Prep1 prevents premature adipogenesis of mesenchymal progenitors

Transcriptional regulators are crucial in adipocyte differentiation. We now show that the homeodomain-containing transcription factor Prep1 is a repressor of adipogenic differentiation since its down-regulation (DR) in both ex vivo bone marrow-derived mesenchymal stromal cells (MSC) and in vitro 3T3-L1 preadipocytes significantly increases their adipogenic differentiation ability. Prep1 acts at a stage preceding the activation of the differentiation machinery because its DR makes cells more prone to adipogenic differentiation even in the absence of the adipogenic inducers. Prep1 DR expands the DNA binding landscape of C/EBPβ (CCAAT enhancer binding protein β) without affecting its expression or activation. The data indicate that Prep1 normally acts by restricting DNA binding of transcription factors to adipogenic enhancers, in particular C/EBPβ.

Prep1 reaches the nucleus and binds DNA only as a dimer with Pbx1 16,17 . Indeed, Prep1 cooperates with Pbx1 to regulate Hox gene expression 12,14,18 . We recently described the Prep1 and Pbx1 binding sites to the DNA of mouse embryos, embryonic stem cells and embryonic fibroblasts and showed that Prep1 mostly, but not exclusively, binds DNA as a dimer with Pbx1 and that two thirds of the binding sites are cell-type specific 12,[19][20][21] .
Prep1 is also involved in cell differentiation as indicated by the inability of Prep1 i/i B and T cells to properly differentiate 14,22 , although the mechanisms are still unknown.
In this paper we report that in both ex vivo bone-marrow derived MSCs and in in vitro 3T3-L1 cells, Prep1 DR, similar to the effect of Pbx1 DR 11 , results in activation of adipogenesis. Interestingly, Prep1 DR triggers adipogenic differentiation even in the absence of complete hormonal induction, since Prep1 down regulated cells are already poised towards adipogenesis before the addition of the inducing hormones. Prep1 DR results in the expansion of the C/EBPβ DNA binding landscape already before induction and its ability to bind DNA is increased after the addition of the adipogenic inducers. Prep1 does not control a specific "differentiating" mechanism, but rather predisposes the cells towards adipogenic differentiation; therefore its possible role appears to be that of restricting DNA binding of transcription factors, in particular C/EBPβ, for example as a chromatin remodeling factor.

Results
Prep1 Down Regulation stimulates adipogenic differentiation. As a first step to test the role of Prep1 in adipogenic differentiation we used ex vivo bone marrow-derived MSCs. Initially, we analyzed Prep1 expression, both at the transcriptional and protein levels, in undifferentiated MSCs and at different time points following adipogenic induction, using MesenCult ™ Adipogenic Stimulatory Supplement (Stem Cell Technologies). Prep1 protein is present in wild type MSCs, but its expression level rapidly decreases after 3 days of differentiation (Fig. 1A) while its RNA level remains unchanged (Fig. 1C). We investigated the effects of Prep1 DR in vivo by using a hypomorphic murine line in which the levels of Prep1 protein are 2-20% of the wt levels 14 . Therefore, we set up MSC cultures, from both Prep1 i/i and wt mice, studied their growth and differentiation ability, following adipogenic induction. Undifferentiated Prep1 i/i and wt cells, collected from n = 8 mice per genotype, did not show differences in terms of morphology and growth kinetics in culture. In contrast, upon adipogenic induction, we observed in 4 out of 8 bone marrow samples that Prep1 i/i MSCs differentiated much faster than Prep1 +/+ control cells, as evidenced by a faster lipid droplet accumulation. Seven days after induction the  Supplementary Fig. S1. (B) Oil-red O staining of MSCs from wild-type (wt) or Prep1 hypomorphic (Prep1 i/i ) mice bone marrow 7 or 14 days after addition of the differentiation cocktail. Quantification of Oil-red O staining by absorbance at 490 nm is shown. (C) qRT-PCR analysis of Prep1, Pparγ, Adipoq and Glut4 expression in MSCs from wild-type (wt) or