Abstract
Cells undergo a major epigenome reconfiguration when reprogrammed to human induced pluripotent stem cells (hiPS cells). However, the epigenomes of hiPS cells and human embryonic stem (hES) cells differ significantly, which affects hiPS cell function1,2,3,4,5,6,7,8. These differences include epigenetic memory and aberrations that emerge during reprogramming, for which the mechanisms remain unknown. Here we characterized the persistence and emergence of these epigenetic differences by performing genome-wide DNA methylation profiling throughout primed and naive reprogramming of human somatic cells to hiPS cells. We found that reprogramming-induced epigenetic aberrations emerge midway through primed reprogramming, whereas DNA demethylation begins early in naive reprogramming. Using this knowledge, we developed a transient-naive-treatment (TNT) reprogramming strategy that emulates the embryonic epigenetic reset. We show that the epigenetic memory in hiPS cells is concentrated in cell of origin-dependent repressive chromatin marked by H3K9me3, lamin-B1 and aberrant CpH methylation. TNT reprogramming reconfigures these domains to a hES cell-like state and does not disrupt genomic imprinting. Using an isogenic system, we demonstrate that TNT reprogramming can correct the transposable element overexpression and differential gene expression seen in conventional hiPS cells, and that TNT-reprogrammed hiPS and hES cells show similar differentiation efficiencies. Moreover, TNT reprogramming enhances the differentiation of hiPS cells derived from multiple cell types. Thus, TNT reprogramming corrects epigenetic memory and aberrations, producing hiPS cells that are molecularly and functionally more similar to hES cells than conventional hiPS cells. We foresee TNT reprogramming becoming a new standard for biomedical and therapeutic applications and providing a novel system for studying epigenetic memory.
Main
Somatic cell reprogramming requires substantial epigenome remodelling to establish states resembling hES cells. The generation of hiPS cells by the ectopic expression of the transcription factors OCT4, KLF4, SOX2 and MYC (hereafter referred to collectively as OKSM) is the most widely used method9. Despite the high similarity of induced pluripotent stem (iPS) cells and embryonic stem (ES) cells10,11, substantial evidence indicates that iPS cells are epigenetically and functionally distinct from ES cells, including residual somatic cell epigenetic memory and de novo epigenetic aberrations1,2,3,4,5,6,7,8. Previous reports have shown that DNA methylation and histone modifications encode these epigenetic differences, which are transmissible through differentiation1,2,3,4, limiting the potential use of hiPS cells in disease modelling, drug screening and cell therapies12. However, the mechanisms underpinning how aberrant epigenetic states emerge during reprogramming remain unknown.
The observation that cells reprogrammed by somatic cell nuclear transfer (SCNT) retain less epigenetic memory than OKSM-reprogrammed cells13 indicates that epigenetic aberrations are not inherent to reprogramming and can be mitigated. Although the exact mechanisms are unknown, SCNT reprogramming appears to recapitulate the pre-implantation epigenome reset, mediated by the molecular environment within oocytes. Notably, although SCNT stem cells contain less epigenetic memory than hiPS cells13, SCNT reprogramming requires donor oocytes, rendering the method inefficient, complex and unscalable.
Conventional OKSM reprogramming produces hiPS cells in a primed pluripotent state (primed-hiPS cells) resembling post-implantation epiblast cells14,15. Recent developments enable the reprogramming of somatic cells to a naive pluripotent state (naive-hiPS cells) resembling the pre-implantation epiblast, including low global DNA methylation16,17,18. These two reprogramming paradigms provide tractable model systems to study how epigenome resetting is influenced by environments resembling distinct developmental states of pluripotency. Previous studies have focused on changes in DNA methylation when hES cells are switched between primed and naive culture conditions19,20,21, but it is not known whether epigenetic memory and aberrations occur in naive-hiPS cell reprogramming. We therefore set out to study the origins, dynamics and mechanisms of epigenetic abnormalities in naive and primed reprogramming to comprehensively understand the reprogramming process.
Divergent epigenome remodelling in hiPS cells
To investigate epigenome remodelling throughout naive and primed reprogramming, we reprogrammed human fibroblasts into both primed and naive pluripotent states using Sendai viral OKSM transcription factors16, and isolated reprogramming intermediates throughout this process using intermediate cell surface markers22 (Fig. 1a, Extended Data Fig. 1a,b and Supplementary Table 1). We then profiled DNA methylation using whole-genome bisulfite sequencing (WGBS) and analysed gene expression data previously generated by RNA sequencing (RNA-seq) from the same cells22 (Fig. 1a). This enabled base-resolution quantification of the methylome throughout reprogramming. The largest changes in CG DNA methylation during primed reprogramming occur between days 13 and 21, with global levels reaching those similar to hES cells by passage 3 (Fig. 1b and Extended Data Fig. 1c). By contrast, most CG methylation changes in naive reprogramming occur before day 13 (Fig. 1b). As expected, naive conditions result in partial methylation at most CG dinucleotides (Extended Data Fig. 1c). Furthermore, intermediate levels of CG methylation in naive conditions is a result of sparse distribution of methylated CGs on individual DNA fragments, demonstrating that intermediate methylation is not caused by cell heterogeneity (Extended Data Fig. 1d).
CpH methylation (where H represents A, C or T) is a hallmark of pluripotent stem cells, and is mostly attributable to CA methylation (Extended Data Fig. 1e). We found that global CA methylation increases within the first 5 days of naive culture conditions, but after day 13 in primed reprogramming (Fig. 1c). Notably, we observed that CH methylation only accumulates upon changing cells to naive or primed culture conditions, concomitant with increased DNMT3B expression (Fig. 1c and Extended Data Fig. 1e,f).
