ERCC1–XPF cooperates with CTCF and cohesin to facilitate the developmental silencing of imprinted genes

Inborn defects in DNA repair are associated with complex developmental disorders whose causal mechanisms are poorly understood. Using an in vivo biotinylation tagging approach in mice, we show that the nucleotide excision repair (NER) structure-specific endonuclease ERCC1–XPF complex interacts with the insulator binding protein CTCF, the cohesin subunits SMC1A and SMC3 and with MBD2; the factors co-localize with ATRX at the promoters and control regions (ICRs) of imprinted genes during postnatal hepatic development. Loss of Ercc1 or exposure to MMC triggers the localization of CTCF to heterochromatin, the dissociation of the CTCF–cohesin complex and ATRX from promoters and ICRs, altered histone marks and the aberrant developmental expression of imprinted genes without altering DNA methylation. We propose that ERCC1–XPF cooperates with CTCF and cohesin to facilitate the developmental silencing of imprinted genes and that persistent DNA damage triggers chromatin changes that affect gene expression programs associated with NER disorders. Chatzinikolaou et al. show that the nucleotide excision repair complex ERCC1–XPF cooperates with the chromatin organizer CTCF, cohesin subunits and ATRX to facilitate the silencing of a subset of imprinted genes in the developing liver.

Inborn defects in DNA repair are associated with complex developmental disorders whose causal mechanisms are poorly understood. Using an in vivo biotinylation tagging approach in mice, we show that the nucleotide excision repair (NER) structure-specific endonuclease ERCC1-XPF complex interacts with the insulator binding protein CTCF, the cohesin subunits SMC1A and SMC3 and with MBD2; the factors co-localize with ATRX at the promoters and control regions (ICRs) of imprinted genes during postnatal hepatic development. Loss of Ercc1 or exposure to MMC triggers the localization of CTCF to heterochromatin, the dissociation of the CTCF-cohesin complex and ATRX from promoters and ICRs, altered histone marks and the aberrant developmental expression of imprinted genes without altering DNA methylation. We propose that ERCC1-XPF cooperates with CTCF and cohesin to facilitate the developmental silencing of imprinted genes and that persistent DNA damage triggers chromatin changes that affect gene expression programs associated with NER disorders.
To counteract DNA damage, mammalian cells have evolved partially overlapping DNA repair systems to remove DNA lesions and restore their DNA back to its native form 1,2 . For bulky helixdistorting ultraviolet (UV)-induced DNA lesions, the principal repair mechanism is the evolutionarily conserved nucleotide excision repair (NER) pathway 3 . NER recognizes and removes helical distortions throughout the genome, that is, global genome NER, or selectively from the transcribed strand of active genes, that is, transcriptioncoupled repair. The NER structure-specific endonucleases XPG and ERCC1-XPF cleave on the 3 and 5 side of the DNA lesion, respectively, to release a 24-32-nucleotide fragment containing the damaged DNA [4][5][6] . Besides NER, the ERCC1-XPF complex also participates in the repair of DNA interstrand crosslinks (DNA ICLs) 7,8 and for the completion of homologous recombination at DNA replication forks stalled by DNA ICLs 9, 10 .
In humans, mutations in NER genes lead to the skin cancer-prone xeroderma pigmentosum or to a heterogeneous group of premature ageing-like (progeroid) disorders, including Cockayne syndrome (associated genes: Csa, Csb, Xpd, Xpb) and trichothiodystrophy (TTD; associated genes: Xpd, Xpb) [11][12][13][14][15] . Patients with subtle mutations in Xpf show mild xeroderma pigmentosum features and develop cancer during adulthood 16 . Instead, a patient with an Ercc1 mutation showed severe clinical abnormalities coupled with a relatively mild DNA repair defect 17 . This and the fact that a complete defect in NER is compatible with life 18 argues for XPF-ERCC1 having functions outside the canonical NER 11 .
Indeed, NER factors are now known to play a role, in addition to DNA repair, in the regulation of gene expression 19,20 , chromatin looping 21 , the transcriptional reprogramming of pluripotent stem cells 22 and the fine-tuning of growth-promoting genes during postnatal development 23 . At present, however, no solid evidence exists as to how NER is functionally involved in these processes, what are the NER-bound protein factors involved and their in vivo relevance to NER disorders. To tackle this, we used an in vivo biotinylation tagging approach in mice and mutant animals to dissect the functional contribution of ERCC1-XPF during liver development.
