The Hox Transcription Factor Ubx stabilizes Lineage Commitment by Suppressing Cellular Plasticity

During development cells become gradually restricted in their differentiation potential by the repression of alternative cell fates. While we know that the Polycomb complex plays a crucial role in this process, it still remains unclear how alternative fate genes are specifically targeted for silencing in different cell lineages. We address this question by studying Ultrabithorax (Ubx), a multi-lineage transcription factor (TF) of the Hox class, in the mesodermal and neuronal lineages using sorted nuclei of Drosophila embryos and by interfering with Ubx in mesodermal cells that have already initiated differentiation. We find that Ubx is a key regulator of lineage development, as its mesoderm-specific depletion leads to the de-repression of many genes normally expressed in other lineages. Ubx silences expression of alternative fate genes by interacting with and retaining the Polycomb Group (PcG) protein Pleiohomeotic (Pho) at Ubx targeted genomic regions, thereby setting repressive chromatin marks in a lineage-dependent manner. In sum, our study demonstrates that Ubx stabilizes lineage choice by suppressing the multi-potency encoded in the genome in a lineage-specific manner via its interaction with Pho. This mechanism may explain why the Hox code is maintained throughout the lifecycle, since it seems to set a block to transdifferentiation in many adult cells.


ABSTRACT
During development cells become gradually restricted in their differentiation potential by the repression of alternative cell fates. While we know that the Polycomb complex plays a crucial role in this process, it still remains unclear how alternative fate genes are specifically targeted for silencing in different cell lineages. We address this question by studying Ultrabithorax (Ubx), a multi-lineage transcription factor (TF) of the Hox class, in the mesodermal and neuronal lineages using sorted nuclei of Drosophila embryos and by interfering with Ubx in mesodermal cells that have already initiated differentiation. We find that Ubx is a key regulator of lineage development, as its mesoderm-specific depletion leads to the de-repression of many genes normally expressed in other lineages. Ubx silences expression of alternative fate genes by interacting with and retaining the Polycomb Group (PcG) protein Pleiohomeotic (Pho) at Ubx targeted genomic regions, thereby setting repressive chromatin marks in a lineage-dependent manner. In sum, our study demonstrates that Ubx stabilizes lineage choice by suppressing the multi-potency encoded in the genome in a lineage-specific manner via its interaction with Pho. This mechanism may explain why the Hox code is maintained throughout the lifecycle, since it seems to set a block to transdifferentiation in many adult cells. Although lineage-restricted TFs could in principle fulfil this function, it would ask for a highly complex regulatory architecture. On the other hand, TFs expressed in multiple/all lineages with a rather promiscuous binding behaviour would dramatically reduce this complexity and would elegantly explain lineage-specific gene regulation based on the interaction of these broadly expressed TFs with lineage-restricted factors.
Hox TFs represent an excellent model to address this fundamental question, since they are active in all lineages along the anterior-posterior (AP) axis of bilaterian animals. Importantly, the famous and well-known identity switch of whole body parts, the homeotic transformation, which is induced by altered Hox expression 9,10 , shows that Hox TFs have comparable functions in all lineages during development and indicates that they control the development of different lineages in a highly specific manner. The latter conclusion is supported by recent experiments showing that mesoderm-derived vascular wall mesenchymal stem cells (VW-MSCs) can be generated in vitro from induced pluripotent stem cells (iPSCs) simply by inducing the expression of a mixture of Hox genes that are selectively expressed in adult VM-MSCs 11 . However, while this study showed that Hox genes alone are sufficient to induce the generation of one specific cell type of one lineage in vitro, it left the questions open whether this is also the case in vivo, and how Hox TFs unambiguously select among the many possible transcriptional programs only one, which drives a cell or a whole lineage into one specific direction.
One major bottleneck in this direction is that genome-wide Hox chromatin binding studies have been performed so far mainly in cell culture systems 12,13 , specialized epithelial tissues 14 , or mixtures of cell lineages 15 , hampering the identification of common and lineage-specific mechanisms employed by Hox TFs in different lineages in vivo. Furthermore, unlike lineage-restricted TFs, which are often tested in vivo using ectopic expression systems 16,17 , the functional analysis of TFs acting in multiple lineages requires the targeted interference with these factors in individual lineages.
With the availability of conditional genome editing 18,19 and nanobody driven protein degradation systems 20 this is now possible in an efficient manner and allows to elucidate the mode of action of Hox TFs in individual lineages, which are located in an otherwise unperturbed tissue environment at any stage in development. This is particularly important for multi-lineage TFs like the Hox proteins, which extensively control cell communication 21,22 and thus influence their own action in neighbouring lineages.
