Targeted removal of epigenetic barriers during transcriptional reprogramming

Master transcription factors have the ability to direct and reverse cellular identities, and consequently their genes must be subject to particular transcriptional control. However, it is unclear which molecular processes are responsible for impeding their activation and safeguarding cellular identities. Here we show that the targeting of dCas9-VP64 to the promoter of the master transcription factor Sox1 results in strong transcript and protein up-regulation in neural progenitor cells (NPCs). This gene activation restores lost neuronal differentiation potential, which substantiates the role of Sox1 as a master transcription factor. However, despite efficient transactivator binding, major proportions of progenitor cells are unresponsive to the transactivating stimulus. By combining the transactivation domain with epigenome editing we find that among a series of euchromatic processes, the removal of DNA methylation (by dCas9-Tet1) has the highest potential to increase the proportion of cells activating foreign master transcription factors and thus breaking down cell identity barriers.

C ellular features, including identity, behavior, and potency, are initiated during development by transcriptional programs, which are triggered by individual master transcription factors 1 . These proteins are temporally and spatially highly restricted in their expression. The erroneous expression of developmental master transcription factors is a permanent threat to the integrity of any multicellular organism, as they are able to impose their transcriptional programs outside of their natural context. This issue has been exemplified by the so-called reprogramming factors, which are master transcription factors able to overwrite already established cellular programs by enforcing foreign cell identities 2 . Because cell identity switches rarely (if at all) occur under normal circumstances, it is likely that special measures restrict the activation of master transcription factors in somatic cells.
Although it is obvious that the most effective barriers against cell identity changes would control the expression of master transcription factor genes and that several chromatin processes (e.g., DNA methylation, polycomb, nuclear topology) have been implicated in this process, causalities were extremely difficult to establish. This is due to the fact that until recently the lack of experimental options for site-specific manipulation made it challenging to generate unambiguous experimental evidence for the causality of local chromatin features. But even with recent advancements in the field of DNA targeting, most experimental reprogramming approaches still rely on vector-mediated overexpression of fate determinants, bypassing the primary activation of the endogenous reprogramming factor gene. Thus, the potentially most relevant cell identity barriers have evaded identification and functional analysis.
However, two experimental strategies, both based on adaptations of the bacterial CRISPR complex, have recently enabled the circumvention of these two issues and can be used to investigate the presence and identity of such functional barriers: transcriptional engineering, which is the manipulation of gene expression by targeting artificial transcription factors to gene promoters [3][4][5] , as well as epigenome editing, which is the site-specific manipulation of epigenomic features, for which a number of options have been developed recently 6,7 . Gene induction by transcriptional engineering aims for the targeted manipulation of gene expression of endogenous genes 8 . While the epigenomic and/or transcriptional effects are defined by the specific dCas9 fusion protein used, the target locus is specified in the protospacer sequence encoded in the short synthetic guide RNAs (gRNAs). The system most frequently used to date to achieve gene activation is a fusion of an enzymatically dead version of Cas9 with four copies of the transactivator domain of the viral transcription factor VP16 (dCas9-VP64) 9 . Transcriptional engineering has been successfully used for transcription activation of a number of genes, including master transcription factors like MyoD, Ascl1, and Sox2 5, 10,11 . Although examples remain sparse and we lack comprehensive approaches, it emerges already that endogenous master transcription factors behave in a particularly rigid manner compared to control genes 5,12,13 .
During neurogenesis, a temporal sequence of master transcription factors controls the progression of neural development. The sex-determining region Y-box transcription factor 1 (Sox1) is amongst the first transcription factors induced during neural development in xenopus 14 , chicken 15 , rodents 16 , and humans 17 . In mice, Sox1 shows strong and specific expression in neuroepithelial cells 18,19 , the progenitors of all neural cells. Sox1 has some sequence identity (51%) to Sox2, but unlike its paralog it is not expressed in pluripotent cells and only becomes detectable in the newly formed neuroectoderm around the onset of somitogenesis 20 . Shortly thereafter, Sox1 expression disappears, except in the adult neural stem cell niches of the hippocampus and the adult subventricular zone, where it marks a population of progenitor cells with long-term neurogenic potential 21 . Interestingly, Sox1-positive NSCs can be propagated only poorly in vitro, as cultured cells irreversibly lose Sox1 expression 22 . This conversion coincides with a progressive loss of neuronal differentiation potential and parallels the natural development in vivo. Despite its undeniable relevance as an early lineage marker, the functional roles of Sox1 are, compared to its paralogs Sox2, Sox9, Sox10 and Sry, still poorly understood 23 .
Here we show that Sox1 is a master cell identity factor: the induction of its endogenous gene copy restores the neuronal differentiation potential of glial progenitor cells. Moreover, chromatin features, in particular DNA methylation at its promoter, tightly control the capability of Sox1 to be trans-activated. By combining epigenome editing and transcriptional engineering, we demonstrate that the selective removal of this barrier increases the number of responsive cells significantly, proving the causal role of the chromatin mark.