A R T I C L E S a 10 WT allele 11

Generation of biotin-tagged XPF mice
We generated knock-in mice expressing the NER structure-specific endonuclease XPF fused carboxy terminally before the stop codon of the last exon 11 with a 15-amino-acid tandem affinity purification (TAP) tag biotinylatable sequence 24 and a 3× Flag tag separated by a tobacco etch virus protease cleavage site for easy tag removal (Fig. 1a). After transfection in 129/SV embryonic stem cells expressing the Protamine 1-Cre recombinase transgene to efficiently excise the neomycin cassette in the male germ line 25 and selection of properly targeted clones ( Fig. 1a and Methods), we used two independent transfected clones to generate germ-line-transmitting chimaeras (Fig. 1b,c). TAP-tag-fused heterozygous males (avXpf +/− ) were backcrossed and maintained in a C57BL/6J background. Homozygous avXpf +/+ knock-in mice were then crossed with mice broadly expressing the HA-tagged bacterial BirA biotin ligase transgene under the control of the human hnRNPA2B1/CBX3 methylation-free island 26 (Fig. 1d and Supplementary Fig. 1A). BirA is a bacterial ligase that specifically recognizes and efficiently biotinylates the 15-aminoacid avidin within the short 15-amino-acid tag (Fig. 1e,f), thus creating a high-affinity 'handle' for the in vivo isolation of XPF-bound protein complexes from protein extracts isolated from biotin-tagged XPF (bXPF) mice by binding to streptavidin. Unlike Ercc1 −/− or Xpf −/− mice 27,28 , bXPF animals are born at the expected Mendelian frequency (Fig. 1g), grow normally (Fig. 1h) and show no developmental defects or other pathological features (Fig. 1i). Importantly, primary bXPF mouse embryonic fibroblasts (MEFs) show no hypersensitivity to UV (Fig. 1j) or to the DNA crosslinker mitomycin C (MMC; Fig. 1k) and no detectable differences in UV-induced unscheduled DNA synthesis when compared to wild-type (WT) control MEFs (Fig. 1l). Thus, bXPF animals develop normally to adulthood and are NER-and DNA ICL-repair-proficient.  A proteomics strategy reveals ERCC1-XPF protein conjugates involved in genome organization and chromosome architecture To isolate and characterize NER-associated protein complexes during postnatal murine development, we prepared nuclear extracts from P15 avXPF; BirA livers (designated as bXPF) and livers expressing only the BirA transgene using high-salt extraction. Next, we confirmed that bXPF can still interact with known protein partners involved in NER, that is, ERCC1 (ref. 11), or transcription, that is, TAF10 (ref. 23) ( Fig. 2a and Supplementary Fig. 8A). We then separated the bound protein interactome by one-dimensional SDS-PAGE followed by in-gel digestion (∼12 slices) and peptides were further separated and analysed with high-resolution liquid chromatography-tandem mass spectrometry (nLC-ESI-MS/MS) on a hybrid linear ion trap Orbitrap mass spectrometer (Fig. 2a). From three biological replicates, we identified a total of 306 proteins (Supplementary Table 1 . Using a hierarchical clustering approach, we confirmed that the 306 bXPF-bound proteins are capable of classifying the bXPF knock-in and BirA transgenic livers into the expected groups ( Fig. 2c). At the confidence interval used, that is, false discovery rate (FDR) < 0.05, the significantly over-represented GO terms found (Fig. 2d) involved 65 out of the initial 140 XPF-bound core proteins; this set of proteins showed a significantly higher number of known protein interactions (that is, 63 interactions) than expected by chance (that is, 28 interactions; Fig. 2e) indicating a functionally relevant and interconnected protein network. Using this data set, we were able to discern four major XPF-bound protein complexes involved in: transcription silencing; transcription initiation; DNA replication; and DNA repair ( Fig. 2f; shown in this order). These findings confirm previously documented interactions of ERCC1-XPF with components of the TFIID complex 23 and NER 11 , whilst they reveal interactions of ERCC1-XPF with factors associated with transcription repression and DNA replication; the latter probably reflects the functional role of the ERCC1-XPF complex in homologous recombination repair.

