from these cells by double immunoaffinity purification as previously described [52]. MBD4 concentration in e-MBD4.com was estimated by polyacrylamide gels silver-staining using His-tagged MBD4 protein as a standard. Identification of proteins was carried out by Taplin Biological Mass Spectrometry Facility (Harvard Medical School, Boston, MA).

For glycerol density gradient, samples were loaded onto a 4.5 mL glycerol gradient (15%-35%) and spun at 45,000 rpm in a Beckman SW60 rotor for 16 h. Fractions were collected from the bottom of the tube.

Antibodies employed were as follows: monoclonal antibodies anti-Flag M2-Peroxidase (Sigma), anti-TIP49B (612,482, BD Transduction), anti-MLH1 (NA28, Calbiochem), anti-PMS2 (556,415, BD PharMingen), anti-TIP49A (ab51500, Abcam); polyclonal antibody anti-MBD4 (A-1009, Epigentek).

The His-tagged protein was cloned in pET28b vector and expressed in the BL21-CodonPlus-RIL-pLysS (Stratagene) strain. An 800 mL culture was grown in LB medium at 37 °C until D₆₀₀ of 0.5 was reached before induction with 100 µM IPTG for 2 h at 25 °C. Cells were lysed in 20 mL of buffer containing 10 mM Tris–HCl pH 7.65, 500 mM NaCl, 10% glycerol, 0.01% NP40, 10 mM Imidazole, 0.2 mM PMSF and protease inhibitor cocktail tablets (Roche) on ice in the presence of lysozyme at 1 mg/mL and sonicated on ice for 3 × 1 min. His-tagged proteins were immunoprecipitated from clarified supernatant with Ni–NTA-agarose (Qiagen), washed with 50 mM Imidazole and eluted with 300 mM Imidazole using a buffer containing 10 mM Tris–HCl pH 7.65, 150 mM NaCl, 10% glycerol, 0.01% NP40. The eluate fraction was diluted two times with sodium phosphate buffer (50 mM sodium phosphate pH 7.0, 1 mM DTT, 1 mM EDTA), incubated with SP sepharose fast flow beads (GE Healthcare), extensively washed with sodium phosphate buffer containing 300 mM NaCl and eluated with sodium phosphate buffer containing 500 mM NaCl. The eluate fraction was desalted with PD-10 Sephadex G-25 columns (GE Healthcare) equilibrated with TGEN buffer containing 150 mM NaCl. His-tagged MBD4 purified proteins were quantified using the Bradford assay with BSA as a standard.

Flag-tagged MBD4, His-tagged MLH1 and HA-tagged PMS2 proteins were cloned in pFastBac vector (Invitrogen). The cloned vectors were transformed into bacterial DH10Bac competent cells for making recombinant bacmid. The recombinant bacmid was then extracted and transfected into Sf9 cells by Cellfectin II Reagent (Invitrogen). After viral amplification, SF9 cells were infected (10⁶ cells per mL) with baculoviruses expressing either Flag-MBD4 alone or in combination with His-MLH1 and/or HA-PMS2 for 2 days at 27 °C. Cells were harvested and resuspended in 25 mL lysis buffer containing 10 mM Tris–HCl pH 7.65, 500 mM NaCl, 10% glycerol, 0.01% NP40, 20 mM Imidazole, 0.2 mM PMSF and protease inhibitor cocktail tablets (Roche). The lysate was dounced 30 times, sonicated, and centrifuged for 10 min at 12,000 rpm. The clarified supernatant was incubated at 4 °C with anti-FLAG M2 affinity agarose resine (SIGMA), washed 3 times with lysis buffer and 3 times with wash buffer containing 10 mM Tris–HCl pH 7.65, 500 mM NaCl, 10% glycerol, 0.01% NP40 and 40 mM Imidazole. The immunoprecipated proteins were eluted with Flag peptide (0.5 mg/mL). The eluted fraction was diluted 3 times with 100 mM sodium phosphate pH 7.0 and incubated with SP sepharose fast flow beads (GE Healthcare). Beads were washed 3 times with wash buffer containing 50 mM sodium phosphate pH 7.0, 100 mM NaCl, 10% glycerol, 0.01% NP40, 3 times with wash buffer 2 containing 50 mM sodium phosphate pH 7.0, 100 mM NaCl and eluted with 500 mM NaCl. The eluate fraction was concentrated and loaded onto a 4.5 mL glycerol gradient (10%-30%) and spun at 34,000 rpm in a Beckman SW60 rotor for 18 h. Fractions were collected from the bottom of the tube and protein containing fractions were pooled and dialyzed in buffer containing 50 mM Tris–HCl pH 7.5, 500 mM NaCl and 10% glycerol.