Results
Targeted activation of Sox1 leads to heterogenous response. Neural progenitor cells (NPCs) do not express the neural stem cell factor Sox1. First, we tested whether transcriptional engineering can be used to significantly activate this early lineage marker in NPCs. For this, we generated clonal NPC lines stably expressing the transcriptional trans-activator dCas9-VP64 that can be targeted to specific genomic loci through simultaneous delivery of gRNAs. The cells continued to produce mostly glial progeny when differentiated (see below). To test the capacity of these cells for targeted gene activation, we used an expression construct containing two gRNAs (A1-9). Those were designed to target with high predicted specificity the promoter of Actc1, encoding an actin gene expressed in heart and skeletal muscle tissue (see Methods and Breunig et al., 2018 24 , Fig. 1a). We analyzed the consequences of transcriptional targeting by qPCR, which confirmed strong induction of Actc1 (more than 100-fold, approximating physiological levels of muscle tissues) (Fig. 1b,  Supplementary Fig. 1a). To induce Sox1 expression, we used an equivalent construct targeting the Sox1 promoter (S1-9, Fig. 1a). In contrast to Actc1, the transcriptional activation of Sox1 was significantly lower (ca. four-fold, Fig. 1b). These results indicated some insufficiency of the transcriptional engineering approach when targeting the developmental transcription factor Sox1.
To rule out differences in transfection efficiency as the reason for the disparity in transcriptional outcomes, we replaced the plasmids for gRNA expression with lentiviral particles. Antibiotic selection was applied to ensure that every cell expressed gRNAs targeting the Sox1 promoter. Moreover, to rule out that the specific choice of gRNAs is the source of the insufficient Sox1 induction, we generated seven different lentiviral constructs, each targeting a different site in the Sox1 promoter (SoxProm, Fig. 1a) and applied a mixture of viral particles with a high titer (MOI 4). This did not significantly potentiate the transcriptional level of Sox1 in transduced and selected cells (Fig. 1c), and neither did lentiviral vectors containing two alternative Sox1 targeting gRNAs (S4-7, Fig. 1a, c), indicating that the individual choice of gRNA sequences, their delivery or selection are likely not responsible for the scarce response.
To test whether the limited induction originates from a uniform but very minor gene activation, or a heterogeneous cellular response with few cells strongly activating Sox1, we generated NPCs from ESCs containing a fluorescent reporter (gift from Prof. Austin Smith). Here, GFP has been knocked in-frame into the endogenous copy of the transcription factor Sox1, leaving the promoter unchanged 18 . NPCs heterozygous for the GFP reporter (Sox1 wt /Sox1 GFP ) and stably expressing dCas9-VP64 were used henceforth to analyze transcriptional induction by flow cytometry or to separate GFP-positive and -negative cells by fluorescence activated cell sorting (FACS). As expected, cells lacking dCas9-VP64 or Sox1 targeting gRNAs appeared almost exclusively GFP-negative in flow analysis ( Fig. 1d  no gRNA non-targ. S1-9 e Fig. 1 Transcriptional editing of Sox1 leads to gene induction. a Schematic overview of the Sox1 and Actc1 locus in NPCs. Heterozygous knock-in of GFP into the Sox1-ORF and gRNA targets are shown. b Moderate upregulation of Sox1 compared to Actc1. gRNAs were transfected separately into NPCs. mRNA of Actc1 and Sox1 was quantified using qRT-PCR, and NPCs without transfection were used as a control population (no gRNA). Non-targeted loci were quantified as a control for unspecific effects. Actc1 mRNA upregulation is significantly higher than Sox1 mRNA upregulation (two-sided Student's t-test, p < 0.005). Data shown as the mean and standard error of the mean; n = 3 biological replicates (generated independently on different days in different clonal lines). c Alteration of gRNA sequences and positions does not significantly affect the efficiency of transcriptional engineering. Three different gRNA lentiviruses were transduced and cells were selected for gRNA expression. Data shown as the mean and standard error of the mean; n = 3 biological replicates (generated independently on different days in different clonal lines). d Flow cytometry reveals that a small population of cells is responsive to transcriptional engineering. SoxProm gRNAs were transduced into Sox1 GFP NPCs, and cells were selected for gRNA expression. Flow analysis of cells reveals that a minority of cells respond to transcriptional activation with induced with GFP fluorescence. e High Sox1 GFP levels correlate to high Sox1 protein levels. Sox1 GFP -positive and -negative cells were sorted from control populations (with either no gRNAs or non-targeting gRNAs) and activated populations, respectively. Western Blot reveales significantly higher Sox1 protein levels in the GFP-positive population from NPCs transduced with the SoxProm gRNAs only. Shown is one biological replicate; Image cropped for clarity targeting gRNAs also responded only in part. Only a minor proportion of the cells reacted to activator targeting with a significant induction of GFP protein (resulting in 1-6 % Sox1 GFPpositive cells, average 2.9%, p-value < 0.0001 tested with twosided t-test, n = 18 biological replicates generated with three different clonal NPC lines), while the majority of cells remained unaffected for all gRNA combinations used (Fig. 1d, Supplementary Fig. 1b). Sorting GFP-positive (subsequently referred to as Sox1 GFP positive or responsive) cells and GFP-negative (subsequently referred to as Sox1 GFP negative or non-responsive) cells and analysis by qPCR showed that the Sox1 GFP positive population is the main source of Sox1 induction, as significantly more Sox1 mRNA is found in those cells ( Supplementary Fig. 1c). This effect is even more pronounced on the protein level where the expression of Sox1 protein is almost exclusively detected in GFP-positive cells (~30-fold increase over no gRNA control, Fig. 1e).