Here, we probe the Hox TF Ubx in the mesodermal and neuronal lineages using sorted nuclei of Drosophila embryos and by interfering with Ubx function specifically in the mesoderm lineage that is specified and fully committed to the mesodermal fate. To this end, we generated a GFP-Ubx gene fusion at the endogenous locus using CRISPR-Cas9 and homologous recombination. We show that Ubx is a key regulator of lineage development and diversification, as it controls the mesodermal and neuronal transcriptional programs with high specificity despite interacting with many genes in both lineages. Intriguingly, our study demonstrates that Ubx controls lineage specification and differentiation by restricting cellular plasticity within each lineage. In the mesoderm Ubx executes this function by silencing alternative fate genes, and we show that this repression requires the interaction of Ubx with the Polycomb Group (PcG) protein Pleiohomeotic (Pho) on co-bound chromatin sites. We furthermore find that Ubx stabilizes Pho binding to Ubx targeted genomic regions, and that this interaction is critical for gene repression by controlling the balance of H3K27me3 as well as H3K27ac at these sites. Taken together, our study demonstrates that Hox TFs control not only the segmental identity but also the developmental programs intrinsic to each lineage with high precision, and that one of the prominent functions of this multilineage TF family is the repression of alternative lineage programs, which restricts cellular plasticity in a lineage-specific manner.

Drosophila embryonic mesodermal and neuronal lineages
To elucidate mechanisms that enable a multi-lineage TF to instruct the development of divergent lineages, we recorded on the tissue level chromatin binding of the broadly expressed Hox protein Ubx in the mesoderm and nervous system and determined the transcriptomes of these two lineages during two stages of Drosophila embryogenesis.
To this end, we used the isolation of nuclei tagged in specific cell types (INTACT) method 13 by inducing the expression of the nuclear membrane protein Ran GTPase activating protein (RanGAP) fused to a biotin ligase acceptor peptide and the E. coli Biotin ligase (BirA) in the mesoderm via the pan-mesodermal twist (twi) 14  We analysed the nuclear transcriptome by high-throughput RNA sequencing (RNAseq) using INTACT-sorted mesodermal or neuronal nuclei obtained from stage 10 to 13 (4-9h after egg lay (AEL)) and stage 14 to 17 embryos (10-18h AEL). Robust and reproducible data were obtained for all samples in biological duplicates. In total, transcripts (containing over 10 RPKM) corresponding to 4588 coding genes were identified for the early mesodermal, 4133 coding genes for late mesodermal, 5336 coding genes for the early neuronal and 4833 coding genes for the late neuronal nuclei populations, which included genes typical for the lineage type and developmental stage (Supplementary Tab. 1). Pearson correlation coefficient analysis revealed that the transcriptome of the mesodermal lineage was clearly distinct from the neuronal one at both time points (r=0.72 for stages [10][11][12][13] ; r=0.53 for stages [14][15][16][17] (Fig. 1F, Supplementary   Fig. 1I). This was also reflected in a high number of genes differentially expressed in the mesodermal as well as the neuronal lineages when comparing identical stages ( Fig. 1G, H). Importantly, GO term classification revealed a significant enrichment of processes typical for the respective lineage (mesoderm both stages: p-value < 2.2e-16, neuronal stages 10-13 : p-value < 1.26e-6, neuronal stages [14][15][16][17] : p-value < 5.3e-5) (Figs 1G, 1H). In addition to elucidating differences in tissue profiles, Pearson correlation coefficient analysis also showed that global gene expression in the mesodermal lineage changed substantially over the selected time points (r=0.78 for mesoderm stages 10-13 + stages [14][15][16][17] (Fig. 1F, 1H, Supplementary Fig. 1I). Tissue-and stagedependent differences and similarities were very well reflected in the distances calculated by principal component analysis (PCA) (Fig. 1E). PCA analysis also showed that in contrast to the mesoderm the neuronal transcriptomes were very similar at both developmental time frames ( Supplementary Fig. 1I), which we assumed to be a consequence of the earlier onset of nervous system differentiation 27,28 .
