KDM4B-mediated reduction of H3K9me3 and H3K36me3 levels improves somatic cell reprogramming into pluripotency

Correct reprogramming of epigenetic marks is essential for somatic cells to regain pluripotency. Repressive histone (H) lysine (K) methylation marks are known to be stable and difficult to reprogram. In this study, we generated transgenic mice and mouse embryonic fibroblasts (MEFs) for the inducible expression of KDM4B, a demethylase that removes H3 K9 and H3K36 trimethylation (me3) marks (H3K9/36me3). Upon inducing Kdm4b, H3K9/36me3 levels significantly decreased compared to non-induced controls. Concurrently, H3K9me1 levels significantly increased, while H3K9me2 and H3K27me3 remained unchanged. The global transcriptional impact of Kdm4b-mediated reduction in H3K9/36me3 levels was examined by comparative microarray analysis and mRNA-sequencing of three independent transgenic MEF lines. We identified several commonly up-regulated targets, including the heterochromatin-associated zinc finger protein 37 and full-length endogenous retrovirus repeat elements. Following optimized zona-free somatic nuclear transfer, reduced H3K9/36me3 levels were restored within hours. Nevertheless, hypo-methylated Kdm4b MEF donors reprogrammed six-fold better into cloned blastocysts than non-induced donors. They also reprogrammed nine-fold better into induced pluripotent stem cells that gave rise to teratomas and chimeras. In summary, we firmly established H3K9/36me3 as a major roadblock to somatic cell reprogramming and identified transcriptional targets of derestricted chromatin that could contribute towards improving this process in mouse.

Different somatic cell types within an individual are morphologically and functionally diverse, yet they share the same genetic information. Epigenetic modifications, such as DNA methylation and histone modifications, control cell-specific gene activity and stabilize phenotypic differences. 'Reprogramming' of these epigenetic marks occurs twice in normal mammalian development, first during gametogenesis and next during preimplantation embryogenesis 1 . This second wave resolves the early parental asymmetry in histone modifications, DNA methylation and chromatin proteins to allow for correct embryonic gene activation 2,3 . It generates methylation marks in both DNA and histones that coincide with the first differentiation event during development, namely the specification of outer trophoblast and inner cell mass (ICM) of the early blastocyst 4 . The ICM later segregates into hypoblast and epiblast that develop into extra-embryonic and embryo tissues, respectively. Early epiblast cells are naturally pluripotent, i.e. capable of engendering all somatic and germ cell types in an adult animal, but lose this ability as development proceeds 5 . However, their transient pluripotency can be permanently captured in embryonic stem cell (ESC) lines. Taken together, these two reprogramming periods restore cellular potency to a pluripotent ground state 5 .
Radical manipulations, such as induced pluripotent stem cell (iPSC) derivation or somatic cell nuclear transfer (SCNT) cloning can artificially reprogram somatic cells 6 . Derivation of iPSCs requires ectopic expression of defined pluripotency factors, resulting in cells that are highly similar to ESCs in morphology, gene expression and germline competence 7 . During SCNT, a somatic cell nucleus is transplanted into an enucleated oocyte where its epigenetic marks are cleared by ill-defined reprogramming factors. In some cases it regains totipotency, i.e. the ability to form all embryonic and extra-embryonic lineages in a viable animal. Compromised development of the NT embryo is associated with aberrant methylation patterns of DNA 8 and histones 9 and dysregulation of gene expression 10 . Several pharmacological chromatin modifiers have been implicated in facilitating epigenetic reprogramming 11 . These drugs induce hyperacetylated, transcriptionally permissive chromatin and can improve iPS colony formation 12 as well as cloning efficiency to term in mouse 11 and pig 13 . Despite these advances, the reprogramming efficiency of somatic cells remains low due to a strong resistance to erase the epigenetic memory of previous lineage decisions and restart embryonic gene transcription 14 .
Overexpressing histone lysine demethylases (KDMs), which reduce transcriptionally activating H3K36 and H3K4 modifications, increase efficiency of iPSC reprogramming 22,23 . Particularly H3K9 marks, which persist through multiple cell divisions 24,25 pose a critical epigenetic barrier in cell reprogramming. We showed that by overexpressing KDM4B, H3K9me3 was demethylated and ESC reprogramming into cloned mouse embryos was improved 26 . A subsequent study confirmed that providing SCNT embryos with exogenous Kdm4d, which only targets H3K9me3, or depleting histone lysine methyltransferases (KMTs) in donor cells markedly increased clone development in vitro and in vivo 27 . By contrast, accumulation of compact heterochromatin domains disrupts ESC self-renewal and alters differentiation potential 15,28 . Reprogramming into iPSCs is also promoted by Kdm3a overexpression 29 and genetic H3K9 KMT depletion 20,21,[29][30][31] , possibly by restricting initial binding and expression of pluripotency factors located in heterochromatic regions 20 . Collectively, this suggest that heterochromatin obstructs re-establishing of pluripotency.