To determine how lasting the induced expression of Sox1 is, we separated Sox1 GFP -positive and -negative cells by FACS (Fig. 2a). Both populations continued to proliferate with minor phenotypic changes, but flow analysis showed that populations differed strongly in Sox1 levels at 7 d after sorting. Sorted cells lacking Sox1-targeting gRNAs resembled the initial population, indicating that naturally occurring Sox1 GFP -positive cells are not a stable subpopulation, but rather short lived. In contrast, NPCs with targeted Sox1 induction gave rise to a population expressing higher GFP levels on average and containing many more GFPpositive cells (22%, Fig. 2a). Taken together, these data show that NPCs respond heterogeneously but can at least partially retain the activation of the developmental transcription factor Sox1.  Fig. 2a, b), a number of genes showed altered expression levels between the groups (1060 genes upregulated >4-fold, p-value < 0.05 tested with the Wald test, false discovery rate of 0.1), 482 genes downregulated (>4fold, p-value < 0.05 tested with the Wald test, false discovery rate of 0.1). The highest induced factors (>10-fold) contained many rarely studied genes, some of which have been shown to functionally contribute to early brain development (for example, the kinase Nuak2 25 , the retinoic acid early transcript Raet1a 26,27 or the proteolipid Neuronatin (Nnat) 28 ). Among those transcripts significantly induced (4-30-fold) were several genuine neural stem cell factors (e.g., Nestin, Dcx, Trnp1, FoxG1, Fig. 2c) and many more genes functionally involved in neural stem cell homeostasis, such as the growth factor Igf2 29 , the notch ligand Dll4 30 , the neural stem cell mitogen Amphiregulin 31 and the neurogenesis factor Fgf13 32 (Fig. 2b, c). These and most other genes significantly upregulated through Sox1 induction were also found significantly higher in the transcriptomes of early neural stem cells (either neural differentiation of embryonic stem cells (NSCs) or FACS-sorted neural rosettes (NRs), see methods) ( Fig. 2b, c, Supplementary Fig. 1d). Downregulated genes showed the opposite general tendency ( Supplementary Fig. 1c); however, this effect was far less prominent and the associated genes did not fall in a clear functional category. To corroborate these findings, we cultured NPCs with and without Sox1 GFP induction to analyze the expression of neuroepithelial markers by immunocytochemistry. Although Sox1 GFP -positive cells appeared overall morphologically similar (they did not form neural rosettes in vitro under NPC culture conditions) and did not induce Prominin (Cd133), the cells tended to cluster and induced several neural stem cell markers that were absent in NPCs, including Occludin (Ocln), and zona occludens 1 (Zo-1) and elevated others (Nestin, Notch1) that were already weakly detectable in control NPCs (Supplementary Fig. 3a, b).
Since transcriptomes and functional markers indicate at least a partial reversion toward a neural stem cell identity, we asked whether these changes translate into the restoration of cellular potency. To that end, we separated Sox1 GFP -positive and -negative NPCs by FACS and tested their differentiation potential. Under differentiation conditions, non-targeted cells or cells with control gRNAs gave rise almost exclusively to glial progeny (as indicated by the glial marker Gfap); however, Sox1 GFP -positive cells produced significantly more neuronal (as indicated by the marker for young neurons Tuj1) and less glial progeny (Fig. 2d, e; Supplementary Fig. 1c). When given time to mature (21d), the neurons predominantly expressed markers of glutamatergic neurons (vGlut1, Supplementary Fig. 3d, e). While Calbindin positivity could also be detected, tyrosine hydroxylase (Th) was absent. These data support the notion that the activation of the stem cell factor gene Sox1 induces transcriptional NSCs programs, which are sufficient to release NPCs from their glial commitment and restore neuronal differentiation potency.
The variable response to Sox1 induction is cell-intrinsic. Next, we investigated whether technical limitations of the CRISPRbased method could be the source of the heterogeneous response to Sox1 targeting. Immunoblot analysis revealed that Sox1 GFPpositive and -negative populations exhibit comparable levels of dCas9-VP64 protein (Fig. 3a). Apart from dCas9 protein, individual lentiviral integration sites of gRNA expression cassettes could influence gRNA levels and thus also contribute to the disparity in the cellular response. We generated clonal NPC lines stably expressing gRNAs targeting the Sox1 promoter; because each clone is derived from one single progenitor cell, all cells of one clone share the same gRNA integration site. We subsequently transfected these clonal NPCs with vectors containing dCas9-VP64 and selected for its expression. None of the tested clones failed to respond to transcriptional targeting, but neither did any clone respond with a significantly increased Sox1 GFP fraction compared to the earlier experiments ( Supplementary Fig. 4a), indicating that individual differences in gRNA integration sites and levels are likely negligible in this experimental setup and do not contribute to the observed cellular heterogeneity.