Prep1 hypomorphic (Prep1 i/i ) mice before (uninduced, UN) and 3 days after (Dif (3d)) induction of adipogenic differentiation. Each graph is representative of 3 independent experiments. Level of significance is indicated as follows *p ≤ 0.05, **p ≤ 0.01. adipogenic differentiation of Prep1 i/i cells was almost complete, while much less Prep1 +/+ control cells contained lipid droplets (mature adipocytes) (Fig. 1B). Quantification of lipids by Oil-red O staining at terminal differentiation (14 days) showed a two-fold (1.67-fold) higher level in Prep1 i/i than in control cells (Fig. 1B). Both the observation that Prep1 expression decreases during wt adipogenic differentiation and the data showing effects of Prep1 ex vivo DR suggest that Prep1 plays a negative regulatory role in the adipogenic program. In addition, we correlated the expression of Prep1 and selected markers of adipogenic differentiation in wt and mutant MSCs, before and after adipogenic induction (3 days). As expected, we detected a significant reduction of Prep1 expression in undifferentiated Prep1 i/i MSCs, as compared to wt cells (Fig. 1C). Moreover, Prep1 RNA levels of both wt and mutant cells remain virtually unchanged between undifferentiated and early differentiated cells. Consistently, we observed that the expression of critical mediators of adipogenesis is higher in Prep1 i/i cells, already at early time points after adipogenic induction (Fig. 1C). Interestingly, the mRNA levels of the early adipogenic markers Pparγ (p = 0.035) and adiponectin (Adipoq; p = 0.05) are significantly increased in Prep1 i/i MSCs, even in the absence of adipogenic induction, suggesting that Prep1 DR itself favors differentiation of MSCs towards an adipogenic fate. All the data obtained indicate that indeed Prep1 i/i MSCs differ from their normal counterparts already at the undifferentiated state.
We observed these results only in MSC cells derived from 50% of the Prep1 i/i bone marrows (n = 8). This is in perfect agreement with the non-penetrant embryonic and adult phenotype of these mice, as only 75% of the embryos die in utero and only 50% of the adult Prep1 i/i mice develop tumors 14,23 .
Given the above observations obtained in Prep1 i/i mesenchymal stromal cells and the intrinsic high levels of heterogeneity of MSCs, we decided to adopt a more homogeneous cell system to further investigate Prep1 function in the adipogenic process. In particular, we adopted the 3T3-L1 cell line frequently used as a model of in vitro adipogenesis, and infected it with a lentivirus vector carrying a scrambled control virus (henceforth referred to as C cells) or two different Prep1 shRNAs (named P and P2 cells, respectively) ( Supplementary Fig. S2A). We verified that Prep1 expression level is maintained at a down-regulated status at least for three days after induction ( Supplementary Fig. S2B). Figure 2A,B show that six days after induction, P cells presented at least twice more lipid droplets than C cells (p < 0.01). Similar results were obtained using the second shRNA2 ( Supplementary  Fig. S2C). The mRNA levels of the early adipogenic markers Pparγ1 and Pparγ2 are significantly increased (Fig. 2C). Therefore, a negative effect of Prep1 on adipogenesis was evident from both ex vivo derived MSCs and the already committed preadipocyte cell line 3T3-L1.
Prep1 DR activates core set of adipogenic genes at a pre-induction stage. In order to explore at the transcriptional level the mechanism leading to the enhanced differentiation of P cells, we performed RNA-seq in C and P cells before (day −2) and 24 hours after induction of differentiation (accessible in GEO at the accession number GSE100057). An expression ratio >2 or <0.5 was used as threshold to identify up-and down-regulated genes, respectively. In C cells, the adipogenic-inducing cocktail affected the expression of 2,800 genes after the first 24 hours of differentiation (Table 1). In fact, in P cells the total number of affected genes was lower because ~700 genes were already up-(446 genes) or down-regulated (259 genes) with respect to C cells before the induction (Table 1).