Here we investigated the consequences of Kdm4b overexpression on somatic H3K9me3 and H3K36me3 (H3K9/36me3) levels and gene expression in mouse embryonic fibroblasts (MEFs). We show a strong decrease in H3K9/36me3, which was accompanied by changed expression levels of 15 genes, including up-regulation of the heterochromatin-associated zinc finger protein 37 (Zfp37). Reduced H3K9me3 and H3K36me3 levels strongly increased reprogramming efficiency into cloned embryos and iPSCs, beyond what has been observed with embryonic donors.

Results
Dox-induced expression of Kdm4b-EGFP. MEFs were engineered for doxycycline (Dox)-inducible expression of a functional (F) and an inactive mutated (M) form of Kdm4b, with both transgenes fused to EGFP 26 . Selected F-Kdm4b MEF lines were Dox-induced for two days and expression of the Kdm4b-EGFP transgene analyzed by flow cytometry. Up to 82% (71 ± 2%, n = 24 replicates across six different cell lines) of F-Kdm4b cells displayed EGFP fluorescence (Fig. 1a). Within 24 hours after Dox removal, cells returned to a profile that was indistinguishable from non-induced cells, confirming reversibility of inducible transgene expression (Fig. 1a). Compared to non-induced controls, there was a 97-fold increase in transgene-derived F-Kdm4b expression (Fig. 1b). This did not affect endogenous Kdm4b mRNA levels (Fig. 1c). Global activity for demethylases containing the catalytic JMJC domain increased 2.5-and 2.8-fold in Dox-induced F-Kdm4B MEFs and ESCs, respectively (Fig. 1d). Overall, JMJC family demethylase activity was ~20-fold higher in ESCs than in MEFs (Fig. 1d).

Induced F-KDM4B specifically reduces H3K9/36me3 in MEFs.
To determine the effect on histone methylation, induced and non-induced F-Kdm4b MEFs were analysed for H3K9me1/2/3, H3K36me3 and H3K27me3 levels by semi-quantitative immunofluorescence (Fig. 2a). In F-KDM4B-EGFP positive nuclei, signal intensity for H3K9me3 and H3K36me3 was 10-fold and 4-fold lower, respectively, than in non-induced controls, while H3K9me1 levels were 1.5-fold increased (Fig. 2b). No changes were apparent for H3K9me2 and H3K27me3. To independently validate these results, we performed western blot analyses on bulk histones (Fig. 2c). Histones were extracted from two cell lines with an average induction of 70 ± 6%. Upon induction of F-Kdm4b, H3K9me3 and H3K36me3 levels dropped ~100-fold and ~5-fold, respectively, compared to non-induced controls, accompanied by a ~2-fold increase in H3K9me1 levels (Fig. 2d). H3K9me2 and H3K27me3 levels were not significantly altered.
We also tested whether H3K9 acetylation (H3K9ac) changed upon H3K9/36me3 reduction. Following induction for 48 h, H3K9ac levels were not affected as determined by immunofluorescence (Fig. S1a) and quantification of pixel intensities (Fig. S1b). Likewise, H4K12ac and H4K16ac were not significantly changed after 48 h induction (data not shown).