To investigate whether partial or heterogeneous gene induction is a general feature of targeted gene induction using dCas9, we analyzed the consequences of VP64 targeting on a control promoter (Actc1). As depicted in Fig. 3b, nearly all NPCs (stably expressing dCas9-VP64 and selected for gRNA plasmids) significantly upregulate Actc1 protein. Although this shows that a vast majority of these cells received functional levels of the CRISPR machinery (able to induce the expression of a silenced endogenous gene), only a small fraction of those simultaneously activate Sox1 in gRNA co-delivery experiments (Fig. 3b).
Recent publications indicate that the compaction of chromatin at the target locus might influence the binding of Cas9 33 . Different binding efficiencies in responding and non-responding cells could cause a heterogeneous cellular response despite comparable cellular levels of dCas9 protein and gRNAs. We therefore quantified the amount of targeted dCas9-VP64 in Sox1 GFP -positive and -negative populations using chromatin immunoprecipitation (ChIP). As depicted in Fig. 3c, dCas9-VP64 occupied the Sox1 promoter at comparable levels in Sox1 GFP -positive and -negative cell populations, strongly suggesting that processes downstream of transactivator binding interfere with the activation of Sox1 in non-responsive cells.
To further test whether the observed reluctance of a majority of NPC cells to activate Sox1 is a particularity due to the use of dCas9-VP64, we employed a composite CRISPR activator, VPR, additionally equipped with trans-activator domains of the NFkappa-B p65 subunit and the R transactivator of Epstein-Barr virus (RTA) 3,34 . Since VPR has been shown to be significantly more potent than VP64 on a series of promoters 3 , this experiment also allowed us to test whether stronger activation   signals would overcome the incomplete Sox1 induction. Flow cytometry and qPCR revealed that although more GFP mRNA can be detected in the bulk population when using dCas9-VPR, the responsible Sox1 GFP positive fraction is not elevated compared to dCas9-VP64 (Fig. 3d, Supplementary Fig. 4a). This result suggests that the ability of a large fraction of NPCs to evade gene induction is not specific to VP64. Taken together, these data indicate that the observed heterogeneous cellular response is not due to delivery, expression or binding of the CRISPR tools, but instead is a feature of the Sox1 gene.
Repressive chromatin at Sox1 correlates to unresponsive NPCs.
The previous experiments showed that the partial reluctance to respond to Sox1 induction is not caused by methodological variations; thus, we aimed to test whether heterogeneity in the chromatin of NPC populations might act as a cellular barrier and could explain at least in part the cellular response. We assessed the presence of chromatin marks frequently associated with gene repression on the Sox1 locus in Sox1 GFP positive and negative NPCs. For this, we performed ChIP to quantify the trimethylation of lysine 9 (H3K9me3) and lysine 27 (H3K27me3). As indicated Actc1 Sox1 no gRNA A1-9 + S1-9 A1-9 + S1-9  Fig. 3 Methodological variance is not the source of unresponsiveness to trans-activation. a dCas9 Protein levels do not differ between populations. Sox1 GFPpositive and -negative cells were sorted from control and activated populations (S1-9), and an immunoblot for dCas9 was performed. dCas9 levels did not significantly vary between the respective samples; Image was cropped for clarity. b Resistance to activation is gene-dependent. Actc1 gRNAs (A1-9) and Sox1 gRNAs (S1-9) were co-transduced into NPCs and immunostained for the target proteins after 7 days. While almost all cells induced Actc1 expression, only a small subpopulation was able to activate the Sox1 signal. Dashed yellow lines mark magnified areas; scale bar: 100 µm. c dCas9 binding does not vary between Sox1-positive and -negative NPCs. Sox1 GFP -positive and -negative cells were sorted from the activated population (SoxProm), and ChIP for dCas9 was performed at the Sox1 locus as well as on the promoters of Gapdh, Actc1, and Oct4 (serving as negative controls); dCas9 is strongly enriched at the Sox1 locus; no significant difference in DNA binding between responsive and unresponsive NPCs was detected. Data depict one of three biological replicates performed on different days in different clonal lines., mean and standard error of the mean of n = 3 technical replicates shown; n.s., not significant. d dCas9-VPR does not overcome the heterogeneity of gene activation observed with dCas9-VP64. NPCs stably expressing Sox1 targeting gRNAs (S1-9) were transfected with either dCas9 only, dCas9-VP64 or dCas9-VPR. Although higher Sox1 mRNA levels were seen when gene induction was conducted with dCas9-VPR, no significant difference in the number of Sox1 GFP -positive cells was detected between the dCas9-VP64-and dCas9-VPR-expressing NPCs (data depict one of three biological replicates, mean and standard error of the mean of n = 3 technical replicates shown) in Fig. 4a, the Sox1 promoter is marked by H3K9me3 to a similar degree as repressed control genes (Oct4, Actc1) in unresponsive cells, while NPCs with induced Sox1 expression have significantly