A R T I C L E S
Ablation of Ercc1 gene triggers aberrant expression of imprinted genes during hepatic development ERCC1-XPF is a highly conserved heterodimeric complex. Ercc1 −/− mice are growth-defective, show progeroid features in several organs and die of liver failure within a month after birth 29,30 . Streptavidin pulldown identified CTCF co-purifying with bXPF under native (micrococcal nuclease digested) chromatin conditions ( Supplementary Fig. 1B). This and our findings that XPF interacts with proteins known to be involved in chromatin organization and gene silencing, that is, CTCF, SMC1A, SMC3 and MBD2 (refs 31-33), prompted us to examine their relevance to the developmental defects seen in Ercc1 −/− animals 34 . Co-immunoprecipitation experiments confirmed that a portion of endogenous ERCC1 interacts with CTCF, SMC1A, SMC3 and MBD2 in P15 livers; the protein-protein interactions were not affected by treating the extracts with benzonase ( Fig. 3ai and Supplementary Fig. 8B). An antibody raised against CTCF confirmed the reciprocity of the interactions ( Fig. 3aii and Supplementary Fig. 8B). We also challenged the specificity of ERCC1 interactions in Ercc1 −/− livers ( Fig. 3bi and Supplementary Fig. 8B); notably, CTCF, SMC1A, SMC3 and MBD2 interact in Ercc1 −/− livers ( Fig. 3bii and Supplementary Fig. 8B).
The functional role of CTCF, the cohesin and MBD2 in genomic imprinting or the postnatal silencing of distinct genes is well documented 32,33,35,36 . Imprinted genes are expressed in a parent-of-originspecific manner already established in the gametes 37 . During embryogenesis only one parental allele is expressed, while the other allele is silenced 35 . Soon after birth, the remaining active allele of several imprinted genes is also silenced 38 . Using liver gene expression data sets 39 Table 3). Unlike Csb m/m , Xpa −/− or Xpd TTD livers, we find that 22 out of the 68 imprinted genes show significantly aberrant gene expression profiles in P15 Ercc1 −/− livers (P < 0.05; fold change > ±1.2) with no preference for genes expressed from the maternal or the paternal allele (Supplementary Table 4); the great majority of imprinted genes showed increased messenger RNA levels (17 out of 22; Fig. 3c) and associate with GO terms related to developmental and endocrine disorders ( Supplementary Fig. 1C). A deregulation in the expression of eight imprinted genes was also observed in Csb m/m /Xpa −/− livers (Fig. 3c). For further studies, we focused on insulin growth factor-2 (Igf2), the paternally expressed gene 3 (Peg3), the protein delta homologue 1 (Dlk1) and the growth factor receptor-bound protein 10 (Grb10) genes in the Ercc1 −/− and WT animals during hepatic development [42][43][44][45] .
The Igf2, Peg3, Dlk1 and Grb10 genes showed a progressive postnatal decline in mRNA levels compared with E18.5 that in Csb m/m , Xpa −/− and WT livers led to barely detectable levels at day P60 (Fig. 3d). Instead, in Ercc1 −/− livers, we find a gradual but steady postnatal increase in the mRNA levels of these genes when compared with age-matched WT livers (Fig. 3e). To test whether the increased mRNA levels in P15 Ercc1 −/− livers reflect an embryonic defect in genome imprinting or a defect in the postnatal silencing of imprinted genes, we extended our studies in E18.5 Ercc1 −/− fetal livers. Despite the marginally smaller size of the ERCC1-null embryos (Fig. 3f), we find no significant differences in the mRNA levels of Igf2, Peg3, Dlk1 and Grb10 genes in E13.5 Ercc1 −/− livers compared with agematched WT controls (Fig. 3g). To determine the allelic source of imprinted gene transcripts in P15 Ercc1 −/− livers, we identified a single polymorphic site within exon 9 of the Peg3 gene in P15 Ercc1 −/− livers that are generated when FVB Ercc1 −/+ females are crossed with C57BL/6J Ercc1 −/+ males; Peg3 is maternally methylated and is expressed solely from the paternal allele in the liver 44 . Our analysis revealed that the increased expression of the Peg3 gene was not due to reactivation of the maternal allele, because transcripts were still derived solely from the paternal allele (Fig. 3h). Further analysis revealed increased Igf2, Grb10, Peg3, Meg3, Atp10a, H13 and Airn mRNA levels in the kidney, white adipose tissue, pancreas, cerebellum and spleen of the P15 ERCC1-defective animals, compared with age-matched WT animals ( Fig. 3i and Supplementary Fig. 1D). Interestingly, Ercc1 inactivation in aP2-Ercc1 F/− animals occurs at ∼P15 in mature adipocytes 46 when imprinted gene expression has already been established. Thus, unlike other NER-defective animals, a defect in ERCC1-XPF triggers the aberrant postnatal expression of a subset of imprinted genes.