The DNA substrates for enzymatic activity assays were prepared by annealing equimolar amounts of the corresponding synthetic oligonucleotides in a buffer containing 10 mM Tris–HCl (pH 7.5), 1 mM EDTA and 100 mM NaCl. DNA substrates were 5’-end labelled on the top or the bottom strand as indicated with [γ-³²P]ATP and T4 polynucleotide kinase.

Underlined characters indicate the position of the mismatch. 5mC, 5hmC, 5fC, 5caC and Ø indicate 5-methylcytosine, 5-hydroxymethylcytosine, 5-formylcytosine, 5-carboxylcytosine and abasic site respectively.

Bot5mCG: 5’ GCA TCG TAC GGA ATC GCT TCT AGC CGG ACA TTA GCG AT5mC GAT CGA TCA GGC TCG TAG GTA CTC GAC GGC AAT CGT TAG 3’.

Bot5hmCG: 5’ GCA TCG TAC GGA ATC GCT TCT AGC CGG ACA TTA GCG AT5hmC GAT CGA TCA GGC TCG TAG GTA CTC GAC GGC AAT CGT TAG 3’.

Bot5fCG: 5’ GCA TCG TAC GGA ATC GCT TCT AGC CGG ACA TTA GCG AT5fC GAT CGA TCA GGC TCG TAG GTA CTC GAC GGC AAT CGT TAG 3’.

Bot5caCG: 5’ GCA TCG TAC GGA ATC GCT TCT AGC CGG ACA TTA GCG AT5caC GAT CGA TCA GGC TCG TAG GTA CTC GAC GGC AAT CGT TAG 3’.

BotabG: 5’ GCA TCG TAC GGA ATC GCT TCT AGC CGG ACA TTA GCG ATØ GAT CGA TCA GGC TCG TAG GTA CTC GAC GGC AAT CGT TAG 3’.

CG/GC homoduplex substrate: TopCG + BotCG; CG/GT mismatch substrate: TopCG + BotTG; CG/GT mismatch hemi-methylated substrate: TopCG + BotTG5mC4 or TopCG5mC4 + BotTG; CG/GT mismatch full-methylated substrate: TopCG5mC4 + BotTG5mC4; CG/GU mismatch substrate: TopCG + BotUG; CC/GT mismatch substrate: TopCC + BotTG; CT/GT mismatch substrate: TopCT + BotTG; CA/GC mismatch substrate: TopCA + BotCG; CC/GC mismatch substrate: TopCC + BotCG; CA/GG mismatch substrate: TopCA + BotGG; CA/GA mismatch substrate: TopCA + BotAG; CG/GG mismatch substrate: TopCG + BotGG; CG/G5mC substrate: TopCG + Bot5mCG; 5mCG/G5mC homoduplex substrate: Top5mCG + Bot5mCG; CG/G5hmC homoduplex substrate: TopCG + Bot5hmCG; 5mCG/G5hmC homoduplex substrate: Top5mCG + Bot5hmCG; CG/G5fC substrate: TopCG + Bot5fCG; 5mCG/G5fC substrate: Top5mCG + Bot5fCG; CG/G5caC substrate: TopCG + Bot5caCG; 5mCG/G5caC substrate: Top5mCG + Bot5caCG; CG/Gab substrate: TopCG + BotabG.

Reaction mixtures (10 µL) containing 20 mM Tris–HCl pH 7.65, 50 mM NaCl, 3 mM MgCl₂, 5% glycerol, 1 mM DTT, 0.1 µg/µl BSA and 2 nM of end-labeled substrates was incubated for 20 min (excepted for kinetic experiment) at 37 °C with 60 nM of MBD4 (excepted as indicated). When indicated, PMS2 and MLH1 proteins were added to reaction mixture to a final concentration of 100 nM. The reaction was stopped by adding 10 µL formamide buffer (90% formamide, 10 mM EDTA, 0.1% blue bromophenol) and heating 5 min at 95 °C before loading on a 12% denaturing polyacrylamide gel. When indicated, the reaction was pre-treated with 1 µL of 1 M NaOH 10 min at 95 °C before the addition of formamide buffer. Gels were dried and quantified on a Typhoon 8600 Variable Mode Imager.