lower H3K9me3 levels, almost to the degree of a strongly expressed housekeeping gene (Gapdh). However, H3K27 methylation did not show modulation between Sox1-expressing and non-responsive cells (Fig. 4b). DNA methylation has a much more complex relationship with transcriptional activity than histone methylation in general and particularly during neurogenesis 35 . To investigate the role of DNA methylation in the trans-activation of the master transcription factor Sox1, we conducted oxidative bisulfite sequencing (oxBS) on the Sox1 promoter (Fig. 4c). We also performed bisulfite sequencing (BS), which cannot distinguish between occurrences of hydroxymethylation (5hmC) and methylation (5mC), to infer approximate levels of DNA hydroxymethylation ( Supplementary  Fig. 5a). We found that the capacity to induce Sox1 correlates strongly to 5mC levels in a small region around the transcription start site, which is located inside the Sox1 CpG island. While the targeting of VP64 to the Sox1 promoter does not affect DNA methylation, since the Sox1 GFP -negative population displayed comparably high methylation levels on the analyzed CpG sites to control cells without dCas9 targeting (≥9 of 18 CpGs were methylated to 50% or higher) the responding Sox1 GFP -positive population had greatly reduced DNA methylation levels (18 of 18 CpGs lower than 20%, Fig. 4c, Supplementary Fig. 5b). By and large, the BS and oxBS data coincide, providing little indication for the presence of 5hmC at the Sox1 promoter, except in the responding Sox1 GFP -positive cell population, where the significant difference between the two methods indicates high levels of hydroxymethylation (ca. 20% of total (Fig. 4c, Supplementary  Fig. 5a-c)). All tested cell populations showed equal DNA methylation on a control region (Actc1), indicating that the detected differences are likely specific for the Sox1 promoter and do not represent a global cellular reduction of DNA methylation ( Supplementary Fig. 5d, e). To test whether the local loss of DNA methylation is a direct consequence of trans-activation, we targeted dCas9-VP64 to Actc1, which results in strong gene induction (Fig. 1b) but not in DNA methylation changes ( Supplementary Fig. 5f). Taken together, these experiments suggest that epigenetic differences could contribute to the disparate responsiveness of NPCs to activate Sox1.
Targeted epigenetic manipulation increases response of NPCs. Next, we aimed to test whether the detected epigenetic marks are a cause or a consequence of Sox1 transcription and whether they functionally hinder the activation of the master transcription factor. For this, we combined dCas9-VP64 with targetable chromatin enzymes at the Sox1 promoter in NPCs (Fig. 5a, b, Methods). To generate less restricted chromatin, e.g., to reduce DNA methylation, we included Tet1, which executes the first enzymatic steps during DNA de-methylation. To reduce H3K9me3, we employed JMJD2, a histone de-methylase acting on the repressive marks H3K9me3 and H3k36me3. To induce more receptive chromatin, we also tested simultaneous targeting with p300, a ubiquitous histone acetyltransferase. We found that although the targeting of these enzymes to the Sox1 promoter alone only has minor effects on GFP and Sox1 expression (Fig. 5a), p300 and Tet1 are able to significantly enlarge the responsive population when combined with dCas9-VP64 by twoand three-fold, respectively (Fig. 5b). Despite the observed differences in H3K9 trimethylation between Sox1-positive and -negative NPCs, the effect of targeting the enzymatically active domain of JMJD2 was not statistically significant.
In light of the observed DNA methylation changes correlating to Sox1 induction (Fig. 4c), the effects of Tet1 caught our attention. We therefore tested whether dCas9-Tet1 targeting to the Sox1 promoter had any effect on DNA methylation of the Sox1 promoter. Indeed, in contrast to dCas9-VP64 alone (Fig. 5c), dCas9-Tet1 significantly decreased DNA methylation levels around the Sox1 transcription start, even in unsorted populations (Fig. 5d). Since several publications indicate that Tet proteins can in certain cases directly influence transcription independently from DNA methylation 36 , we replicated the experimental setting but included an enzymatically compromised mutant of Tet1 37 . As shown in Fig. 5c, d and Supplementary Fig. 4c, d, the effect of targeted Tet1 on DNA methylation is largely dependent on its enzymatic activity. Furthermore, combining the catalytic mutant of Tet1 with VP64 led to a significantly lower number of GFPpositive cells than its wild type form (Fig. 5a, b), supporting the functional involvement of DNA modifications in this process. To further confirm the relevance of DNA methylation in the context of transcription factor-mediated Sox1 induction and changes of cellular identity, we also applied Zebularine, a strong Dnmt inhibitor with limited toxicity 38 in combination with dCas9-VP64 on the NPCs. As anticipated, neither Zebularine, dCas9-VP64 or gRNAs alone resulted in a significant number of Sox1 GFP -positive cells; however, when used together, the induction was comparable to that triggered by the combination of dCas9-VP64 and dCas9-Tet1 (Fig. 5a, b).