Genomic imprinting is established already in gametes by parentof-origin-specific epigenetic marks, such as DNA methylation 47 . We, therefore, examined the DNA methylation status of the Igf2, Peg3, Dlk1 and Grb10 promoters, the Peg3, H19/Igf2 (from now on designated as H19) and Meg3/Dlk1 intergenic differentially methylated region (DMR) (from now on designated as Meg3) ICRs and the Grb10 DMR in P15 Ercc1 −/− and WT livers. Using a bisulfite conversion and sequencing assay on well-defined CTCF-and RNAPII-bound loci in Peg3, H19 and Meg3 ICRs and Grb10 (CGI2)/Meg3 DMRs ( Fig. 4c and Supplementary Fig. 2A,G), we detected no difference on methylation in Ercc1 −/− compared to WT livers. The H19/Igf2 genomic domain is regulated by long-range chromatin interactions in the liver, a process mediated in part by the DMRs and their DNA methylation state 48,49 . However, similar to ICRs, the DNA methylation state was also preserved in the Igf2 DMR1 regulatory region in both WT and Ercc1 −/− livers (Fig. 4c). Loss of ERCC1 in the liver resulted in reduced methylation of the Peg3 promoter (Supplementary Fig. 2B; as shown), but no change in DNA methylation was observed between Ercc1 −/− and WT livers at the CTCF peak within the Grb10 region, the Igf2 promoters 2 and 3, and the Grb10 promoter ( Supplementary Fig. 2C-F; as shown). Thus, in P15 Ercc1 −/− livers, the increased mRNA levels of imprinted genes occurs without affecting DNA methylation in ICRs or the DMRs. ii.  Disruption of Ercc1 leads to the dissociation of the CTCF-cohesin-MBD2 interacting factors from the promoters and ICRs of imprinted genes ChIP followed by qPCR on the well-characterized CTCF-and RNAPII-bound H19, Igf2, Dlk1/Meg3, Peg3 and Grb10 target sequences (Supplementary Fig. 2A) showed that bXPF assembles with CTCF, SMC1A, SMC3 and MBD2 on promoters ( Supplementary   Fig. 3A) and the ICRs/DMR (Fig. 5a) in P15 WT livers but not in CTCF-negative regions ( Supplementary Fig. 3B). We also find ATP-dependent helicase ATRX to recruit to the respective promoters and the ICRs (Supplementary Fig. 3A and Fig. 5a); ATRX was found to interact with bXPF in our proteomics studies and is known to cooperate with cohesin in silencing a subset of imprinted genes during postnatal murine brain development 32 . ChIP for CTCF or SMC1A and re-ChIP for SMC3, ERCC1, Flag-tagged XPF, MBD2 and ATRX showed that these factors co-occupy the H19 and Peg3 ICRs (Fig. 5b). Instead, ChIP signals for all factors tested were significantly reduced in promoters and the ICRs in P15 Ercc1 −/− livers compared with WT controls (Fig. 5c and Supplementary  Fig. 3C). Thus, the ERCC1-XPF complex is required for the optimal promoter and ICR assembly of protein complexes that associate with the postnatal silencing of imprinted genes in the developing liver.