Enzymatic activities assays were done as described above. Products of reactions together with the products of the G + A and the C + T Maxam–Gilbert cleavage reactions performed on the same substrates were loaded on an 8% denaturing polyacrylamide gel.

As the protein MBD4 can repair the product of 5mC deamination, we asked how the cell uses the specific property of this enzyme to regulate the steady state of DNA methylation level. To this end we used WT (Mbd4^+/+) and KO (Mbd4^−/−) primary MEFs. We first hypothesized that the absence of MBD4 would generate alterations in gene promoter methylation patterns, which would in turn affect their transcriptional status. DNA methylation distribution was analyzed by reduced representation bisulfite sequencing (RRBS) [53]. RRBS provides single-nucleotide resolution and quantitative DNA methylation measurements for the majority of CG-rich regions such as CGIs and promoters [54]. Summary of the data quantity after each step of filtration is shown in Supplementary Fig. 1a and both samples showed near complete bisulfite conversion of non-CpG cytosines (> 99%). Since only half of all CGIs overlap with TSSs despite the fact that more than 70% of annotated gene promoters are associated with a CGI [55], we decided to analyze separately promoters from CGIs (Supplementary Fig. 1b). Our data covered 77% to 87% of cytosines theoretically covered by RRBS in both CGI and promoter regions for WT and KO cells (Supplementary Fig. 1c). Since approximately 95% of the detected 5mC were found to occur in CpG context (Supplementary Fig. 1d), we decided to focus our study only on 5mCG. The methylation level of CGIs and promoters was assessed by two different parameters: the percentage of methylated CpGs per region and the methylation level of each 5mCG. In agreement with previously published data [56], CGIs and promoters were found to be poorly methylated (6.3–7.4% of 5mCG) with little or no difference between WT and KO cells (Fig. 1a-c and Supplementary Tables 1–4). Importantly, the methylation levels of 5mCG in both CGIs and promoters significantly decreased in the absence of MBD4 (P < 10^–10 and P < 10^–8, respectively; Fig. 1d). For a significant portion of 5mCGs that were originally methylated over 20% in WT cells, their methylation levels dropped to less than 10% in KO cells (Fig. 1e-f and Supplementary Tables 1–4). These results strongly indicate that, although the number of methylated CpGs only modestly decreased in KO cells (Fig. 1a-c), their methylation levels significantly decreased in the absence of MBD4 (Fig. 1d-f). It is reasonable to conclude that MBD4 plays a protective role in preventing the demethylation of methylated CGs. Notably, the loss of methylation primarily affected promoters and CGIs with low methylation levels (Supplementary Fig. 1d-e).

We next investigated how DNA methylation defect affects gene expression. Genome-wide transcriptome analysis of Mbd4^−/− cells identifies a total of 1,258 genes (with P < 0.01 and ∣log₂ fold change∣ > 2)) having strong transcriptional de-regulation compared to the control Mbd4^+/+ cells where 802 of these genes were found up-regulated and 456 were found down-regulated (Fig. 1g and Supplementary Table 5). We hypothesized that the hypomethylated CGI promoters are up-regulated. To test this, we selected genes containing the most significantly demethylated proximal region (methylation level WT/KO > 2, number of CpG > 10, P < 0.01, distance to nearest TSS < 5 kb) and correlated their methylation level (Fig. 1h) to their transcriptional states (Fig. 1i). Accordingly, around 40% of the hypomethylated CGI promoters are transcriptionally up-regulated (P < 0.01) in the absence of MBD4. Clonal standard and bisulfite sequencing further confirmed the hypomethylation phenotype at proximal regions of Zic5, Tox and Brzap1 genes, and revealed that the methylation loss is not accompanied by C > T transitions (Fig. 1j).

These data, taken as a whole, illustrate that MBD4 is required for both preserving the methylation status of its target genes and maintaining them in a repressive state.