A heatmap with hierarchical clustering (Fig. 2D) showed the ratios between the RPKM (reads per kb per million aligned reads) of each gene versus the average of two control replicas at time −2 days. We included only those genes whose expression ratio was >2 or <0.5 and Benjamini-Hochberg adjusted p-value < 0.1. This analysis showed that a subset of the genes regulated by the inducing hormones in C cells, is already affected in the same direction in P cells before differentiation, i.e. in the absence of the hormones. This strongly suggests that endogenous Prep1 normally prevents adipogenic differentiation by some mechanism that can be counteracted by the inducing hormones.
The list of genes that are similarly affected by Prep1 DR at day −2 and by induction of adipogenic differentiation of control cells at day 1, is shown in Supplementary Table S1. Gene Ontology (GO) shows a statistically significant enrichment for the gene categories highlighted in Fig. 2D and listed in Supplementary Table S2. The genes up-regulated in P cells belong to GO categories such as response to cAMP (p < 0.0004), response to glucocorticoid (p < 0.005), fat cell differentiation (p < 0.02). These data are in agreement with the observed phenotype displaying higher adipogenesis.
Overall, these results show that in 3T3-L1 cells Prep1 DR induces a major alteration in the gene expression pattern even before the adipogenic induction confirming the hypothesis that Prep1 acts as an inhibitor for the adipogenic differentiation program.

Prep1 DR affects pathways induced by all three adipogenic inducers. Given the observed changes
in gene expression following Prep1 DR, one would expect that Prep1 may not specifically affect the action of a single adipogenic inducer. Therefore, to test such hypothesis, cells were incubated in media containing only one of the hormones needed for differentiation, i.e. insulin (Ins), dexamethasone (Dex) or IMBX 24 starting two days after reaching confluence and the expression of key adipogenic markers was followed for 3 days. Pparγ1 and Pparγ2, two important mediators of adipogenesis, were induced more strongly in P rather than C cells also in the presence of Ins or Dex only. The same was observed for C/EBPα (Fig. 3A). Pparγ was even slightly induced in presence of serum only (FCS). The smaller increase of Pparγ after the addition of only IBMX may be due to the absence from the nucleus of the glucocorticoid receptor which must be pre-bound to chromatin for IBMX full activation 25 . Normalization was accomplished by assessing Vinculin expression. Similar results were observed also upon partial combinations of the adipogenic inducers.
We have tested the effect of Prep1 DR on other known targets of the adipogenic inducers. The insulin pathway appeared to be activated in P cells as shown by the strong and early phosphorylation of Insulin Receptor Substrate 1 (pIrs1; Fig. 3B) at already 5 min after induction with respect to C cells, that show major phosphorylation after 40 min. Such faster phosphorylation of Irs1 is accompanied by a stronger phosphorylation of Akt kinase at both Thr 308 (Fig. 3C) and Ser 473 ( Supplementary Fig. S6A). In addition, the increased expression levels of Sterol Regulatory Element Binding Transcription Factor 1 (Srebf1) in P cells (Fig. 3D) are consistent with activation of the Akt pathway 26,27 .
In the glucocorticoid pathway, dexamethasone-dependent silencing of Pref1 was faster in P rather than C cells, as shown by its dramatic inactivation at 72 hours (Fig. 3E). Finally, also the cAMP pathway showed earlier and stronger activation. Indeed, even though two very early players of the pathway, Klf4 and Egr2 (aka as Krox20) do not change their expression at the RNA level in P as compared to C cells ( Fig. 3F and Supplementary  Fig. S6B), conversely, Klf5, a transcriptional regulator of Pparγ2 and a downstream target of C/EBPβ, is significantly up-regulated in P as compared to C cells, already a few hours after induction (Fig. 3G,H). Figure 3I also shows that the RNA level of Cebpd is significantly increased in P cells, as compared to control cells; a similar behavior is shown by the protein (Supplementary Fig. S6C). These results are also confirmed by the RNA-seq data Clustering heatmaps of gene expression changes assayed by RNA-seq from P and C cells before differentiation (−2 days) and control cells 1 day after differentiation (1 day), where n = 704 represents the number of the genes that changed their expression. The logarithm of ratios for each normalized RPKM is shown. Gene ontology (GO) terms of 3T3-L1 genes enriched by both differentiation of control cells (from day −2 to day +1) and by Prep1 down-regulation at day −2 are shown.