Inspection of CG DNA methylation changes at regulatory elements revealed stepwise changes during primed reprogramming, but only one major change during naive reprogramming between days 7 and 13 (Fig. 1d). Fuzzy clustering identified five distinct classes of dynamic methylation at regulatory elements (Fig. 1e and Supplementary Table 2), with methylation changes generally occurring after, and being inversely correlated with, the expression change of linked genes (Fig. 1e and Extended Data Fig. 1g,h). This suggests that methylation changes at regulatory elements do not drive expression change during reprogramming but maintain repression, similar to reprogramming in mouse cells23.
We then identified the transcription factor motifs associated with methylation changes at regulatory elements (Fig. 1f). Elements with increasing methylation during reprogramming (clusters 1–3) were enriched for the AP-1, JUN and FOS motifs, as was the transient cluster (cluster 5), which was also enriched for OCT4–SOX2 motifs (Fig. 1f). This is consistent with human and mouse studies suggesting that transcription factors at somatic enhancers are sequestered to transiently active elements bound by OKSM, which recruits transcription factors away from the loci maintaining somatic cell identity22,23. Demethylated regulatory elements featured OCT4–SOX2 motifs, and were associated with pluripotency genes, where expression increased after day 3 (cluster 4; Fig. 1e,f). Inspection of methylation changes driven by OKSM in fibroblast medium (up to day 7) revealed that 1,030 enhancers but only 39 promoters feature CG methylation loss of more than 20%, with these enhancers being enriched for AP-1 and pluripotency transcription factor motifs (Extended Data Fig. 1i). These time-course methylome profiles reveal that the first wave of epigenetic remodelling at regulatory elements is driven by OKSM, followed by distinct methylation states coincident with transitioning to primed and naive culture conditions.
Emergence of aberrant DNA methylation
Several reports indicate that hiPS cells feature differentially methylated regions (DMRs) compared with hES cells that can be categorized as either somatic cell epigenetic memory or acquired aberrant methylation states that are unique to hiPS cells, which are not present in the cell of origin or hES cells1,2,3,4,5,7,13,24. Despite reports of DNA methylation differences between hiPS cells and hES cells, their temporal dynamics during reprogramming are not well characterized. We thus first identified CG-DMRs between multiple primed-hiPS cell and hES cell lines (Extended Data Fig. 1j). We identified 2,727 CG-DMRs (methylated CG (mCG)/CG difference >0.2; P ≤ 0.05), with 86.5% showing lower CG methylation levels in hiPS cells (Fig. 2a, Extended Data Fig. 1k and Supplementary Table 3). CG-DMRs could be classified as acquiring aberrant DNA methylation or retaining somatic cell epigenetic memory by comparing the DNA methylation levels between primed-hiPS cells and the fibroblasts that they originated from (Fig. 2b). This revealed that in primed-hiPS cells, 60.4% of the CG-DMRs were hypo-methylated relative to hES cells and showed less than 20% difference in methylation levels relative to fibroblasts, indicating somatic cell epigenetic memory, and an additional 24.2% of the CG-DMRs that were hypo-methylated relative to hES cells harboured higher methylation in primed-hiPS cells relative to fibroblasts, indicating partial epigenetic memory (Fig. 2b). Conversely, a majority of hyper-methylated CG-DMRs (54.2%) exhibited aberrant DNA methylation acquired during reprogramming, with methylation levels more than 20% higher than both fibroblasts and hES cells (Fig. 2b). Time-course analysis revealed that aberrant methylation begins to emerge between days 13 and 21 of primed reprogramming and continues to increase between day 21 and passages 3–10 (Fig. 2c). With memory CG-DMRs, minor transient demethylation (mCG/CG < 0.1) occurred in primed reprogramming (Fig. 2d), concordant with global CG methylation change (Fig. 1b). However, transitioning cells to naive medium triggered substantial demethylation in memory CG-DMRs by day 13 (Fig. 2d,e and Extended Data Fig. 1l,m). For hyper-methylated memory CG-DMRs, we observed demethylation to levels similar to those in hES cells by day 13 (Extended Data Fig. 1l). Overall, we found that aberrant CG methylation does not begin to accumulate upon OKSM induction during early reprogramming, and begins to emerge only after day 13 of primed reprogramming (Fig. 2c). Of note, aberrant CG hyper-methylation loci in primed-hiPS cells were not aberrant in naive reprogramming (Fig. 2c), indicating that aberrant hyper-methylation is a feature of primed and not naive reprogramming.
TNT reprogramming resets the epigenome
During early development, the pre-implantation embryo undergoes an epigenetic reset involving a wave of global demethylation, during which genomic imprints are protected from demethylation27. By combining our new understanding of epigenomic reconfiguration during reprogramming, we hypothesized that we could avoid somatic cell epigenetic memory and aberrant DNA methylation by reprogramming through a transient naive-like state, similar to the demethylation observed during embryonic development. Thus, we devised two experimental systems. In the first system, we reprogrammed fibroblasts with a transient naive culture treatment for 5 days after the initial 7 days of culturing in fibroblast medium, followed by culturing in primed medium for the remainder of the reprogramming (Fig. 3a), to give rise to transient-naive-treatment hiPS cells (TNT-hiPS cells). In the second system, we first established naive-hiPS cell colonies by extended naive culturing and then transitioned the cells to a primed pluripotent state to give rise to naive-to-primed hiPS cells (NTP-hiPS cells) (Fig. 3a)…