We next profiled genome-wide Ubx binding in the same lineages and identical time windows by chromatin immunoprecipitation coupled to massively parallel sequencing (ChIP-seq) using 1x10 6 INTACT-sorted mesodermal and neuronal nuclei and an Ubx specific antibody generated and verified in the lab (see Materials and Methods). The data was benchmarked by the identification of Ubx binding to known target loci. One example is the well-characterized interaction of Ubx with the decapentaplegic (dpp) enhancer, which is required for dpp activation in parasegment 7 of the embryonic visceral mesoderm 29,30 . We found Ubx to interact with the dpp visceral enhancer using chromatin from INTACT sorted mesodermal but not neuronal nuclei ( Fig. 2A), showing that our data reflected Ubx interactions in vivo. However, this analysis also uncovered that Ubx bound a large fraction of genes in both lineages ( Supplementary Fig. 2D, E), although this TF is known to have different functions in the developing mesoderm and nervous system [31][32][33][34] . To resolve this discrepancy, we overlapped Ubx binding events and transcriptome profiles. We found that only 19 to 27% of the Ubx chromatin interactions occurred in the vicinity of genes actively transcribed either in the mesodermal or neuronal lineages at the different stages (Fig. 2B, Supplementary Fig.   2A), with more than 80% of these interactions occurring at intron, intergenic and distal enhancer regions (Fig. 2E). This included Mef2, teashirt (tsh) and β-Tubulin at 60D (βTub60D) in early and thin (tn), α-actinin (Actn) and Tropomyosin 1 (Tm1) in late mesodermal nuclei, while in early neuronal nuclei deadpan (dpn), huckebein (hkb) and Neurotrophin 1 (NT1) in late neuronal nuclei Neuroglian (Nrg), target of PoxN (tap) and castor (cas) was among the Ubx bound active genes. In contrast, the majority of the Ubx chromatin interactions (72 to 80%) were close to inactive genes in the two lineages ( Fig. 2B, Supplementary Fig. 2A) While it is well documented that Hox TFs function as activators and repressors depending on the context 35,36 , the high number of Ubx chromatin interactions at non-transcribed genes was unexpected. By determining the gene functions associated with Ubx interactions, we found a substantial overrepresentation of GO terms characteristic for the respective lineage among the Ubx targeted and expressed genes (mesoderm: p-value 2.2e-16, neuronal stages 10-13 : p-value 3.3e-6, neuronal stages 14-17 : p-value 0.002) (Fig. 2C, D), while Ubx interactions at inactive genes occurred frequently at genes controlling processes active in other lineages (Supplementary Fig. 2A-2C). For example, ectodermal and neuronal but not mesodermal GO terms were highly enriched among the inactive genes bound by Ubx in the early mesodermal lineage (neuronal 10-13 : p-value 0.0014, ectoderm 10-13 : p-value 0.05).
In order to comprehensively analyse the binding behaviour of Ubx to active and inactive genes, we used the WEADE tool, which identifies and visualizes overrepresentation of functionally related biological GO terms assembled to higherorder GO term sets and allows the representation and comparative analysis of multiple gene sets 37 . This analysis uncovered a high correlation between the lineage-specific transcriptional profiles and the genome-wide Ubx interactions (Fig. 2G). For example, many genes expressed in the early mesoderm controlled stem cell, cell cycle and translation related processes, and Ubx interactions were found enriched in the vicinity of these genes, while Ubx hardly interacted with genes encoding gene functions that were not represented among the active genes, like metabolism, signalling, stimulus and transport/trafficking related functions (Fig. 2G). A similar correlation was found among the inactive genes, as Ubx interactions were again highly enriched at gene classes found to be overrepresented among the non-expressed genes, while Ubx did not interact with underrepresented gene classes (Fig. 2G). This result suggested that Ubx binding controlled global gene expression in the mesoderm, the activation as well as repression, in a comprehensive manner. In order to test whether this binding behaviour was a general characteristic of TFs, we analysed GO term enrichment of genes bound by Tinman (Tin), a NK homeodomain TF expressed exclusively in the mesoderm 38 . To this end, we used genome-wide binding data of Tin profiled at similar embryonic stages 39 , and analysed higher-order GO term representation among the genes bound by Tin that were either transcribed or silent in the mesoderm. In contrast to Ubx, Tin interactions were mostly independent of gene classes represented among the active and inactive genes (Fig. 2G). Indeed, Tin preferentially interacted with genes controlling stem cell processes irrespective of whether these genes were actively transcribed or silent (Fig. 2G). We assumed the differential binding behaviour of Ubx and Tin to reflect the more restricted function of Tin in the mesoderm, as this TF controls the determination and specification of the cardiac, visceral and dorsal mesoderm 40,41 , while Ubx seemed to generally control development of the mesoderm (Fig. 2G).
In sum, these results illustrated that the broadly expressed Hox TF Ubx played a prominent role in orchestrating the transcriptional program in the mesodermal (and neuronal) lineages. In addition, our analysis revealed that a substantial fraction of Ubx interactions were found in the vicinity of inactive genes, which encoded many mesoderm-unrelated functions. Thus, we hypothesized that one important function of Ubx in tissue development could be the repression of alternative transcriptional programs, which instruct the development of other lineages.