To confirm that targeted DNA de-methylation has a functional impact on cell identity, we released NPC bulk populations from self-renewal in the presence or absence of dCas9-Tet1. Figure 6a, b shows that, even without pre-selection of Sox1 GFP positive cells, significantly more neuronal progeny are generated when both Tet1 and VP64 are targeted to the Sox1 promoter, indicating that DNA methylation has a pronounced effect on the impact of a bound trans-activator and can serve as a robust barrier for cell identity changes (Supplementary Fig. 6e).
To test whether these findings are a peculiarity of Sox1 or whether other linage factors would show similar behavior, we chose four master transcription factors for further investigation (Ngn2, NeuroD4, Oct4 and Nkx2-2). While none of these factors are expressed in NPCs (Fig. 6c, d), only two contain high degrees of DNA methylation around their TSS (Oct4 and Nkx2-2, Fig. 6e). As for Sox1, we used two gRNAs to target dCas9-VP64 in the vicinity of the respective TSS (Ng1-9, Ne1-9, O1-9, Nk1-9). Through immunocytochemistry we observed that only one of the master transcription factors, NeuroD4, is upregulated in a significant proportion of NPC cells (ca 50%) in response to transcriptional engineering; the other three remain silent in most cells. To investigate whether (as for Sox1) targeted de-methylation can break down a barrier of transactivation, we combined transcriptional engineering with epigenome editing. While dCas9-dTet1 has minimal effects, dCas9-Tet1 multiplies the proportion of responding cells (Fig. 6d), but only for those genes where DNA methylation was present (Oct4 and Nkx2-2, Fig. 6e). Ngn2 showed minimal response in either condition, indicating that different safeguarding mechanisms are operating at the Ngn2 locus. Taken together, these data strongly suggest that DNA methylation, among other chromatin modifications, tightly controls Sox1 (and other master transcription factor) induction and makes their genes insensitive even to potent transcriptional activators. However, manipulating these modifications accordingly can release the functional block and make cells undergo cell fate changes.

Discussion
Here we show for several example master transcription factors, such as Sox1, Nkx2-2 and Oct4, that the induction of developmental transcriptional programs must overcome functional barriers. These are constituted (at least in part) by chromatin features on the promoter of the master transcription factor itself. We found that some of these epigenetic features have the potential to decouple the effects of a bound transcription factor on the transcription of another (and thus play important roles in cell identity robustness). It is clearly evident that the targeted activation of transcription varies in the order of three to four magnitudes depending on the gene, but the molecular basis for this variation and to what extent cellular or experimental heterogeneity might contribute are currently unknown.
Here, we use transcriptional engineering of the widely used neuro-epithelial marker Sox1 to increase the potency of neural progenitor cells. In line with our results, Sox1 overexpression in vivo has been shown to bias neural progenitors to neuronal commitment, while progenitors from Sox1-deficient mice generate fewer neurons 19 . Thus far, few examples have been reported in which cell identity could be manipulated using the transcriptional activation of an endogenous gene 5,10,39 . We find that with transcriptional engineering strong Sox1 protein induction can be achieved but that only small subsets of cells are able to respond to the transactivation signal. This is to our knowledge the first study examining the sources of heterogeneity in cellular responses during transcriptional editing, separating experimental from biological heterogeneity. Due to the widespread use of overexpressing exogenous gene copies in cellular reprogramming models, the concept of cell identity barriers has mainly concentrated on the targets of reprogramming factors 40,41 .
Here, we investigated instead the chromatin processes interfering with the activation of an endogenous master transcription factor, thereby protecting the existing cell identity. For this we employed epigenome editing, a technique with numerous applications whose effectiveness along the lines of cell identity has recently been demonstrated on an enhancer element of the muscle reprogramming factor MyoD 42 . In contrast to this and other publications employing dCas9-Tet1, DNA de-methylation alone had limited impact on transcription of Sox1. Only when additional transcriptional stimuli were applied, absence or presence of DNA methylation really made a significant difference. If this is as common among master transcription factors, as our data indicates, it could solve the paradox that although DNA methylation is needed for development, repressed developmental genes are rarely activated when DNA methylation is eliminated 39,43 .