Gene expression profiling. The global transcriptional impact of Kdm4b-mediated reduction in H3K9/36me3 levels was examined by microarray, mRNA-seq and qPCR. Three independent female F-Kdm4b MEF lines, either induced or non-induced, were used for all gene expression studies. On the microarray, a total of 137,489 probes, representing 24,203 different genes, were detected (Table S1). In total, 599 were significantly altered (Fig. 3b, Table S1). From these, 168 (63 with P < 0.05) genes had a more than 1.5-fold higher and 186 (33 with P < 0.05) a 1.5-fold lower transcript abundance when Kdm4b was induced (Fig. 3a). The maximal difference was a 4.6-fold up-and 2.1-fold down-regulation (Table S1). We also performed microarray analyses of F-Kdm4b ESC clones. In total, 6012 were significantly altered (Table S1). From the same 24,203 genes as on the MEF array, 1467 (540 with P < 0.05) genes had a more than 1.5-fold more and 616 (568 with P < 0.05) a 1.5-fold less transcript when Kdm4b was induced, with a maximal 4.8-and 6.2-fold up-and down-regulation, respectively SCIeNtIFIC RepoRts | 7: 7514 | DOI:10.1038/s41598-017-06569-2 (Table S1). Four array probes were found to match the Kdm4b gene itself. However, none of these covered amino acids 1-424, corresponding to the overexpressed truncated version of Kdm4b. This explains why Kdm4b overexpression was not detected on the arrays. From the same MEF RNA that was used for microarrays, mRNA was isolated for sequencing. Across all six samples, an average of 83 megabases were read with a Q30 > 91. 9 and a mean quality score per base of 35.2. These represented 19,478 different genes. From these, 104 genes were significantly altered (Fig. 3b). Out of the total, 1106 (35 with P < 0.05) genes had a more than 1.5-fold higher and 2273 (66 with P < 0.05) a 1.5-fold lower transcript abundance when Kdm4b was overexpressed (Fig. 3a). The maximal difference was a 26-fold up-and 31-fold down-regulation (Table S1). The Kdm4b gene itself was 15-fold overexpressed (P < 5E-09). Comparing microarray and mRNA-seq results, we found 16,061 genes to be present in both data sets. From these, 15 genes were deregulated in the same direction, using a 1.5-fold change and P < 0.05 as cut-offs (Fig. 3c, Table 1). This overlap was significant (P < 0.05). Taking into account the ESC expression array, there were four additional 1.5-fold up-regulated genes (Ly6a, Ly6c2, Ly6f, Sprr2a3) but only one (Ly6a) was significant between all array (MEF, ESC) and mRNA-seq (MEF) data sets ( Table S1). All of these were confirmed to change their expression in the same direction by qPCR validation, with 12 being up-and 3 being down-regulated (Fig. 3d). This shows a 100% concordance between microarray, mRNA-seq and qPCR data, represented by grey and white bars ± SEM, respectively (n = 3 replicates); Bars with an asterisk (*) differ P < 0.05 between treatments (NI vs I); bars with different letters (a,b) differ P < 0.05 between cell types (MEF vs ESC) of the same treatment by two-tailed unpaired t-tests. which was significant for 40% of candidates (Zfp37, 2810474O19Rik, Gpr56, Slc29a3, Nnat and Lrrn1). Overall, the fold-changes in expression levels of all five genes were remarkably similar between microarray, mRNA-seq and qPCR data (Table 1).
Functional annotation clustering of differently expressed (DE) genes by Gene Ontology (GO) and QIAGEN's Ingenuity ® Pathway Analysis (IPA ® , QIAGEN Redwood City) revealed DE genes as enriched for canonical pathways associated with pluripotency in ESCs, including oncostatin M signaling 32 and transcriptional regulation of pluripotency (Table S3). Cell compromise and maintenance were the main functional categories, with cell cycle as the top associated network (Table S3).
Reduced H3K9/36me3 levels are restored after nuclear transfer. We first adapted our zona-free NT cloning procedure from ESCs to MEFs. MEFs were incubated in pronase prior to lectin-sticking and couplet formation 33 . This pre-treatment tripled mitotic MEF fusion efficiency from 57/306 = 19% (n = 7) to 250/400 = 63% (n = 6), using otherwise standard conditions (P < 0.0001). We then examined how the reduced H3K9/36me3 levels from induced F-Kdm4b donors changed after NT and subsequent in vitro embryo development under non-induced conditions. NT reconstructs from induced and non-induced F-Kdm4b MEFs were fixed at various time points after NT and analyzed by immunofluorescence for the presence of H3K9/36me3 (Fig. 4a). Within ten minutes after fusing MEFs with the cytoplast, the EGFP signal became undetectable in the NT reconstruct and remained so even after formation of a single pseudo-pronucleus. Reduced H3K9/36me3 signals were still observed in NT reconstructs from induced compared to non-induced donors ten minutes and one hour after fusion (Fig. 4b). At eight hours post fusion, both the newly formed pseudo-pronucleus and extruded Shown are the ratios of I over NI relative expression levels, normalized on the geomean of three housekeeping genes. Dotted horizontal line indicates NI level. Error bars: standard error of ratios determined from 3-6 biological repeats per gene. Bars with an asterisk differ between NI vs I at *P < 0.05; **P < 0.005; ***p < 0.001 as determined by two-tailed t-test on mean expression levels.
pseudo-polar body displayed similar H3K9/36me3 intensities in NT reconstructs derived from induced and non-induced donor cells (Fig. 4b). Thus the initially different H3K9/36me3 levels in MEFs were restored relatively rapidly.