The generic Hox transcription factor Ubx functions as a major regulator of lineage programs
Our analysis of Ubx function in the mesodermal and neuronal lineages was so far based on correlating lineage-specific Ubx binding profiles with RNA-seq probed gene expression. In order to elucidate which genes are under direct Ubx control, we profiled the transcriptional output induced in mesodermal cells devoid of Ubx protein, while leaving Ubx levels in all other cell and tissue types unchanged. We focused our analysis on the mesoderm, as its control by various lineage-restricted TFs is well described [40][41][42][43][44] . In order to deplete Ubx in the mesoderm, we used the targeted degradation of GFP fusion proteins. To this end, we first generated an endogenously the GFP-Ubx fusion protein in a lineage-specific manner, we used the deGradFP system, which harnesses the ubiquitin-proteasome pathway to achieve direct depletion of GFP-tagged proteins 20 . We first functionally verified the system by ubiquitously degrading the GFP-Ubx fusion protein using the armadillo (arm)-GAL4 driver 47 , which resulted in a strong reduction of GFP-Ubx protein levels (Supplementary Fig. 3E-G, O), and in animals resembling the Ubx null mutant phenotype ( Supplementary Fig. 3M, n) 10 . In a next step, we applied deGradFP to specifically interfere with Ubx function in the mesoderm using the Mef2-GAL4 driver 48 . This combination substantially decreased GFP-Ubx protein accumulation in the mesoderm (Fig. 3C, D, G), and the expression of the direct Ubx target gene dpp in the visceral mesoderm ( Fig. 3H-J) 30,49 . Consequently, we only observed the well-described loss of the third midgut constriction in these embryos (Fig. 3E, F), which is caused by the absence of Ubx activity in this tissue 50 Neurotrophin 1 (NT1) and myospheroid (mys), while 1452 genes exhibited reduced expression, including βTub60D, Ankyrin 2 (Ank2) and Notchless (Nle) (Fig. 3K). PCA analysis confirmed that the mesodermal transcriptomes in the absence (mesoderm [14][15][16][17] Ubx Degrad ) or presence (mesoderm [14][15][16][17] of Ubx were substantially different (Fig. 3N).
Strikingly, 85% (1227/1452) of the genes with reduced and 90% (1299/1393) of the genes with increased expression were bound by Ubx in mesodermal nuclei in wild-type embryos, implying that most of the expression changes were a direct consequence of altered Ubx chromatin interactions. To have a global view of the biological processes directly controlled by Ubx, we compared overrepresentation of higher-order biological GO terms between the Ubx degradation and control transcriptomes sets using the WEADE tool. We observed that processes were not randomly changed in the absence of Ubx but were mostly in agreement with Ubx chromatin binding (Fig. 3M). For example, translation and cell cycle processes, which were enriched among the expressed as well as Ubx bound active genes in the control transcriptome, were now represented among the transcripts with reduced accumulation in the absence of Ubx ( Fig. 3M), supporting that these processes were directly activated by Ubx. On the other hand, stimulus and signalling related processes, which were only to a minor extent represented among the genes expressed in the mesoderm, were found enriched among the genes with enhanced activity in the absence of Ubx (Fig. 3M). Consistent with a repressive function of Ubx, these processes were overrepresented among the inactive genes bound by Ubx in control mesodermal nuclei (Fig. 3M).
Intriguingly, we found differentiation processes to be enriched among the genes up- Taken together, these results showed that Ubx controlled a large number of genes encoding diverse functions. The strong correlation between Ubx binding with specifically repressed or activated classes of genes suggested that Ubx indeed exerts a dominant influence over the mesodermal transcriptome at the two stages analysed.
In addition, this analysis highlighted that Ubx repressed many genes controlling the establishment of non-mesodermal, alternative lineages, indicating that Ubx could have a pivotal role in restricting cellular plasticity, thereby ensuring that lineages adopt a unique identity.

Ubx represses alternative fate genes by organizing the epigenetic landscape
One question emerging from these results was how Ubx mediated the repression of non-mesodermal fate genes. As gene expression critically depends on the epigenetic status of the gene regulatory control regions 52 In a next step, we asked how the epigenetic landscape changed in the absence of Ubx, in particular at genomic regions with overlapping Ubx and repressive H3K27me3 marks. In total, 1216 H3K27me3 peaks and 1480 H3K27ac peaks located in gene regulatory regions, which corresponded to 2768 genes, changed their chromatin state in the absence of Ubx, as they experienced either a reduction/loss or an increase/gain in tri-methylation/acetylation at lysine 27 of histone 3 (Fig. 4E, Supplementary Fig. 4E).
Importantly, 80% of these genomic regions possessed an Ubx binding peak in control mesodermal cells, indicating that Ubx was critically required for establishing or maintaining the majority of these histone marks. Intriguingly, 1058 of these genes were primed for gene expression, as they experienced a loss/reduction of H3K27me3 and/or gain/increase of H3K27ac in their regulatory regions ( One result that puzzled us was that the majority of genes primed for activation (894 of 1058) remained silent on the RNA level when Ubx was lineage-specifically degraded ( Fig. 4D). In order to resolve this discrepancy, we analysed H3K27me3 and H3K27ac changes at introns, intergenic and distal enhancer regions, which we collectively labelled enhancers (Fig. 4F), as well as at promoters. We found that H3K27ac marks were equally increased at promoters and enhancers of genes unchanged or upregulated in the absence of Ubx (Fig. 4F). However, while H3K27me3 marks were significantly decreased at enhancers as well as promoters of up-regulated genes, unchanged genes showed a significant increase in H3K27me3 at their enhancers (Fig.   4F). As a high proportion of Ubx and H3K27me3 peaks co-occurred at enhancers ( Supplementary Fig. 4D), this result revealed that Ubx played a major role in repressing the expression of alternative fate genes by controlling the deposition/maintenance of H3K27me3 marks at enhancers.