With this approach, we were able to separate cause from consequence and investigate, which chromatin features functionally interfere with the consequence of a bound trans-activator. Although our results suggest that several tested chromatin processes might form an intrinsic barrier against cellular conversion Fig. 5 Targeted DNA demethylation lowers a barrier of Sox1 induction. a Epigenetic editing alone has a minor effect on Sox1 gene induction. NPCs stably expressing Sox1-targeting gRNAs (S1-9) were transfected with dCas9 tethered to VP64 or enzymatic domains of canonical chromatin modifying enzymes (JMJD2a, p300, Tet1). Flow cytometry revealed that none of the epigenome editing constructs superseded the effect of dCas9-VP64. Data and mean shown are derived from three biological replicates, performed on different days in different clonal lines. b Epigenetic engineering enhances the number of Sox1-positive NPCs. NPCs were transduced with gRNAs and subsequently transfected with different chromatin modifiers tethered to dCas9. Flow analysis reveals a significantly higher proportion of Sox1-positive cells when VP64 was combined with Tet1 or P300, respectively (but not JMJD2a), compared to combination with dCas9 only. In addition, the effect of the combination with Tet1 was significantly higher than the combination with its catalytically dead mutant dTet1 (data are shown as single biological replicates and mean; *p < 0.05; **p < 0.01; ***p < 0.005 calculated by two-sided Student's t-test. n = between 3 and 5 biological replicates, generated and analyzed on separate days, in three different clonal lines). c, d Oxidative bisulfite sequencing reveals efficient de-methylation of the Sox1 promoter by dCas9-Tet1. NPCs were transduced with gRNAs (S1-9) and either subsequently transfected with dCas9-Tet1 and dCas9-dTet1 or not. c Oxidative bisulfite sequencing after 11 days shows that on unsorted populations no change in the proportion of methylated CpGs is apparent between gRNA receiving and control cells provided that no dCas9-Tet1 is present. d However, dCas9-Tet1 transfection and selection lead to an efficient reduction of DNA methylation at the Sox1 promoter. Furthermore, no demethylating effect of dCas9-dTet1 can be seen (data depict the means and standard error of the mean of NGS reads of at least n = 1000 individual sequences generated from four individual biological replicates generated and analyzed on separate days, in three different clonal lines) (and the reduction of individual features might not be fully penetrant), we concentrated further on DNA methylation for three main reasons: firstly, we found the significant reduction of DNA methylation around the Sox1 transcription start site promoter in responsive NPCs. Secondly, critical roles for DNA methylation during neural cell fate choices, in particular in glial vs. neuronal commitment, have been suggested 35,44 . Thirdly, we were able to reproduce the effects of the DNA de-methylase Tet1 with a fully independent approach, applying an inhibitor of DNA methyltransferases. Importantly, we could detect similar principles when activating Oct4 and Nkx2-2, showing that the observed effects are not unique to Sox1 and have broader implications.
How DNA methylation might interfere with the cellular ability to respond to trans-activation remains elusive; however, many neural (and ubiquitous) transcription factors are predicted to bind sites in the Sox1 promoter 45

Oct4
No gRNA O1-9 O1-9 + dCas9-Tet1 Oct4 Oct4 80 Fig. 6 DNA methylation as a barrier to transcriptional engineering is not an exclusive feature of Sox1. a, b Combination of transcriptional engineering with epigenome editing increases the neuronal differentiation potency of NPCs. NPCs stably expressing dCas9-VP64 were transduced with Sox1-targeting gRNAs (S1-9) and subsequently transfected with dCas9-Tet1 or dCas9-dTet1. NPCs were stained for Sox1 or differentiated for 7 more days, respectively. Sox1 staining revealed a higher number of Sox1-positive cells when dCas9-VP64 is combined with dCas9-Tet1 but not with dCas9-dTet1. After differentiation, more cells differentiated into the neuronal lineage, as indicated by the higher Tuj1 positivity after epigenome editing compared to transcriptional editing alone. Data are shown as the mean and standard error of the mean of n = 3 biological replicates, performed on different days in different clonal lines. ***p < 0.005 calculated by two-sided Student's t-test. Dashed yellow lines mark magnified areas; scale bar: 100 µm. c, d Activation of different master transcription factors reveals varying responsiveness to transcriptional engineering and sensitivity to DNA demethylation. NPCs expressing dCas9-VP64 were transfected with gRNAs (Ng1-9, Nk1-9, O1-9, Ne1-9) targeting different master transcription and reprogramming factors (Ngn2, Nkx2-2, Oct4, NeuroD4). Induction varied considerably between different targets, although three of four master transcription factor genes mostly resisted gene activation. Transfection of dCas9-Tet1 but not dCas9-dTet1 almost quintupled the cell population responsive to Nkx2-2 transactivation and tripled that inducing Oct4. Data shown as the mean with standard error of the mean of n = 3 technical replicates; scale bar: 100 µm. e DNA methylation levels at master transcription factor promoters predict responsiveness to epigenome editing. Oxidative bisulfite sequencing was performed in NPCs to quantify DNA methylation levels at regions including the TSS of the tested master transcription factors. While the promoter of Ngn2 and NeuroD4 appear almost completely unmethylated, that of Nkx2-2 and Oct4 exhibit high methylation levels at the TSS. Data derived from two biological replicates are shown as the mean and standard error of the mean of all analyzed CpGs (Ngn2: 13, Nkx2-2: 8, Oct4: 9, NeuroD4: 6) inside a 100-300 bp region surrounding the TSS (for the genomic position see Supplementary Table 1); Dots show the methylation levels of single CpGs in analyzed loci overlap CpG sites (Fig. 4c). Several of these factors, including E2F-1 46 , Sp1 47 and YY1 48 have been reported to bind their DNA motive in a methylation-sensitive manner and/or play important roles during neurulation 49 and neurogenesis 50 . Moreover YY1 and YY1, Sp1 and E2F-1 motifs are also occurring at the Nkx2-2 and Oct4 promoter respectively. However, other factors and DNA motifs could be equally relevant in this context. It is currently unclear how many of the myriad of individual chromatin features have functional roles, but it is not unlikely that many of those are only relevant when new conditions arise, may these be local (a new transcription factor binding event) or global (cellular identity changes). We show that transcriptional engineering and epigenetic editing can be combined to investigate such conditional causalities, paving the way to comprehending the complex interplay between transcription factor binding and effects, chromatin features and barriers, as well as cellular identity and heterogeneity.