Reduced H3K9/36me3 levels improve in vitro development of cloned embryos. To evaluate
whether MEFs with reduced H3K9/36me3 levels would improve donor cell reprogrammability, we determined their in vitro development. Culture was performed without Dox using induced vs non-induced F-Kdm4b cells and additional M-Kdm4b controls. Prior to each NT run, cells were validated for induction of Kdm4b by monitoring EGFP fluorescence. The average induction rate was 76 ± 8% (range 60-88%; n = 3). Reconstructs from induced F-Kdm4b MEFs developed similarly until the morula stage (induced vs. non-induced 9% vs. 7%) but 6-fold better into blastocysts (induced vs. non-induced 18% vs. 3% p < 0.0001) compared to non-induced donors ( Table 2). By contrast, no significant differences were detected in control experiments using induced vs non-induced M-Kdm4b donor MEFs. A small number of F-Kdm4b embryos (N = 33) from different experiments (n = 4) were transferred into four recipients but none of them implanted. Treatment with different histone deacetylase inhibitors (HDACi) can significantly improve developmental efficiency of SCNT embryos 11 . To explore a possible relationship between HDACi and Kdm4b, we simultaneously treated Kdm4d-SCNT embryos with either trichostatin A (TSA) or scriptaid (SCR). These reagents improved SCNT morula and blastocyst rate ~4-15-fold, respectively, over non-induced controls (Table S4). However, neither reagent further increased development beyond the level already achieved by overexpressing Kdm4b (Table S4). Thus combining HDACi with Kdm4b had neither additive nor synergistic effects on reprogramming SCNT preimplantation stages, suggesting that they may exert their effects through a similar pathway. Improved reprogramming into iPSCs. iPSC formation is a complementary cell reprogramming approach to SCNT (Fig. 5a). In both Col1a1 4F2a/WT ('WT/iPS') and Col1a1 4F2a/F-Kdm4b ('Kdm4b/iPS') MEFs, dome-shaped, tightly packed colonies with clear borders appeared around 12 days post-induction and one week following culture in iPSC medium 34,35 . Upon Dox withdrawal after 12 days in culture, both MEF genotypes reverted to a fibroblast morphology and very few colonies remained. When Dox was removed nine days later, the number of stably induced, Dox-independent alkaline phosphatase-and NANOG-positive colonies was significantly higher in the F-Kdm4b-expressing compared to M-Kdm4b and wild-type MEFs (Fig. 5b). Non-induced MEFs of either genotype did not result in colony formation. Following repeated passaging without Dox (Fig. 5c), 4F2a/F-Kdm4b-derived iPSCs were assessed for pluripotency in functional assays. Following injection of three independent iPSC 4F2a/F-Kdm4b lines into SCID mice, large tissue masses, ranging from 10-20 mm in diameter, were harvested after seven weeks from five out of six hind legs (Table S5). Histological examination showed differentiation into ectoderm (squamous epithelium), endoderm (ciliated epithelium) and stromal mesoderm (Fig. 5d). These features were consistent with intramuscular grade 3 teratomas. To fully evaluate pluripotency, two iPSC 4F2a/F-Kdm4b lines were tested for their ability to contribute to chimeras. We obtained four live coat-colour chimeras (Table S6, Fig. 5e). Following repeated test matings, no germ line transmission was obtained from these chimeras.

Discussion
Here we show that H3K9/36me3 poses major obstacles during somatic cell reprogramming and identify potential transcriptional targets of reduced H3K9/36me3 levels. We first generated mice that conditionally overexpress KDM4B. These provide a new animal model to study nuclear reorganization during cell differentiation, Different letters indicate significant differences between MEFs from the same groups (induced or non-induced) at different time points, ***P < 0.0001, **P < 0.01, *P < 0.05 by two-tailed unpaired t-tests.   26 . This supports the notion that KDM4B-mediated demethylation converts H3K9me3 into H3K9me1. Apart from repressive H3K9me3, KDM4B also demethylates H3K36me2/3, which is mainly associated with transcriptional activation 38 . While its demethylase activity for this target is much weaker, decreasing H3K36me2 levels by only 10% 38 , we have found a 4-5 fold decrease in H3K36me3 levels. How would reduced H3K9/36me3 make the donor genome easier to reprogram? H3K36me3 is present at transcribed genes, peaking toward the middle and 3′ ends of coding regions but mostly missing from promoters 39 . Interestingly, trimethylation of active H3K36 and repressive H3K27 do not coexist on the same H3 tail 39 . This broad mutually exclusive distribution of H3K36me3 over active euchromatin may prevent spreading and accumulation of silencing marks. Indeed, removal of H3K36me1/2/3 writers results in global loss of H3K36me3, redistributing H3K27me2/3 from its endogenous sites to active gene bodies and mis-regulating gene expression 39 . We may not have detected subtle changes in H3K27me3 abundance or staining pattern. For this, it might be worthwhile to interrogate regions enriched in H3K36me3 and H3K27me3, such as those involved in maintaining X-linked gene expression and X chromosome inactivation in females 40 , in particular when they have a negative impact on NT reprogramming 41 .