Effect of Differentiation in P cells* Number of genes.
Comparison of P v. C cells before differentiation $ Number of genes. The data refer to the number genes up-(>2x) or down-regulated (>0.5x) as obtained by the RNA-seq data (padj < 0.1). *Differentiated C or P cells 24 hrs after the addition of the inducing cocktail are compared with non differentiated cells (two days before the addition of the inducing cocktail). $ Comparison between C and P cells 2 days before or 24 h after the addition of the inducing cocktail.

Comparison of P v. C cells 24 h after induction $ Number of genes.
(Supplementary Table S1) that show the increase of the Cebpd expression. Overall these data suggest that Prep1 is a general inhibitor of adipogenic differentiation.
In conclusion, it appears that Prep1 DR favors adipogenesis independently of the adipogenic inducer employed, activating essentially many of the adipogenic target genes, and acting at a very early stage preceding the induction of differentiation.
Prep1 DR strongly increases C/EBPβ binding to chromatin. C/EBPβ is an essential and very early transcription factor in adipogenesis. Hence, the effects of Prep1 DR might be explained by its increase or activation 28 . The expression of Cebpb, however, was not significantly changed upon Prep1 DR as shown by RNA-seq analysis (Supplementary Table S1) and qPCR ( Supplementary Fig. S6D). Since C/EBPβ must be phosphorylated in order to acquire its DNA-binding activity 29,30 , we checked whether the levels of total and phosphorylated C/ EBPβ isoforms (LAP*, LAP, LIP) were affected by Prep1 DR. However, both total and phosphorylated levels of C/ EBPβ (LAP is phosphorylated at Thr188, LIP is phosphorylated at Thr37) were equally affected in P v. C cells, at all times tested after the induction (Fig. 4A).
Since C/EBPβ is the main player in the formation of multifactor complexes recruited on enhancers in chromatin hotspots already a few hours upon induction of differentiation 4 , we analyzed the effect of Prep1 DR on the binding landscape of C/EBPβ in 3T3-L1 cells comparing specific ChIP-seq at day −2 and 4 hours after induction in P v. C cells (accessible in GEO at the accession number GSE100057). The specificity of the C/EBPβ antibody used for ChIP was checked by western blot and proved that C/EBPβ is properly immunoprecipitated (Supplementary Fig. S9). Four hours after induction corresponds to the formation of transcription factor hotspots during adipogenesis 9 . The data obtained through our assays are reported in Table 2. Venn diagrams illustrating the effect of Prep1 DR on C/EBPβ DNA binding are shown in Fig. 4B. Prep1 DR induced a 2-fold increase of highly significant DNA-binding sites (p < 10 −5 ) in P versus C cells before differentiation at day −2, and a further 2-fold increase 4 hours after induction of differentiation ( Fig. 4B and Supplementary Fig. S10). The increase was distributed among all types of peaks, i.e. transcription start site associated (TSSA), intragenic (IG), close (CI)and far-intergenic (FI) ( Table 2), but their distribution among these categories was not affected. As C/EBPβ is the common transcription factor recruited in the hotspots established at 4 hours after the induction 4 , we tested the correspondence between the C/EBPβ peaks and the hotspots. Table 2 shows that 54% of the C/EBPβ peaks in differentiated control cells are present within hotspots; however, this fraction is increased to 74% in differentiated P cells. We conclude, therefore, that Prep1 DR strongly increases C/EBPβ binding to chromatin; this may occur because the absence of Prep1 provides a genomic environment better suited for C/EBPβ binding to DNA.  Regulatory elements regulate gene expression during adipogenic differentiation of 3T3-L1 cells even when they are Mbs away from a corresponding promoter 6 . We associated each peak located closer than 20 kbs to a transcription start site (TSSA), or located intragenically (IG), to the corresponding gene and found that the number of C/EBPβ-bound genes was also significantly increased in P cells both before and after differentiation (Fig. 4C). Moreover, at time day −2, i.e. before differentiation, GO analysis of the 544 genes bound by C/EBPβ exclusively in P cells (Fig. 4C) was enriched for such GO terms as "Regulation of fat cell differentiation" (p < 10 −6 ) and "Positive regulation of fat cell differentiation" (p < 10 −4 ). The same GO categories and few others related to adipogenesis (Fat cell differentiation, White fat cell differentiation) were enriched at 4 hours after induction in P cells; however, these genes were bound by C/EBPβ also in C cells.