CTCF, SMC1A, SMC3 and ATRX are known to display allelespecific binding at imprinted loci 32,33,35,36 . We, therefore, envisioned a similar scenario for ERCC1-XPF in the developing liver. ChIP followed by allele-specific restriction digest analysis of amplified DNA in P15 bXPF/SPRET/EiJ F1 livers that are polymorphic within the Peg3 and H19 ICRs showed that CTCF, SMC1A and SMC3 are preferentially enriched on the maternal and the paternal allele of the H19 and Peg3 ICRs, respectively (Fig. 5d); MBD2 and ATRX are preferentially recruited at the paternal and the maternal alleles of the H19 and Peg3 ICRs, respectively. The observation that CTCF and ATRX are localized to opposite alleles in H19 and Peg3 ICRs was unexpected as ERCC1-XPF assembles with these factors on ICRs (Fig. 5a) and interacts with CTCF and ATRX in P15 livers ( Fig. 2 and Fig. 3a,b). We, therefore, reasoned that the ERCC1-XPF complex resides in both the paternal and the maternal alleles of H19 and Peg3 ICRs. In line, allele-specific restriction digest analysis of amplified immunoprecipitated or pulled-down DNA in P15 bXPF/SPRET/EiJ F1 livers showed no allele preference for either ERCC1 or bXPF on the ICRs (Fig. 5d).

SphI
To test whether DNA damage directly contributes to the aberrant localization of CTCF and ATRX in Ercc1 −/− MEFs, naive primary WT MEFs were exposed to UVC, H 2 O 2 and MMC, a potent DNA crosslinker that, similar to the ERCC1 defect, triggers DNA ICLs that are efficiently processed by ERCC1-XPF 51,52 in a mechanism (g) Immunofluorescence detection of CTCF in primary MMC-treated MEFs exposed to ATM (ATMi) or ATR (ATRi) inhibitors. Note the absence of CTCF translocation to heterochromatin in MMC-treated MEFs exposed to ATMi (indicated with arrows). (h) Immunofluorescence detection of ATRX in primary MMC-treated MEFs exposed to ATM (ATMi) inhibitor (see also Supplementary  Fig. 3). Note the absence of ATRX accumulation to heterochromatin in MMCtreated MEFs exposed to ATMi (indicated with arrows). (i) qPCR mRNA levels (expressed as fold change to untreated control MEFs) of Igf2, Peg3, Dlk1 and Grb10 genes in primary MEFs exposed to MMC, UV or H 2 O 2 (n = 3 biological replicates distinct from NER 9 and the accumulation of DNA damage-associated γ H2AX foci in the nucleus (Supplementary Fig. 4I). As with Ercc1 −/− MEFs, we find that CTCF and ATRX are predominantly localized in heterochromatic regions in MMC-treated MEFs but, importantly, not following exposure of MEFs to UVC irradiation or to H 2 O 2 -induced oxidative DNA damage (Fig. 6f). The γ H2AX foci accumulate only in the nucleoplasm and unlike with CTCF, they do not appear in the chromocentres of Ercc1 −/− and MMC-treated MEFs ( Supplementary Fig. 4I,J). Translocation of CTCF was also visible in G0 synchronized, serum-starved, MMC-treated MEFs ( Supplementary Fig. 5A). Likewise, ATRX accumulates with CTCF in heterochromatin of MMC-but not of UVC-or H 2 O 2 -treated cells ( Supplementary Fig. 5B,C). In line, we find increased mRNA levels for all genes tested in MMC-but not in UVC (4 J m −2 )-or H 2 O 2 -treated cells (Fig. 6i); increased mRNA levels were also detected at the much higher dose of 10 J m −2 of UVC probably reflecting other types of damage 53 . Similar findings were seen for a larger subset of imprinted genes in primary MEFs that also show increased mRNA levels for genes known to be induced following exposure to 4 J m −2 of UVC ( Supplementary Fig. 6A) or following treatment with H 2 O 2 ( Supplementary Fig. 6B). Inhibition of ATM (ATMi) with KU-55933 inhibitor 54 in MMC-treated MEFs significantly abrogated the accumulation of CTCF and ATRX in heterochromatin (Fig. 6g,h). In ATMi-treated cells, we also evidenced the normative expression levels for Igf2, Peg3, Dlk1 and Grb10 genes compared with untreated control cells (Fig. 6j). Inactivation of ATR with the ATR/CDK inhibitor NU6027 in MMC-treated MEFs led to similar results to those seen following ATM inactivation, albeit to a smaller magnitude (Fig. 6g,h,j). We find that CTCF does not interact with components of the Fanconi anaemia, that is, FANCA or FANCD2 and, unlike in Ercc1 −/− MEFs ( Supplementary Fig. 5D), CTCF does not translocate to heterochromatin in Fanca −/− or Fancd2 −/− MEFs ( Supplementary Fig. 5E). Importantly, exposure of MEFs to MMC led to substantially reduced ChIP signals on the ICRs/DMR and the promoters for all genes tested ( Fig. 7a and Supplementary Fig. 6C).