H3K9me3, on the other hand, is a repressive modification. It is recognized and bound by the heterochromatin protein HP1 42, 43 , causing heterochromatin formation 44 and epigenetic silencing 45 . Therefore, H3K9me3-initiated heterochromatinization can restrict access to chromatin-binding factors and prevent activation of developmentally important genes. In addition, H3K9me3 may inhibit local deposition of activation marks, such as H3K9 acetylation, which we found globally unchanged. These changes in donor cell epigenotype altered transcription for 15 genes. Given that H3K9me3 generally marks gene-poor regions and accumulates over repetitive elements, such a relatively low number of differentially expressed genes is not surprising. Most genes were up-, not down-regulated, consistent with expectations from the published role of H3K9me3 as a transcriptional repressor. Whilst some megabase-scale H3K9me3 regions contain pluripotency genes (e.g. Nanog, Sox2, Prdm14), these candidates were not significantly effected in our screen. Among the changed transcripts, the only direct linkage to heterochromatin was Zfp37. The corresponding zinc finger protein is expressed in brain and testis, where it specifically associates with the heterochromatin adjoined to nucleoli 46 . It is not known what regulatory role ZFP37 plays in these tissues. Only one gene, Ly6a (or Sca-1), was up-regulated in both ESCs and MEFs. This gene is part of a large family of mostly GPI-linked cell-surface glycoproteins in lymphocytes. Monoclonal antisera against Ly6a are commonly used to identify hematopoietic stem cells but is also found on other leukocyte subsets 47 . The in vivo function of this protein remains largely unclear 47 . Overexpressing some of these candidate genes in donor cells or embryos may shed light on their function and directly improve cloned blastocyst formation.
Following NT, all donor transcripts and proteins will be transferred and ~1000-fold diluted in the cytosol of the recipient oocyte, where they could influence early embryonic gene expression. Numerous studies have shown that even though the donor cell genome is largely repressed by the late one-cell stage, some highly expressed donor transcripts remain detectable at both early and late mouse preimplantation stages 27,48,49 . It would be informative to measure the extent of this donor-specific gene and protein "memory" in Kdm4b-derived SCNT embryos.
In addition to the protein-coding and unannotated transcripts, H3K9me3-marked regions harbour repetitive sequences, which were also represented in our mRNA-seq data set. In ESCs, most H3K9me3 chromatin is targeted to repeat-rich intergenic and intronic regions and absent from coding sequences or regulatory elements, such as enhancers or promoters. The major repeat target regions for H3K9me3 outside constitutive heterochromatin are full-length (>6 kb) ERVs and LINEs 50 . This subpopulation of intact retrotransposons accumulates highest levels of H3K9me3, whereas the vast majority of degenerate and truncated ERVs and LINEs fail to be enriched for this mark. Following loss of H3K9me3 accumulation, the most significantly deregulated targets were major satellites, ERVs and intact LINEs, especially of the L1Md subfamily 50 . However, ERV and LINE-specific H3K9me3 enrichment and transcriptional dysregulation was not observed in immortalized MEFs 50 . Instead, genome-wide coverage of H3K9me3 in immortalized MEFs mainly resides in megabase domains that are not enriched for LINEs, ERVs, or other repeats but rather contain genes 20 . We observed increased transcriptional output from all full-length ERV (>6 kb) repeat regions. This may contribute to the better reprogramming of Kdm4b-expressing donor cells because activation of repeat sequences is important for preimplantation development 51,52 . Our observation is consistent with the finding that Kdm4d partially relieves repression of murine ERV retrotransposons in 2-cell SCNT embryos 27 .
Globally reduced H3K9/36me3 levels in induced donor cells initially persisted in reconstructed embryos but returned to non-induced control levels within eight hours of NT. This is different from the prompt restoration of H3K9me3 marks previously observed for ESCs, which occurred within minutes of NT 26 . The difference is consistent with the greater H3K9me3 depletion in MEFs relative to ESCs, which may take longer to reprogram by histone methyltransferases active in the recipient cytoplasm. Indeed, RNA-seq and proteomics of metaphase-arrested oocytes has detected all known H3K9me3-specific KMT mRNAs and KMT1E protein, respectively 53 . This fast-acting cytosolic machinery would normally reprogram the incoming mouse sperm upon fertilisation, restoring H3K9 methylation in unmethylated male pronuclei within three hours of entry into the oocyte 54 . The rapid gain of methylation observed in MEF nuclei suggests fast enzymatic re-methylation of H3K9 sites rather than slow incorporation of already methylated histones. Concomitant with normalizing H3K9me3, there was a complete loss of KDM4B-EGFP signal in NT reconstructs, probably caused by rapid dissociation from the prematurely condensing donor chromatin and dilution in the oocyte cytoplasm 55 .