Differentiated P cells (+4 hrs)
The distribution of the enriched GO categories between P and C cells is shown in Supplementary Table S3, which also shows an overall increase in all peaks' categories. It is important to notice that virtually none of the genes are exclusively bound by C/EBPβ in control cells before induction. We also checked whether the expression of these genes is changed at day −2 using our RNA-seq data. While more than half of C/EBPβ-bound genes from "Fat cell differentiation" category in P cells was up-regulated, none of the 6 control-specific genes changed their expression (data not shown). This is in agreement with the possibility of Prep1 regulating C/EBPβ access to DNA and hence allowing premature induction of expression of adipogenic genes.
We also analyzed the ChIP-seq C/EBPβ DNA binding motifs using MEME software. While Prep1 DR did not affect C/EBPβ binding to the canonical site 9 , the binding to the additional AP1 consensus sequences was greatly reduced (from 35% to less than 15%) in peaks exclusively bound in P cells (Fig. 4D). This is in line with Prep1 DR regulating C/EBPβ DNA-binding efficiency.

Discussion
In this paper we have shown that Prep1 DR in 3T3-L1 preadipocytes and in ex vivo Prep1 hypomorphic MSCs predisposes to adipogenic differentiation (Figs 1 and 2). First, and most importantly, the effect of Prep1 DR, i.e. the activation of some markers of adipogenic differentiation, precedes the addition of the inducers. Moreover, Prep1 DR did not affect one single pathway activated by insulin, glucocorticoids or cAMP, i.e. the inducers of in vitro differentiation. Finally, the major effect of Prep1 DR was an increased binding of C/EBPβ to DNA which, however, was not explained by a direct effect on the level or activation of this protein.
Since Prep1 belongs to a family of proteins which are essential in embryonic development, it was initially considered as a direct regulator of development 16,31 . Although the Prep1-deficient phenotype, i.e. the lethality of the Prep1-deficient embryos in different organisms [13][14][15]18,32 , suggests a developmental function, the effects were never shown to be targeted on developmentally essential genes. Absence of Prep1 causes phenotypes that may be attributed to basic cellular activities, such as apoptosis and DNA damage 18 ; however, differentiation of several cell lineages was observed 13,14,33 , implying that differentiation mechanisms may be altered. Nevertheless, the analysis of the DNA binding landscape displayed by Prep1 did not detect any distinct involvement in development/ differentiation-related genes. Rather, the Gene Ontology analysis of the genes bound and regulated by Prep1 in E11.5 mouse embryos identified gene categories that belong to "basic cellular machineries or function" such as transcription regulation, DNA metabolism, signaling and chromatin modifications 12 . In addition, while in vivo Prep1 DR is not necessarily incompatible with life (Prep1 i/i hypomorphic mice show a partially-penetrant phenotype) 14,23 , induction of differentiation of Prep1 DR ES cells has shown dramatic effects 19 .