In sum, these results showed that Ubx mediated the repression of genes encoding non-mesodermal functions by controlling the epigenetic status, in particular H3K27me3, at Ubx bound chromatin sites located in enhancers.

Ubx interacts with Pho at H3K27me3 marked inactive genes encoding alternative fate genes
To identify factors that together with Ubx could mediate the repression of alternative fate genes thereby restricting lineage identity, we performed a DNA motif search using all Ubx peaks either overlapping with H3K27me3 repressive or H3K27ac active marks.
We found in both cases motifs for Ubx, Extradenticle (Exd), a TALE class homeobox TF functioning as Hox cofactor in invertebrates and vertebrates 58,59 , and Trithorax-like (Trl), a GAGA factor activating and repressing gene expression by chromatin modification 52 , among the highest ranking motifs. In contrast, the DNA binding motif for the zinc finger protein Pleiohomeotic (Pho) was specifically enriched only among the co-occurring Ubx and H3K27me3 binding events (Fig. 5A). Interestingly, it has been shown just recently that in the absence of Pho, which recruits PcGs to PREs 60-62 , H3K27me3 marks were reduced in Polycomb regions and redistributed to heterochromatin 63 . In combination with our finding that H3K27me3 levels were reduced specifically at up-regulated genes (Fig. 4F), we hypothesized that Pho could function together with Ubx in the lineage-specific repression of alternative fate genes.
Consistent with this idea Pho was found expressed in mesodermal cells also expressing Ubx (Fig. 5B -C''). Furthermore, we confirmed an interaction of Ubx and Pho proteins in a complex in vitro and in vivo by performing co-immunoprecipitation (Co-IP) experiments in cellulo using Drosophila S2R+ cells transfected with tagged versions of Ubx and Pho as well as in vivo using GFP-Ubx embryos (Fig. 5F, G). This result suggested that Ubx and Pho could interact with the same chromatin regions to control gene expression, thus we analysed high-resolution Pho maps retrieved from embryonic (stage 9 to 12) mesodermal cells 64 . We found 4814 chromatin regions (which represent 62% of the Ubx and 12% of the Pho binding events) to be co-bound by Ubx and Pho and marked by H3K27me3 in stage 14-17 mesodermal cells (Fig. 5D,   6a, Supplementary Fig. 5A), with 40% of these events occurring at promoters and 49% at enhancers (Supplementary Fig. 5C). Consistent with the reported role of Pho in transcriptional repression [60][61][62] , 76% (3578/4715) of the genes bound by Ubx and Pho as well as marked by H3K27me3 were not expressed in the mesoderm (Fig. 5D), and GO term analysis revealed that the majority of the genes encoded non-mesodermal and stem cell-related functions (p-value 2.87e-11) (Fig. 5E). This was different for the remaining 1137 genes also bound by Ubx and Pho and marked by H3K27me3 but expressed in mesodermal cells, as they encoded gene functions controlling processes typical for the tissue type and developmental stage (Fig. 5E). By analysing the distribution of H3K27me3 and H3K27ac at regulatory regions, we discovered that those chromatin regions associated with inactive genes had a higher coverage of H3K27me3 and lower coverage of H3K27ac at shared Ubx/Pho binding regions, while it was the opposite for active genes (Fig. 5I). Furthermore, the canonical Pho binding motif was found overrepresented only among the Ubx/Pho/H3K27me3 chromatin regions associated with genes inactive in the mesoderm (Fig. 5H), while other DNA binding motifs, including the ones for Ubx, Exd and Trl, were overrepresented in Ubx/Pho/H3K27me3 chromatin regions associated with inactive as well as active genes.
These results demonstrated that Ubx interacted with Pho and that their interaction on H3K27me3 marked chromatin regions occurred preferentially at inactive genes in mesodermal cells.

Ubx is required for stabilizing Pho binding to H3K27me3 chromatin regions
Our results indicated that Ubx could mediate the repression of alternative fate genes by recruiting or stabilizing Pho binding at Ubx targeted genomic regions in the mesodermal lineage, thereby allowing the two proteins to act in a combinatorial fashion. In support of this hypothesis, we found 55% (769/1391) of the genes upregulated in the absence of Ubx, which were highly enriched for non-mesodermal and stem cell related functions, to be located in the vicinity of Ubx/Pho binding regions ( Supplementary Fig. 6B). In contrast, only a minor fraction (15%) was in the vicinity of chromatin sites targeted by Pho only, and we did not detect any significant enrichment of GO terms among these genes ( Supplementary Fig. 6B). To provide additional evidence for a combined action of Ubx and Pho, we analysed the binding of Pho to Ubx/Pho/H3K27me3 binding regions in the absence of Ubx by performing ChIP experiments on INTACT sorted control and Ubx depleted mesodermal nuclei.