The apparent re-instatement of H3K9me3 levels was assessed by IF, which only provides a semi-quantitative read-out of global methylation levels. It is possible that some loci remained hypomethylated, for example the previously described reprogramming-resistant genomic regions in MEFs 27 . These heterochromatin regions are enriched for KMT1A/B-deposited H3K9me3 and low DNase I accessibility. Continued activation of locally H3K9me3-depleted regions may persist for some time, making the genes in these regions worthwhile candidates for further investigation of embryonic genome activation in Kdm4b-induced vs non-induced SCNT embryos. Thus even a very short-lived decrease in H3K9me3 levels could facilitate initial binding of oocyte reprogramming factors, triggering a ripple effect that persists after the global differences in H3K9me3 have disappeared.
In spite of their rapid apparent restoration, reduced H3K9me3 donor levels correlated with an almost tripled in vitro development of NT reconstructs. This is a ~5-fold greater increase than previously observed with ESCs using the same approach 26 , suggesting that a greater reduction in H3K9me3 levels is beneficial for NT-mediated reprogramming into blastocysts.
H3K9me3-mediated heterochromatin formation also acted as a reprogramming barrier for generating bona fide iPSCs. This is supported by several studies using both gain-of-function 30 and loss-of-function 20,21,30 approaches to reduce H3K9me3 to promote conversion of MEFs into iPSCs. The H3K9me3 barrier in mouse iPS reprogramming can be lowered by knocking down various H3 KMTs, primarily KMT1 variants 20,21,30 . Since iPSC generation is relatively slow compared to NT, extending over days and weeks, the immediate downstream gene networks necessary for reacquisition of pluripotency, are likely different. Accordingly, the only indispensable reprogramming factors for iPSC generation, Oct4/Pou5f1, has been shown to be dispensable in SCNT reprogramming 56 .
While we were able to generate cloned blastocysts with Kdm4b-EGFP MEFs, we did not obtain live offspring. Hence we cannot conclude that induction of Kdm4b in donor cells improved in vivo development of cloned embryos. Considering the substantial technical difficulties of cloning live mice, achieved by only a very small number of groups, our lack of success may be attributable to inherent technical difficulties. In another study, implantation and development to term was significantly higher in Kdm4d-injected than in control SCNT embryos 27 .
In summary, we demonstrate improved in vitro reprogramming into cloned embryos and iPSCs following Kdm4b overexpression in MEFs. It is plausible that targeted reduction of repressive H3K9me3 marks led to a derestricted genome with greater reprogrammability. Targeting other histone modification, such as H3K36me1/2 via KDM2 overexpression 22,23 and H3K79me via KMT4 knockout 57 , could further improve reprogramming efficiency. It remains to be seen if this cell-based epigenetic therapy is applicable to other mammalian species, such as, livestock or humans. In cattle, for example, serum starving donor cells reduces their histone methylation levels, which increased reprogramming into totipotency 58 . This provides a mandate to develop specific small molecule inhibitors (e.g. for KMT1) to improve epigenetic reprogramming without genetic intervention.

Materials and Methods
Transgenic Kdm4b mice. All methods were carried out in accordance with the New Zealand (NZ) Animal Welfare Act 1999 and approved by the Ruakura Animal Ethics Committee. To make germline chimeras, we used ESCs (C57/BL6 × 129SV) engineered for Doxycycline (Dox)-inducible expression of a functional (F) and an inactive mutated (M) form of Kdm4b, with both transgenes fused to EGFP 26 . Each transgene was integrated at the collagen 1 alpha 1 (Col1a1) locus and controlled by a tetO promoter, driven by the M2 reverse tetracycline transactivator (M2-rtTA) in the Rosa26 locus 26 . Chimeras and offspring were PCR-screened for rtTA and Kdm4b-EGFP transgenes and absence of the wild-type loci (Table S7).
Quantitative PCR. Target sequences (Table S6) were amplified using the Syber Premix ExTaq (Takara) with a Rotorgene 6000 (Corbett) or LightCycler ® 2.0 (Roche). Comparative quantification in the Rotorgene Software was used to determine amplification efficiency of each reaction 59 . Transcript levels were determined relative to the geometric mean of four housekeeping genes (Gapdh, Actb, Hprt, Gusb) while normalizing for amplification efficiency 60 . Reverse transcriptase was omitted in one sample for each cDNA synthesis reaction ("-RT" control). Overexpression plasmids served as positive controls and water as negative control ("no template"). Each sample was run in technical triplicates. Product identity was confirmed by melting curve analysis.