gRNA design and plasmid generation. gRNA targeting sequences were designed using a free online platform (www.benchling.com) that employs a published algorithm for gRNA binding efficiency 51 . Specificity scores above 40 were set as requirements, and only sequences with a 5'G were considered in order to fit the requirements of the human U6 promoter. gRNA sequences were designed in the region starting 250 bp upstream of the transcription start site (TSS) up to the TSS. For the STAgR constructs, the gRNAs of each pair were required to be at least 100 bp apart to avoid spatial hindrance. STAgR plasmids were generated using the protocol provided by Breunig et al., 2018 24 . For lentiviral gRNA plasmids, the vector pLKO.1 (Addgene plasmid 10878) was modified as described in Koeferle et al. 52 . For subcloning of STAgR constructs into the modified pLKO1 backbone, the gRNA cassettes were amplified by PCR using the amp_gRNA_pLKO_fwd and amp_gRNA_rev primers. 25 μl Phusion High-Fidelity PCR Master Mix with HF Buffer (NEB, M0531S), 0.5 μl of each primer (100 μM), 1 ng template in a total reaction volume of 50 μl. The backbone was digested with AgeI-HF (NEB, R3552S). Insert and backbone were gel-purified and mixed in a molar ratio of 3:1, and 2.5 µl of the mix were incubated in 2.5 µl 2x Gibson Mastermix for 30 min at 50°C. For cloning of single gRNAs into the lentiviral backbone, gRNA sequences were ordered as strings (see Supplementary Table 3). Strings were amplified using the libgen_fwd and libgen_rev primers. The PCR mix contained 0.1 ng DNA template, 25 µl 2x Phusion High-Fidelity PCR Master Mix with HF Buffer, 0.5 µl of each primer (100 µM), in a total reaction volume of 50 µl. The backbone was digested with AgeI-HF. Insert and backbone were gel-purified and mixed in a molar ratio of 3:1, and 2.5 µl of the mix were incubated in 2.5 µl 2x Gibson Mastermix for 30 min at 50°C. Transformation, PCR and plasmid preparations were done according to routine laboratory practice. gRNA sequences are listed in Supplementary Table 3.
Cell culture, transfections and viral transductions. Murine Sox1 GFP NPCs, NSCs and NRs were derived from Sox1 GFP ESCs (gift from Austin Smith) as described 53 ) and cultured in NeuroCult TM Proliferation kit (mouse, Stemcell Technologies, Catalog #05702) media. For expansion of NPCs the media was supplemented with 10 ng ml −1 human recombinant bFGF (Stemcell Technologies, Catalog # 78003), 20 ng ml −1 human recombinant EGF (Stemcell Technologies, Catalog #78006) and 1 µg ml −1 Laminin (Sigma, L2020). Cells were grown in a monolayer in Laminin coated cell culture dishes at 37°C and 5% CO 2 . For the establishment of clonal dCas9-VP64 expressing cell lines, 300000 cells were seeded into 6-well plates. In total 2.5 µl Lentivirus were added to the medium. After 3 days, selection with 8 µg ml −1 Blasticidin S (ThermoFisher, R21001) was started and kept throughout the duration of experiments. After seven days, resistant cells were seeded as single cells in 96-well plates. Clones were expanded and dCas9 expression verified by qPCR and western blot. For transfections, 350000 cells were plated into six-well plates and 2 µg or 3.2 µg of STAgR plasmid and/or dCas9 expression plasmid respectively were transfected using Lipofectamin® 2000 (ThermoFisher Scientific, 11668027) according to the manufacturer's protocol. dCas9 expression from plasmids was selected with 150 µg ml −1 Hygromycin B (ThermoFisher, 10687010). For transduction of gRNAs, 250000 cells were plated into T25 cell culture flasks. A volume of lentivirus equivalent to 1 × 10 6 particles was added to the medium. To select for gRNA expression, 0.08 µg ml −1 Puromycin (Thermo-Fisher, A1113803) was added to the medium. qPCR primer and antibodies are listed in Supplementary Table 4 and 5.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Raw Sequencing data is publicly available at GEO under accession number GSE119480 (RNAseq) and at SRA under accession numbers PRJNA490128, PRJNA522700, and PRJNA522707 (Bisulfite and oxidative Bisulfite Sequencing). Raw Data underlying all Figures are provided as a Source Data File. All relevant data can also be inquired from the authors.