Furthermore, we also quantified the levels of H3K27me3 and H3K27ac marks at these loci. We selected five Ubx/Pho/H3K27me3 genomic regions that changed their histone marks towards activation (less H3K27me3 and/or more H3K27ac) when Ubx was depleted in the mesodermal lineage, which resulted in the activation of the associated genes encoding neuronal functions. This included ventral nervous defective (vnd), a NK2 class TF encoding gene critical for patterning of the neuroectoderm as well as the formation and specification of ventral neuroblasts 65 , HGTX, a homeodomain TF encoding gene promoting the specification and differentiation of motor neurons innervating the ventral body wall muscles 66 , Neurotrophin 1 (NT1), a cytokine encoding a gene that regulates motor neuron survival and axon guidance 67 , hamlet (ham), a gene encoding a PRDM class TF that regulates neuron fate selection in the peripheral nervous system 68 , and Ptx1, again a homeodomain TF encoding gene which is expressed at high levels in the early embryonic central nervous system and the midgut, and later also in ventral muscles 69 . In addition, we also chose five loci bound by Pho but not by Ubx, which were associated with genes normally not expressed in the mesoderm and which remained silent in the absence of Ubx, including folded gastrulation (fog), Olig family (Oli), bendless (ben), myospheroid (mys) and medial glomeruli (meigo). Strikingly, all five Ubx/Pho/H3K27me3 loci, which were all co-bound by Ubx and Pho in the presence of Ubx (Supplementary Fig. 5A), experienced a dramatic loss of Pho binding when Ubx was degraded (Fig. 6B). In contrast, Pho binding to Pho-only control loci remained unaffected (Fig. 6C). As Pho expression levels were unaltered in the absence of Ubx ( Supplementary Fig. 5B), we concluded that Ubx is required to stabilize Pho binding to chromatin of H3K27me3 marked loci.
These results indicated that the expression of alternative fate genes should be similarly de-repressed in the mesoderm in the absence of either Ubx or Pho. In line with this hypothesis, we found the expression of vnd and Ptx1, which were normally not or only weakly expressed in mesodermal cells of stage 16 control embryos (Fig. 6D, E) to be significantly increased in the mesoderm of Mef2>Nslmb,GFP-Ubx embryos (Fig. 6F It has been recently reported that in the absence of Pho H3K27me3 enrichment was decreased 63,70 , and consistently, we found that H3K27me3 levels at the HGTX and ham associated Ubx/Pho/H3K27me3 loci were significantly decreased in Ubx depleted mesodermal nuclei (Fig. 6B). Interestingly, while H3K27ac levels remained unaffected at the HGTX associated Ubx binding locus, they were increased at the ham associated Ubx/Pho/H3K27me3 region (Fig. 6B). In contrast, H3K27me3 levels at Ubx/Pho/H3K27me3 loci associated with the vnd, NT1 and Ptx1 genes were not considerably decreased, however, these loci had significantly higher H3K27ac levels in the absence of Ubx (Fig. 6B). These results revealed that in the absence of Ubx Pho's ability to interact with the five Ubx/Pho/H3K27me3 loci was strongly reduced and concomitantly H3K27me3 as well as H3K27ac levels were changed. Thus, we asked whether H3K27me3 and H3K27ac levels were generally altered at Ubx/Pho binding sites when Ubx was depleted in the mesodermal lineage. To this end, we calculated H3K27ac/H3K27me3 ratios at all Ubx/Pho/H3K27me3 loci as well as at loci only bound by Pho. Intriguingly, we found that H3K27ac/H3K27me3 ratios were significantly higher at Ubx/Pho/H3K27me3 loci in Ubx depleted mesodermal nuclei when compared to control nuclei, while they were not significantly changed at Pho-only sites (Fig. 6J).
In sum, these results demonstrated that Ubx stabilized Pho binding to chromatin regions at alternative fate genes in the mesodermal lineage, which controlled the proper levels of H3K27ac to H3K27me3 at these sites, thereby ensuring repression of these genes.

DISCUSSION
Cell and tissue types get different during development and their identity needs to be maintained also in adulthood to guarantee the survival of organisms. Here, we provide evidence that multi-lineage TFs of the Hox class stabilize the different lineage choices by restricting cellular plasticity in a lineage specific manner. To this end, we studied the broadly expressed Hox TF Ubx in the mesodermal and neuronal lineages during Drosophila development using a comparative genomic approach and an experimental system to deplete Ubx protein exclusively in the embryonic mesodermal lineage. This approach allowed us to dissect the cell-autonomous function of Ubx in a single lineage that was located in an otherwise normal cellular environment at the transcriptome and chromatin level. Using this experimental set-up, we found that Ubx comprehensively orchestrates the transcriptional programs of the mesodermal as well as of the neuronal lineage, as it bound and regulated a substantially fraction of genes specifically expressed in these tissue lineages. Strikingly, this analysis revealed that the majority of Ubx chromatin interactions were located in the vicinity of inactive genes, and lineagespecific interference with Ubx in the mesoderm demonstrated that about 20% of these interactions were important for repressing the close-by genes. Intriguingly, these genes were highly enriched for alternative cell fates, demonstrating that Ubx had indeed a pivotal role in restricting developmental plasticity in a context-dependent manner.