With the exception of the Dlk1 promoter, we find normative ChIP signals for all gene promoters and ICRs/DMR tested in ATMi-and, in part, also in ATRi-treated MEFs exposed to MMC (Fig. 7a and Supplementary Fig. 6C), suggesting that such chromatin changes depend on functional DNA damage response signalling. In line with Ercc1 −/− livers (Fig. 3h), exposure of C57Bl/6/SPRET/EiJ F1 MEFs to MMC that are polymorphic within the H19 and Peg3 ICRs led to substantially reduced ChIP signals for CTCF, SMC1A and SMC3 from the paternal allele (SPRET/EiJ) and the maternal (C57Bl/6) alleles of Peg3 and the H19 genes, respectively ( Fig. 7b and Supplementary  Fig. 7A) and to increased Peg3 mRNA levels (Fig. 7c).
As for P15 Ercc1 −/− livers, we find the loss of repressive histone H3K9 trimethylation and H3K27 trimethylation marks on the ICRs/DMR (Fig. 7d) and promoters (Supplementary Fig. 7B) in MMC-but not in UVC-or H 2 O 2 -treated MEFs. We also find the concomitant increase of activating acetylated histone H3K9 on the H19 and Peg3 ICRs and on all gene promoters tested ( Supplementary Fig. 7C-D). Abrogation of ATM-and to a lesser extent also of ATR-led again to normative ChIP signals for these histone marks compared with untreated MEFs (Fig. 7d and Supplementary Fig. 7B-D). We find no difference for activating histone H3K4 trimethylation in H19/Igf2, Peg3, Meg3/Dlk1 and Grb10 ICRs/DMR or promoters in these cells ( Supplementary Fig. 7C-D).

DISCUSSION
Using NER-defective animals and an in vivo biotinylation tagging approach in mice coupled to high-throughput proteomics, our data link ERCC1-XPF to important regulators of chromatin structure, to the proper occupancy of CTCF, the cohesin complex, MBD2 and ATRX at four imprinted loci and to the timely silencing of a subset of imprinted genes in the developing liver.
Importantly, ERCC1-XPF is not required for the assembly of the CTCF-cohesin complex and MBD2 in developing Ercc1 −/− livers and is not involved in genome imprinting itself, a process already established in gametes. However, the heterodimeric complex is required for the recruitment of the CTCF-cohesin complex, MBD2 and ATRX to promoters and ICRs during hepatic development. Unlike CTCF or ATRX, the ERCC1-XPF complex is found on both alleles of H19 and Peg3 ICRs. This would allow ERCC1-XPF to interact with CTCF and/or ATRX for optimal gene silencing during embryogenesis or postnatal development.
A central aspect of these findings is their possible relevance to associated developmental disorders. A recently reported patient with ERCC1 deficiency presented with microcephaly, growth retardation and neurological abnormalities 17 . These pathological features also manifest in patients with CTCF haploinsufficiency 55 , with Cornelia de Lange syndrome associated with mutations affecting the cohesin complex or, intriguingly, the RAD21 protein also involved in doublestrand break repair 56 and in male children carrying a mutation in ATRX 57 .
The involvement of CTCF and the cohesin in specific chromatin loop formation 49,58 and the influence of these structures in transcriptional regulation are well documented 36,59 . XPG and XPF were recently shown to be required for establishing CTCF-dependent chromatin looping between the promoter and terminator of the activated RARβ2 gene in HeLa cells 19,21 . This and the data presented herein suggest a similar role for ERCC1-XPF in facilitating long-range looping, thereby altering gene expression in the developing liver. The lack of any comparable defects in the single NER mutant Csb m/m , Xpa −/− or Xpc −/− livers suggests that this mechanism does not require functional NER.