Our RRBS data reveal methylation loss at CpG-rich regions exhibiting low methylation levels in Mbd4^−/− cells. But the vast majority of the mouse genome is highly methylated and CpG-poor, such as repetitive elements that make up 40% of the genome. We then decided to extend our analysis to the globally methylated and CpG-poor genomic landscape by using DNA immunoprecipitation sequencing (DIP-seq). We have carried out a genome-wide comparative analysis of the 5mC (MeDIP-seq) and 5hmC (hMeDIP-seq) patterns in both MEFs WT and KO for Mbd4. We first analyzed the DNA methylation/oxidation pattern of gene bodies by analyzing uniquely mapped reads (Fig. 2a, left panel). Tag densities were calculated in 100 bp sliding windows spanning ± 2 kb of gene bodies using datasets normalized to the total number of unique reads. Interestingly, we observed a strong decrease in 5mC density across genes peaking at the 3’ end of gene bodies. In contrast, no difference was detected by analyzing 5hmC or input reads. These data show that MBD4 is also required to maintain DNA methylation in gene bodies.

A DIP-seq approach allows a high coverage of repetitive elements [57]. In order to highlight a potential function of MBD4 in retro-element repression, we extended our initial DIP-seq and RNA-seq analyses to include “unmappable” multihit reads (Fig. 2b, right panel). The unique reads can be mapped to specific genomic sites while the multihit reads, which represent repetitive elements [58], can only be assigned to their specific repeat family. Reads were aligned to repetitive elements in two passes (see materials and methods for details). The fold change in read counts between MEFs WT and KO for Mbd4 were calculated for each DNA repeat family, and represented as log₂(KO/WT) ratio (Fig. 2c, d and e for MeDIP-seq, hMeDIP-seq and RNA-seq, respectively). We restricted our analysis to retrotransposon families (including long terminal repeats (LTR), long (LINE) and short (SINE) interspersed nuclear elements), which account for more than 90% of the repetitive elements of the mouse genome. While we did not detect any change in 5hmC enrichment at retro-elements in Mbd4^−/− cells (Fig. 2d), we observed a global loss of 5mC affecting significantly 20% (86/410, P < 0.05 and |log₂ fold change|> 0.5) of the covered families (Fig. 2e). Importantly, the hypomethylation phenotype was accompanied by a global reactivation of retrotransposons affecting significantly 14% (56/410, P < 0.01 and |log₂ fold change|> 0.5) of all families. However, we only found six families that were both hypomethylated and reactivated (Fig. 2f). To better understand the limited correlation between the methylome and transcriptome, we conducted separate analyses for SINE (Supplementary Fig. 2a), LINE (Supplementary Fig. 2b), and LTR (Supplementary Fig. 2c-d) elements based on their evolutionary age. This approach enabled us to distinguish Mbd4 deficiency-dependent effects on lineage-specific (mouse) and ancestral retro-element families (Fig. 2g and Supplementary Fig. 2a-d). While the methylation loss was mainly detected for the oldest subfamilies (4% and 27% of the mouse-specific and ancestral retro-element subfamilies, respectively), the transcriptional derepression was more pronounced for the youngest subfamilies (23% and 11% of the mouse-specific and ancestral retro-element subfamilies, respectively). We attributed this apparent discrepancy to different CpG content, which is twofold higher in the youngest retro-elements than in the oldest one (Supplementary Fig. 2). MeDIP-seq data being strongly dependent to the local density of 5mC, and therefore to CpG content, we postulate that a moderate decrease in 5mC density is more detectable at CpG-poor retro-elements. Conversely, we hypothesized that MBD4 proteins are more prevalent at CpG-rich elements, making these regions transcriptionally more susceptible to MBD4 depletion and potentially subject to reactivation through both DNA methylation-dependent and -independent mechanisms. To conclude, our DIP-seq analysis reveal that the MBD4 function in DNA methylation maintenance is not restricted to punctuated and weakly methylated CpG-rich regions (as shown by single base resolution RRBS approach), but is also extended to the globally methylated CpG-poor genomic landscape.