Microarray and mRNA-seq sample preparation. One F-Kdm4b ESC clone 26 , in biological triplicates, and three female F-Kdm4b MEF lines were used for global transcript profiling. A sub-sample from each line showed >70% induction by flow cytometry. RNA was extracted using an RNeasy Kit (Qiagen). Quality, concentration and integrity were verified by NanoDrop ® ND1000 and inspection on formaldehyde-denaturing gels (28 S/18 S rRNA ratio ~2). The subsequent workflow (cDNA processing, Cy3-labelling, hybridization, washing, scanning) was carried out by Nimblegen (Roche, NZ). Briefly, cDNA was hybridized to a Nimblegen Mouse Gene Expression 12 × 135 k Array, containing 44,170 target genes with 3 × 60 mer replicate probe sets per target, based on the MM9 genome build. For mRNA-seq, the Poly(A) Purist ™ MAG kit (Ambion) was used to extract mRNA from total RNA of the same six MEF samples submitted for microarray analysis. The quality was assessed using the Agilent 2100 Bioanalyzer. After submission of 250 ng of high-quality mRNA, the complete workflow (library preparation, quality control, sequencing) was carried out by NZ Genomics Ltd. (Dunedin, NZ). Libraries were generated using a TruSeq RNA Sample Prep Kit v2 (Illumina) and validated on a Bioanalyzer with a High Sensitivity DNA kit. Sequencing used the paired-end 2 × 100 protocol on an Illumina HiSeq. 2000.
Microarray data analysis. Raw data were extracted using Nimblegen's DEVA software. For each probe set, there was one perfect match (PM) and one mismatch (MM) probe, designed to have 100% identity and one mismatch, respectively, to the target sequence. With all MM probes being zero or near-zero, all PM probes were declared valid. Multi-array average normalization 61 and background subtraction was performed using DEVA. Normalized data were re-loaded into R and examined on boxplots, revealing no outliers (outside 3 standard deviations of the mean). Contrasts for each probe and gene were done using the "limma" package in R. mRNA-seq data analysis. More than 90% of the bases had a Phred quality (Q) score >30, i.e. a base call accuracy of 99.9%. Flexbar software was used to remove low quality regions (<Q20) and sequencing adapters from the reads. This removed reads <18 bp after quality trimming and only kept properly paired reads. Each sample was mapped to the mouse reference genome (Mus musculus GRCm38 with NCBI annotations) using STAR version 2.5.0, resulting in ~62-85 Million (≈90%) mapped paired-end reads. The ~58-79 Million (≈85%) uniquely mapped paired-end reads for un-stranded mRNA-seq for each sample were compiled into a tab-delimited text file. Counts for each gene were analysed using the "edgeR" (version 3.10.5) package in R (version 3.2.1) 62 . Fold-changes between induced vs non-induced samples were calculated using a negative binomial model. Exact tests were used to calculate p-values, which were corrected for multiple testing using FDR-corrected q-values 63 . Fold-changes, p-values and FDR values were imported into Excel for further analysis. Significant upand down-regulated genes were called using P < 0.05 and ≥1.5 fold-change as cut-offs.
To identify significantly deregulated repeat types, sequences of repetitive elements were downloaded from the Repeat Masker (RMSK) database (open-3-3-0 version of RMSK, RMSK library release 20120418). After sorting and converting the STAR (version STAR_2.5.0b) output with samtools v0.1.19-44428cd 64 , reads that mapped to repetitive elements were extracted using the feature Counts Version 1.5.0-p3 65 using the "-p" option to count fragments instead of reads. Counts from individual units of the same repeat type were combined and total counts for each repeat type were calculated. Repeats with less than 10 fragments over all the samples were removed. Differentially expressed repeats were determined in edgeR. Count numbers for all repeats were used as input to estimate the effective library size and normalised using the trimmed mean of M method in edgeR. Fold-changes, p-values and FDR values for induced vs non-induced samples were calculated as above and called as significantly different at P < 0.05.

Demethylase activity detection.
A fluorometric assay (Cayman Chemicals, No. 700390) was analyzed on a multi-mode plate reader (Synergy 2, Biotek, USA) using an excitation/emission wavelength of 360/460 nm, respectively. Human recombinant KDM4B and H3K9me3 peptides provided positive controls.