As for Ercc1 −/− mice, exposure of cells to DNA ICLs-but not to other types of DNA lesion also repaired by NER-triggers chromatin changes and aberrant histone post-translational modifications associated with active transcription and the localization of CTCF and ATRX to heterochromatin. In line, animals carrying inborn defects in Ercc1, Xpf or Xpg (required for DNA damage incision) are the only single gene mutations in NER known to manifest with severe developmental defects [60][61][62] . All other single NER mutations result in phenotypically healthy animals with minor to moderate progeroid features 63 . The heterochromatic localization of CTCF and ATRX was accompanied by the dissociation of the CTCF-cohesin complex, MBD2 and ATRX from promoters and ICRs in MMC-treated or Ercc1 −/− MEFs and in postnatal Ercc1 −/− livers. Inactivation of ATM, and to a lesser extent of ATR, abolished the DNA damage-driven accumulation of CTCF and ATRX to heterochromatin, the release of repressor complexes from promoters and ICRs and the optimal gene silencing of the active allele. Thus, the DNA damage-driven reorganization of chromatin structure is reversible and requires functional DNA damage response. Indeed, the high affinity of the ERCC1-XPF complex for persistent DNA ICLs 64 in MMC-treated MEFs could trigger the displacement of the heterodimeric complex from high-order chromatin structures relevant for gene regulation to DNA damage sites requiring DNA repair (Fig. 7d).
It has been challenging to delineate how DNA damage drives the onset of tissue-specific, developmental defects. Here, we provide evidence for a functional link between ERCC1-XPF, DNA ICLs, chromatin architecture and gene silencing during hepatic development. Further studies are necessary to reveal how chromatin organizers respond to DNA damage during development or with disease onset.

METHODS
Methods, including statements of data availability and any associated accession codes and references, are available in the online version of this paper.  23,30,40 . Total RNA was isolated from liver, heart, kidney, spleen and lung of P15 Ercc1 −/− mice using a Total RNA isolation kit (Qiagen) as described by the manufacturer. Quantitative PCR (qPCR) was performed with a DNA Engine Opticon device according to the instructions of the manufacturer (MJ Research). The generation of specific PCR products was confirmed by melting curve analysis and gel electrophoresis. Each primer pair was tested with a logarithmic dilution of a cDNA mix to generate a linear standard curve (crossing point (CP) plotted versus log of template concentration), which was used to calculate the primer pair efficiency (E = 10 (−1/slope) ). Hypoxanthine guanine phosphoribosyltransferase1 (Hprt-1) mRNA was used as an external standard. For data analysis, the second derivative maximum method was applied: ). Primer sequences are listed in Supplementary Table 5.
Allele-specific ChIP analysis. Allele-specific ChIP analysis was performed on P15 bXPF(C57BL/6)-SPRET/EiJ F1 livers. PCR amplification using unbiased primers was performed for the H19 and Peg3 ICR regions. The amplified product was purified and subsequently digested with specific restriction enzymes for 2 h. The digested DNA was resolved on a 2% agarose gel and visualized using an ultraviolet transilluminator (BioRad). Primer sequences are available in Supplementary Table 5. Allele-specific sequences were retrieved from the Mouse Genomes Project (Sanger Institute) and primers were designed with Batch Primer 3.0. All sets of primers were validated for allele specificity by qPCR amplification of strain/species pure or mixed genomic DNA from C57BL/6NJ, FVB/NJ and SPRET/Eij. Bisulfite mutagenesis. Genomic DNA isolated from P15 livers of Ercc1 −/− and littermate control mice was mutagenized with sodium bisulfite by using an EpiTect Bisulfite Conversion Kit (QIAGEN) according to the manufacturer's instructions. PCR amplification was carried out with primers specific for bisulfite-treated DNA. Primer sequences are available in Supplementary Table 5. All promoter and DMR regions were amplified by PCR. The resulting PCR products were ligated into the pCRII vector using a TOPO-TA cloning kit (Invitrogen), according to the manufacturer's instructions. Positive clones were sequenced, and clones were only accepted at >94% cytosine conversion. To ensure that the selected clones were derived from a unique DNA template we used non-converted cytosines and mismatched base pairs as reference points for each clone.