In order to decipher the molecular mechanisms implicated in MBD4-dependent DNA methylation maintenance in vivo, we sought to study the enzymatic properties of MBD4. The properties of purified recombinant MBD4 have previously been analyzed and the reported data suggests that the recombinant protein exhibits G/T and G/U mismatch specific monofunctional glycosylase activity [27, 28, 59, 60]. The native MBD4 complex could, however, have features distinct from those of the MBD4 protein alone. To test this, we purified the epitope-tagged MBD4 complex (e-MBD4.com) from HeLa cells stably expressing hemagglutinin (HA) and FLAG epitope tagged MBD4 (Fig. 3a). MBD4 together with the DNA helicases TIP49A/B and the MMR proteins (MLH1 and PMS2) were identified as major components of e-MBD4.com by both mass spectrometry (Fig. 3b and Supplementary Table 10) and Western blotting (Fig. 3a, lower panel). Fractionation of the e-MBD4 complex on a glycerol gradient confirmed that MLH1 and PMS2 are stable components of this complex (Fig. 3c). This interaction was further validated by immunoprecipitating either the endogenous MBD4 (Fig. 3d) or MLH1 (Fig. 3e) complexes from non-tagged HeLa cell nuclear extracts using specific antibodies. We next analyzed the composition of the MBD4 complex in mouse embryonic fibroblast (MEF) cell lines stably expressing e-MBD4. Western blotting demonstrates that the MEF e-MBD4 complex exhibits the same composition as the e-MBD4 complex isolated from HeLa cells and thus, it should mechanistically function in the same way (Fig. 3f). We conclude that MBD4 forms in vivo a complex with the PMS2/MLH1 heterodimer.

Some common glycosylases are known to exhibit an endonuclease mismatch activity [61, 62]. To test if this is the case for the e-MBD4.com, we carried out nuclease assays on substrate DNA containing different types of mismatches. The data clearly show that: (i) the e-MBD4 complex is able to cleave the G/T (or G/U) mismatch (Supplementary Fig. 3a-b), but not the cytosine, the methyl-, the hydroxymethyl-, the formyl- or the carboxyl-cytosine substrates (Supplementary Fig. 3c), and (ii) the cleavage is achieved at the abasic site on the “T”-containing strand of a G/T mismatch substrate (Supplementary Fig. 3a). Importantly, no NaOH treatment of the reaction products was needed for generation of the cleavage products (Supplementary Fig. 3a). Therefore, the e-MBD4.com exhibits G/T mismatch specific endonuclease activity.

To analyze whether the e-MBD4.com endonuclease activity was dependent on the methylation status of the substrate, we carried out similar experiments, but with fully methylated (on both strands) G/T mismatch containing substrates by using identical amounts of either highly purified MBD4 alone (Fig. 4a) or MBD4 in the context of the e-MBD4.com. The purified MBD4 protein was able to induce some weak, non-methylation dependent cleavage of the substrate (Fig. 4b, upper panel and Fig. 4c). The e-MBD4.com shows ~ threefold higher activity for unmethylated DNA relative to that of the MBD4 protein (Fig. 4b-c). The e-MBD4.com endonuclease activity was, however, strongly methylation dependent and ~ 8–tenfold higher e-MBD4.com induced cleavage (compared to this for the MBD4 protein) for fully methylated substrates was measured (Fig. 4b-c). These data indicate that the association of MBD4 with its partners modulates its enzymatic properties and, as a result, MBD4 acquires a much higher G/T specific endonuclease activity, which isdependent on the methylation of the DNA substrate.

We next addressed the role of the methyl binding domain of MBD4 by generating a stable HeLa cell line expressing R97G mutated e-MBD4 (the substitution of R97 with G results in a dead methyl binding domain [18]). Both silver staining (Fig. 4d) and Western blotting (Fig. 4e) showed that the composition of the purified methyl binding dead e-MBD4 complex (e-MBD4.com R97G) is identical to the native e-MBD4.com. Nuclease assays revealed that the R97G mutant e-MBD4.com does not discriminate between unmethylated, hemi-methylated and fully methylated mismatch-containing substrates. In all cases, cleavage in the absence of carrier DNA is very low (~ 10%) and, with increasing concentrations of carrier DNA, this activity progressively decreases (Fig. 4f-g, right panels). In contrast, the native complex (e-MBD4.com WT) discriminates clearly between methylated and unmethylated substrates and its cleavage efficiency, compared to the mutant e-MBD4.com R97G, is much higher at the respective carrier DNA concentrations (Fig. 4f-g, left panels). In particular, dramatic differences in cleavage efficiencies for both complexes are measured for the fully methylated substrates (Fig. 4f-g). These results demonstrate that the e-MBD4.com endonuclease activity strongly depends on the degree of methylation of the DNA substrate and that the methyl-binding domain of MBD4 targets the MBD4-MMR complex on methylated DNA.