One important question arising from this result was why not more of the inactive genes bound by Ubx were de-repressed in the absence of Ubx. There are several explanations for this finding. First, Hox TFs cross-regulate each other's expression, a phenomenon described as posterior suppression 71,72 . Consistently, the Hox gene Antennapedia (Antp), which is normally expressed anterior to Ubx, is ectopically activated when Ubx function is absent ( Supplementary Fig. 7), allowing Antp now to partially take over the function of Ubx in this lineage. Second, gene regulation is tightly linked to the chromatin status at promoter and enhancer regions. It had been shown in mammalian cells that the turn-over rates of the histone variant H3.3 at regulatory regions were correlated with specific histone modifications, high turn-over when associated with high levels of active histone modifications, like H3K27ac, while much slower turn-over when associated with higher levels of H3K27me3 marks 70 . This implies that the "clearing" of repressive histone modifications takes much longer in comparison to active ones. Third, H3K27me3 marks serve as epigenetic memory to permanently silence genes in the course of development, and a recent study demonstrated that a resetting of the epigenetic status requires cell division to dilute the H3K27me3 mediated silencing effect 73 . In line with these studies, we found that the reduction of H3K27me3 levels was low in comparison to the increase of H3K27ac levels in Ubx depleted mesodermal nuclei. Thus, a de-repression of genes might require either more time or cell divisions or both. However, after its specification at embryonic stage 12 the mesoderm does not divide anymore, which might prevent an efficient clearing of H3K27me3 marks at Ubx targeted chromatin sites. Fourth, derepression of genes is not only a consequence of abolishing repression but also of gaining activation, which not only requires a change of the histone environment at control regions but also the expression and action of the proper sets of TFs. Comparing the transcriptomes of the mesodermal and neuronal lineages revealed that only about 5% of the TFs expressed in an alternative lineage, in this case the neuronal one, were expressed in the Ubx depleted mesodermal cells, including vnd, HGTX, ham and Ptx1, which was obviously not sufficient to induce a lineage switch, in particular as the expression of only about 10% of the mesodermal fate related TFs was reduced in their expression. As we find Antp to be ectopically expressed in the mesodermal lineage in the absence of Ubx ( Supplementary Fig. 7), we assume Antp to partially take over the lineage-specific function of Ubx. Thus, it will be interesting in future to study mesoderm development in a Hox-free environment and determine the fate of the developing cell lineage.
Another highly relevant finding of our study is that the Hox TF Ubx lineagespecifically repressed the transcription of alternative fate genes by organizing the epigenetic landscape and that chromatin changes at Ubx sites were dependent on the interaction of Ubx with the Polycomb recruiter Pho. Strikingly, our study revealed that Ubx was crucial for stabilizing binding of the PcG protein Pho to specific chromatin regions. Although it is known that DNA binding TFs other than Pho interact with PREs, for example Grainyhead 74,75 , a TF that has been shown recently to lineage-specifically displace nucleosomes at enhancers 76 , these studies were mostly performed in vitro and the role of these TFs in Pho chromatin targeting was not addressed. In contrast, we analysed the combinatorial interaction of Ubx and Pho on the chromatin in vivo, in the mesodermal lineage of Drosophila embryos. Importantly, we showed that Pho was no longer able to interact with Ubx bound sites when Ubx protein was depleted in a lineage-specific manner, while Pho-only chromatin interactions were unaffected. As a result, histone modifications changed towards higher H3K27ac/H3K27me3 ratios, resulting in the de-repression of genes primarily associated with Ubx-Pho chromatin regions. This result indicated that not only Pho but very likely the whole PcG protein complex was not properly targeted to Ubx chromatin sites, which is in line with our finding that H3K27me3 levels were generally decreased at enhancers of genes derepressed upon Ubx depletion (Fig. 7). However, we also found H3K27ac levels increased at Ubx chromatin interactions located in the vicinity of de-repressed genes.