Therefore, we proposed that also the phenotype of Prep1 DR 3T3-L1 and of Prep1 i/i mesenchymal adipogenic progenitors may be considered under this angle. The phenotypes observed in both systems appear to be secondary to a fundamental effect, which is already in place before the induction of differentiation, and becomes evident only after the addition of the differentiation cocktail. We hypothesize that Prep1 DR modifies chromatin organization. For example, the increased binding landscape of C/EBPβ might be due to the presence of a more accessible, Prep1-dependent, chromatin architecture. As Prep1 is a regular transcriptional partner of Pbx1 and since Pbx1 DR also shows a similar phenotype to Prep1 DR, one wonders how much of the Pbx1 function is due to a specific differentiation function and how much is due to more basic cellular activities.
In conclusion, we have shown for the first time that Prep1 transcription factor is required for adipogenic differentiation of both a preadipocytic cell line and of ex vivo mesenchymal progenitor cells. It performs its action indirectly, inducing an extensive DNA binding of C/EBPβ at the early stages of the adipogenic process. The mechanism of this induction remains to be elucidated.

In vitro cultures of Mesenchymal Stromal Cells (MSCs) and 3T3-L1 cells.
Mice were sacrificed by cervical dislocation before collecting tibias and femurs. Bone marrow cells were flushed and seeded in culture using MesenCult Basal Medium supplemented with 20% Mesenchymal Mouse Stimulatory Supplement (StemCell Technologies) and 1% Pen-Strept (Life Technologies) (Complete Medium). Cells were grown at 37 °C in humidified atmosphere at 5% CO 2 . Medium was changed every 3 days and cells were trypsinized at confluence and reseeded at 2 × 10 4 cells/cm 2 (passage 1, p1). All experiments were performed at passage 2 (p2). Adipogenic differentiation. MCSs were seeded at p2 at 2 × 10 4 cells/cm2 and at 80% confluence the Complete Medium was replaced with Adipogenic Induction Medium (MesenCult Basal Medium supplemented with 20% AdipogenicStimulatory Supplement (StemCell Technologies) and 1% Pen-Strept (Life Technologies) we followed the manufacturer's recommendations.
For adipogenic differentiation of 3T3-L1 cells, two days after confluence the medium was changed to differentiation medium I containing 1 μM dexamethasone, 0.5 mM 3-isobutyl-1-methylxanthine, and 1 μg/ml insulin in DMEM containing 10% of FBS. Two days after the induction of differentiation the medium was changed do differentiation medium II containing only 1 μg/ml insulin in the presence of 10% of FBS. After 4 days of differentiation the medium was changed to differentiation medium III containing only 10% of FBS.
Oil Red O staining. Differentiated 3T3-L1 and MSC cells were stained by Oil-red O using a standard protocol 34 . Briefly, cells were fixed in 3.7% of formaldehyde in PBS for 30 min at room temperature and then stained in 0.3% of Oil Red O in 60% of isopropanol for 30 min at room temperature. For quantification assays, Oil Red O was extracted from the stained cells by using 100% isopropanol and absorption was measured at 490 nm.
To identify enriched domains we used MACS version 2.0.10.20131028 with default parameters 35 , except for p-value threshold (10e-5).The reads were mapped against the mm9 mouse genome. De novo motif discovery was run to identify consensus sequences enriched in the selected regions versus the whole genome using MEME-SUITE as de novo motif finder algorithm 36 .

RNA-seq.
For RNA-seq, total RNA was purified and the library prepared. For each time point (−2 days and 1 day) and sample (control and Prep1 down-regulated) we used 2 biological replicas. The sequencing was performed using Ion-Torrent system. After eliminating artifacts with FASTX-Toolkit v.0.0.13.2, reads were aligned to the mm9 reference using Tophat v2.0.9. Then GenomicFeatures, GenomicRanges and GenomicAligments R libraries were used to count the number of reads with respect to the annotation reference (UCSC, genome = 'mm9' , tablename = 'refGene'). Differentially expressed genes between different conditions were retrieved using DESEQ R library 37 , filtering the results using padj <0.1 as a threshold. Gene ontology analysis was performed using Gorilla software 38 .
The heat maps were drawn plotting the ratios between each of normalized RPKM versus the average of two control replicas at each time point and sample. qPCR and immunoblotting. RNA extraction was processed according to the RNeasy (QIAGEN) protocol.