We next asked whether the MBD4 glycosylase domain is important for the observed G/T specific endonuclease activity. To this end we substituted amino acid residue D554 with alanine (A) and created a glycosylase dead mutant e-MBD4 protein [63]. Then, we established stable HeLa cell lines expressing the e-MBD4 D554A mutant. Both protein gel analysis and Western blotting show that the glycosylase dead mutant e-MBD4 D554A complex purified from the stable HeLa cell lines has a protein composition identical to the WT e-MBD4 complex (Fig. 5a-b). The mutant e-MBD4.com D554A exhibits, however, no endonuclease activity (Fig. 5c). We conclude that the activity of the MBD4 glycosylase domain is required for its endonuclease activity.

Past studies have demonstrated that MBD4 possesses G/T specific glycosylase activity [27, 28, 60]. Several glycosylases show, however, also AP lyase activity, i.e., the excision of the base is followed by AP lyase-mediated cleavage of the phosphate backbone at the abasic site [61, 62]. If the AP lyase product is generated through β elimination, it retains the abasic residue at its 3’ end (Fig. 5e and [64, 65]). This product migrates slower on sequencing PAGE than the respective product obtained from the Maxam & Gilbert sequencing reaction. Upon treatment of such AP-lyase generated products with NaOH, the phospho-diester bond is cleaved (δ elimination), the abasic residue is released and the mobility of the resulting product becomes identical to that generated by the Maxam & Gilbert sequencing reaction.

Our data suggest that MBD4 possesses an AP-lyase type endonuclease activity. To test this, we have used the above-described procedure (Fig. 5d-e). Treatment with NaOH of either MBD4 or e-MBD4.com reaction products resulted in clear increase of their migration rate in PAGE under denaturing conditions. The migration position of the NaOH treated products was identical to that of the products of the Maxam & Gilbert sequencing reaction obtained upon cleavage at the respective thymine (Fig. 5d). These observations reveal that MBD4 possesses an AP-lyase endonuclease activity, which operates through β elimination (Fig. 5e).

Having demonstrated the AP-lyase endonuclease activity of MBD4, we next sought to analyze the role of the MMR proteins in the regulation of this newly identified activity. We have expressed either separately or co-expressed together the recombinant MBD4, MLH1 and PMS2 proteins in the baculovirus system. We were able to reconstitute the MBD4/MLH1 and the MBD4/MLH1/PMS2 complexes but not the MBD4/PMS2 complex, suggesting a direct physical interaction between MBD4 and MLH1 (Fig. 6a). Reconstitution of the in vitro nuclease assay using individual recombinant proteins shows that MBD4, but not the MMR proteins, has some weak endonuclease activity (Fig. 6b), a result in agreement with the data presented in Fig. 4b and excluding a potential enzymatic function of MMR proteins in e-MBD4.com endonuclease activity. Remarkably, the presence of MLH1 results in a dramatic increase of the endonuclease activity of MBD4 with marked preference for fully methylated substrates (Fig. 6b-c). Note that the enzymatic dead MBD4 D554A mutant protein alone, or complexed with MLH1, did not exhibit endonuclease activity (Fig. 6c). Similarly, none of the purified recombinant proteins exhibited endonuclease activity towards a DNA substrate containing abasic sites (cleavable by the recombinant apurinic/apyrimidinic endonuclease APE1), ruling out the presence of a non-specifically associated contaminating AP-lyase activity (Fig. 6b, right panel and Fig. 6c, right panel). All together, these data reveal that the recombinant MLH1 alone and not some biochemically undetectable contaminant is responsible for stimulating the MBD4 endonuclease activity. The nuclease time-course assay (Fig. 6d-e) shows that the purified MBD4/MLH1 complex cuts much more efficiently the G/T mismatch substrate compared to MBD4 alone. The most drastic differences in the cleavage efficiency are measured for fully methylated substrates, where the cleavage kinetics (as assessed by the initial slope of the nuclease assay time course curve) of the MBD4/MLH1 is at least tenfold higher than for MBD4 alone (Fig. 6d-e). Therefore, by using recombinant components we have been able to reconstitute in vitro the enzymatic properties of the isolated native MBD4 complex and demonstrate the crucial role that MLH1 plays in the MBD4 complex endonuclease activity.