Data analysis.
A two-tailed t-test was used to extract the statistically significant data by means of the IBM SPSS Statistics 19 (IBM), Spotfire (Tibco), Partek (Partek INCoR1porated) and R-statistical package (www.r-project.org). Significant over-representation of pathways and gene networks was determined by DAVID (http://david.abcc.ncifcrf.gov/summary.jsp; through BBID, BIOCARTA and KEGG annotations) as well as by means of the Ingenuity Pathway Analysis software (www.ingenuity.com). For mass spectrometry (MS), the MS/MS raw data were loaded in Proteome Discoverer 1.3.0.339 (ThermoFischer Scientific) and run using the Mascot 2.3.02 (Matrix Science) search algorithm against the Mus musculus theoretical proteome (last modified 6 July 2015) containing 46,470 entries in Uniprot. A list of common contaminants was included in the database 67 . For protein identification, the following search parameters were used: precursor error tolerance 10 ppm, fragment ion tolerance 0.8 Da, trypsin full specificity, maximum number of missed cleavages 3 and cysteine alkylation as a fixed modification. The resulting .dat and .msf files were subsequently loaded and merged in Scaffold (version 3.04.05, Proteome Software) for further processing and validation of the assigned MS/MS spectra. Thresholds for protein and peptide identification were set to 99% and 95% accordingly, for proteins with minimum 1 different peptides identified, resulting in a protein false discovery rate (FDR) of <0.1%. For single peptide identifications, we applied the same criteria in addition to manual validation of MS/MS spectra. Protein lists were constructed from the respective peptide lists through extensive manual curation based on previous knowledge. For label-free relative quantitation of proteins, we applied a label-free relative quantitation method between the different samples (control versus bait) to determine unspecific binders during the affinity purification. All .dat and .msf files created by Proteome Discoverer were merged in Scaffold where label-free relative quantification was performed using the total ion current (TIC) from each identified MS/MS spectra. The TIC is the sum of the areas under all the peaks contained in a MS/MS spectrum and total TIC value results by summing the intensity of the peaks contained in the peak list associated to a MS/MS sample. Protein lists containing the Scaffold-calculated total TIC quantitative value for each protein were exported to Microsoft Excel for further manual processing including categorization and additional curation based on previous knowledge. The fold change of protein levels was calculated by dividing the mean total TIC quantitative value in bait samples with the mean value of the control samples for each of the proteins. Proteins having ≥60% protein coverage, ≥1 peptide in each sample and a fold change ≥1.2 in all three measurements were selected as being significantly enriched in bXPF compared with BirA liver samples. Proteins that were significantly enriched in bait samples were considered these with P value ≤0.05 and a fold change ≥2. Significant over-representation of pathways, protein-protein interactions and protein complexes were derived by STRING 68 (http://string-db.org).

Statistics and reproducibility.
Experiments were repeated at least 3 times. The data exhibited normal distribution (where applicable). There was no estimation of group variation before experiments. Error bars indicate standard deviation unless stated otherwise, that is, s.e.m. For animal studies, each biological replicate consists of 3-5 mouse tissues or cell cultures per genotype per time point or treatment. No statistical method was used to predetermine sample size. None of the samples or animals was excluded from the experiment. The animals or the experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment. The representative images shown in Figs 1d-f,i, 2a, 3a,b,f, 4c, 5b,d and 6a,c-h and Supplementary Figs 1A,B, 2B-G, 4 and 5 have been repeated more than 3 times. Data availability. Previously published microarray data that were reanalysed here are available under accession codes E-MEXP-3930 and E-MEXP-835 (ref. 23,30,40). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository 69 with the data set identifier PXD005897. Source data for Figs 1 and 3-7 and Supplementary Figs 1, 3, 4, 6 and 7 have been provided as Supplementary Table 6 (Statistics source data). All other data supporting the findings of this study are available from the corresponding author on reasonable request.