Interestingly, PREs frequently co-localize with response elements for Trithorax group (TrxG) proteins 52 , a large group of proteins first described for their role in transcriptional activation. In this line, we identified not only the binding motif for Pho but also for the GAGA TF Trithorax-like (Trl) enriched among the Ubx bound and H3K27me3 marked chromatin sites, and we found Trl to interact with Ubx in cellulo (Supplementary Fig.   5D). Trl was initially discovered to activate transcription, in particular of the Ubx gene 77 , but is also required for transcriptional repression, as it binds PREs 78 and physically associates with the Polycomb Repressive Complex 1 79 . Interestingly, Trl had been suggested recently to function as a pioneer factor in early Drosophila development by making genomic regions accessible through the deposition of active histone marks. Furthermore, it had been shown that another TrxG protein, the methyltransferase Trithorax (Trx), mediates together with the p300/CREB-binding protein (CBP) the deposition of H3K27ac marks 80,81 , a histone mark that was enriched at promoters and enhancers upon Ubx depletion. Thus, the TrxG complex and CBP could be recruited in the absence of Ubx to activate gene expression (Fig. 7). GAGA factors like Trl do not only play a role in the activation but is also in the repression of gene expression by interacting with the Polycomb complex 78,82 . Interestingly, a recent study showed that GAGA factors are required for the formation of repressive chromatin loops in Polycomb domains to stabilize gene silencing during early Drosophila development 83 . These results could explain one puzzling result of our study, the increase of H3K27me3 marks in the absence of Ubx, which could be due to Trl interacting with these sites when Ubx levels drop (Fig. 7), thereby promoting or increasing the formation of repressive chromatin loops. In future, it will be important to study TrxG as well as PcG proteins at Ubx targeted control regions to understand how Hox TFs orchestrate the interplay between transcriptional activation and repression in the course of lineage development when cellular plasticity needs to be restricted.
One peculiar feature of Hox TFs is that they are not only active during the development but their input is continuously required throughout the lifetime of an organism to assess the positional values of cells and to maintain their proper identities and functions [84][85][86][87][88][89][90] . Our findings now shed new light on the mechanism ensuring this stability in cellular identity, as they suggest that Hox TFs robustly repress alternative lineage programs (via the interaction with epigenetic factors like Pho) and reliably restrict the plasticity also of adult cells. In the normal context, this is absolutely critical for an organism to function properly, however, in some instances it can be a major hurdle, in particular when cells need or should adopt a new identity, which requires them often to regain plasticity. In organismal life, this happens mostly during dedifferentiation, regeneration and tissue remodelling [91][92][93] . Another situation of cell type conversion is cellular reprogramming, which is extensively studied due to its high potential as therapeutic strategy 94 . But although reprogramming strategies have been improved over time, the direct conversion of one somatic cell type into another one, the so-called transdifferentiation, is still inefficient 6,95 . Interestingly, it has been reported recently that the induction of the Hox code typical for a differentiated cell type in pluripotent stem cells (PSCs), either in combination with other factors 96

Fly stocks and husbandry
For the INTACT method 23

Generation of endogenously tagged Ubx
The GFP-Ubx 3.005 line was generating by using the CRIPR/Cas9 system 45,46 . For the donor DNA a pUC-MCS-5'GFP-MCS was designed, multiple cloning sites flanking the GFP containing an ATP start codon. For the 5' region from the GFP a homologous arm 104 containing the 5'UTR was cloned by using NotI and KpnI restriction sites. The 3' region from the GFP included the first Ubx exon and a large part of the intron for homologous recombination was cloned with BglII and XhoI. The gRNAs were designed to eliminate the first Ubx exon, positioned at the beginning of the 5'UTR and the end of the coding region of the first exon. The excised exon was replaced using the donor DNA and homologous recombination. The microinjection was performed by BestGene using vas-Cas9 (BL51323) as injection line and the resulting progenies (F0) were crossed with TM3/TM6 balancers and resulting F1 was used for single crosses against TM3/TM6 balancers to generate independent stocks. The F1 generation was screened by PCR for the presence of the GFP and the GFP containing stocks of the F2 generation were visually screened for the Ubx patterned GFP expression in vivo.

Generation of the Ubx antibody
The Ubx antibody was generated using the pGEX-purification system (GElifesciences).
The open reading of Ubx-RA was cloned in the pGEX-6P-2 vector using BamHI and XhoI restriction site. The protein was purified according to the protocol (GElifesciences) and eluted by using the PreScission Protease site. The immunisation and antibody handling was performed by the Charles Rivers company.

Purification of affinity-tagged nuclei, ChIP, ChIP-Seq and RNA-Seq
The nuclei were purified as described in Steiner et al. 23

Interactive data mining tool
The enrichment analysis method presented in this paper is implemented as a userfriendly Shiny web-application accessible via http://beta-weade.cos.uni-heidelberg.de..      PcG binding (dark blue) to alternative fate genes (grey: ectodermal genes, yellow: gut related genes) to ensure their lineage-specific repression. Histone tri-methylation marks (H3K27me3, red, ME) are set and maintained. Histones are illustrated in light blue and H3K27ac mark is Pho binding to regulatory regions of alternative fate genes (example for neuronal (purple)), which could have different outcomes: (1) The PcG complex might be further maintained at these sites through Trl (GAGA), an interactions potentially leading to an increase of H3K27me3 levels.
(2) Destabilisation of Pho and PcG at the regulatory regions of neuronal genes (purple). (3) Loss of PcG might increase TrxG binding through Trl (GAGA) in combination with CBP. This interaction could mediate the acetylation of Lys27 at histone 3 (H3K27ac, orange), resulting in the activation of alternative fate genes (example for neuronal (purple)).