Immunofluorescence. Cells were simultaneously fixed and permeabilized in 4% (w/v) paraformaldehyde (PFA)/1% (w/v) Triton X-100 in PBS, blocked with 2.5% (w/v) bovine serum albumin in PBS and incubated overnight at 4-8 °C with primary antibodies 26 (Table S9). Cells were washed in PBS-0.05% Tween ® and incubated for 1 h at room temperature with secondary antibodies. Cells were visualized with epifluorescence (AX-70 Olympus microscope). Images were processed using a Spot RT-KE slider CCD camera and software (Diagnostics Instruments Inc.). NT reconstructs were first fixed in 4% PFA and then permeabilized in 0.1% Triton X-100. Negative controls were processed the same way with blocking buffer instead of primary antibodies. For each reconstruct, the Hoechst-stained region of interest (ROI) was outlined in ImageJ. Mean grey value intensity, measured at 3 random cytoplasmic locations, was subtracted from the mean ROI intensity. This background-corrected mean intensity represents the sum of grey values of all pixels in the ROI divided by the number of pixels, and is referred to as normalized pixel intensity. Following quantification, images were pseudo-coloured using the cyan, green and red lookup tables.
Western blot. Histones were extracted using the EpiQuik TM Total histone extraction kit (Epigentek, Cat-OP-0006-100). Histone extracts (10-15 µg per lane) were separated on a 15% SDS PAGE gel, transferred onto a nitrocellulose membrane and probed with primary antibodies (Table S8). Following incubation with a secondary antibody, the modified histones were visualised with enhanced chemiluminescence. Signal intensity was normalized for the histone H1 signal (Ponceau S stain) and quantified using Quantity One software (Bio-Rad Laboratories Inc.).
Nuclear transfer. NT donor cells were metaphase-arrested with 1.65 μM nocodazole for 2-3 hours, before shaking them off the plate 26 . Oocytes were collected from 10-12 weeks old B6C3 mice and the cumulus cells and zona pellucida removed. Zona-free NT was carried out as described 26 but with two modifications. First, zona-free oocytes were enucleated at 30 °C under polarized light (CRi Oosight Imaging System). Second, shake-off MEFs were treated with 1 mg/ml pronase (in Hepes-buffered CZB medium) for 3 minutes prior to NT to improve fusion efficiency. NT reconstructs were kept in nocodazole until fusion had occurred. At 1 h post-fusion, reconstructs were activated with 10 mM SrCl 2 in Ca-free M16. Only reconstructs with one polar body after 6 h were cultured in M16 drops at 37 °C under 5% CO 2 in air. On day 4, compacted morulae and blastocysts were either processed for molecular analyses or transferred into the uterus of day 0.5 pseudo-pregnant females 26 . For some experiments, cells or embryos were treated with TSA or SCR at 50 nM and 250 nM, respectively, or respective DMSO dilutions (1:200) as controls. For NT, TSA and SCR were provided during activation (6 h) and early culture (3 h for TSA, 12 h for SCR). iPSC generation. Homozygous iPSC mice were imported from the Jackson laboratory (Stock no. 011004). This R26 rtTA strain expresses a Dox-inducible polycistronic four iPS factor (4 F) cassette, linked by self-cleaving 2 A peptides, from the Col1a1 locus (Col1a1 4F2a ). Mice were mated with either homozygous R26 rtTA , Col1a1 F-Kdm4b or R26 rtTA , Col1a1 WT controls. On day 12.5, MEFs were derived from both crosses. In addition to two R26 rtTA alleles, these MEFs carry one copy of the Col1a1 4F2a transgene and either one copy of Col1a1 F-Kdm4b or one copy of Col1a1 WT . Secondary iPSCs were derived from both MEF genotypes by culture in 2 µg/ml Dox. MEFs (passage 1-4) were seeded at ~1 × 10 4 cells/cm 2 in DMEM/F12 with Glutamax TM -I and 10% FBS. After three days, cells were passaged onto 0.1% gelatine-coated tissue culture dishes. After two days, cells were shifted into iPSC medium, comprising of PD0325901 (0.04 μM), GSK3B inhibitor CHIR99021 (3 μM), and recombinant human LIF (20 ng/ml) in DMEM/F12 supplemented with N2 and mixed 1:1 with Neurobasal medium supplemented with B27 and 1 mM L-glutamine ('N2B27') 66 . Culture medium was changed every 2-3 days. On D21, Dox was removed from the medium and the number of alkaline phosphatase-and NANOG-positive colonies counted on D28, i.e. one week after Dox removal. For detecting alkaline phosphatase activity, cells were fixed with 4% PFA and stained with NBT/BCIP reagent for 20 min. Dox-independent colonies were dissociated using 0.05% TrypLE ™ and passaged every three days.

Statistical analysis.
Values are the average of several replicates (n) ± SEM, unless noted otherwise.
Significance of differences for western analysis was determined via two-tailed paired t-tests on normalised band intensities. For the quantification of fluorescent signals and qPCR results, significance was determined by two-tailed t-tests. For comparing in vitro development, significance was determined using the two-tailed Fisher exact test for independence in 2 × 2 tables.