Derangement of a Factor Upstream of RARα Triggers the Repression of a Pleiotropic Epigenetic Network

Background Chromatin adapts and responds to extrinsic and intrinsic cues. We hypothesize that inheritable aberrant chromatin states in cancer and aging are caused by genetic/environmental factors. In previous studies we demonstrated that either genetic mutations, or loss, of retinoic acid receptor alpha (RARα), can impair the integration of the retinoic acid (RA) signal at the chromatin of RA-responsive genes downstream of RARα, and can lead to aberrant repressive chromatin states marked by epigenetic modifications. In this study we tested whether the mere interference with the availability of RA signal at RARα, in cells with an otherwise functional RARα, can also induce epigenetic repression at RA-responsive genes downstream of RARα. Methodology/Principal Findings To hamper the availability of RA at RARα in untransformed human mammary epithelial cells, we targeted the cellular RA-binding protein 2 (CRABP2), which transports RA from the cytoplasm onto the nuclear RARs. Stable ectopic expression of a CRABP2 mutant unable to enter the nucleus, as well as stable knock down of endogenous CRABP2, led to the coordinated transcriptional repression of a few RA-responsive genes downstream of RARα. The chromatin at these genes acquired an exacerbated repressed state, or state “of no return”. This aberrant state is unresponsive to RA, and therefore differs from the physiologically repressed, yet “poised” state, which is responsive to RA. Consistent with development of homozygosis for epigenetically repressed loci, a significant proportion of cells with a defective CRABP2-mediated RA transport developed heritable phenotypes indicative of loss of function. Conclusion/Significance Derangement/lack of a critical factor necessary for RARα function induces epigenetic repression of a RA-regulated gene network downstream of RARα, with major pleiotropic biological outcomes.


Introduction
Retinoic acid (RA), the bioactive derivative of retinol, is a signal fundamental for developmental and cellular processes, whose intracellular physiological level is tightly regulated by a complex metabolic pathway involving both RA synthesis and RA catabolism [1,2]. RA exerts its biological action mainly by binding and activating specialized transcription factors, the RA-receptors (RARs) [3]. When RA is channeled onto the retinoic acid receptor alpha (RARa) in the nucleus, it can rapidly induce transcription of RARa-target genes containing a RA-responsive element (RARE). Specifically, RA binding to RARa triggers both the dissociation of corepressors proteins, and the recruitment of coactivators and histone modifying enzymes that enable chromatin conformation changes compatible with the access and action of RNA polymerase II [4,5].
The temporal dynamics of the cascade of events following RA-RARa-mediated chromatin activation has been mostly derived from studies on the prototypic direct RARa-target gene RARb2. Once expressed in response to RA, RARb2 sustains its own transcription by binding to its own promoter [6], and subsequently activates the chromatin of other downstream RA-responsive direct target genes [7,8]. In the absence of RA, RARb2 chromatin reaches a repressed state, which is however poised for transcription [4,5].
Previously, we demonstrated that when RA signal cannot be integrated at RARa, because RARa is either not expressed, or has acquired genetic mutations that make it non-functional, the chromatin associated with RARb2 falls into an aberrant exacerbated state of repression, which is unresponsive to RA [9]. Moreover, by using different cell systems, we demonstrated that the impaired integration of RA signal at a mutant RARa induces a repression wave that is propagated, in a domino fashion, from RARb2 to targets downstream of RARb2. Specifically, by using mouse embryocarcinoma cells, we found that a dominant negative RARa mutant creates a concerted repression of both RARb2 and its direct target CYP26A1, encoding the cytochrome P450 RA-specific hydrolase, which acts as a neuronal differentiation switch in these cells [8,10]. In an independent study using human mammary epithelial cells, we demonstrated that inhibition of RARa function with various genetic strategies triggers the concerted repression of both RARb2 and another target downstream of RARb2, CRBP1, encoding the cellular retinol binding protein 1, which is pivotal for breast epithelial cell acinar morphogenesis [7].
Based on the observation that the RARb2 chromatin can also be found aberrantly repressed in RARa-positive cancer cells [11], we hypothesized that lack/derangement of upstream factors capable of affecting RARa function is sufficient to induce aberrant chromatin repression at RARb2 and its downstream targets.
In the present study we show that the derangement of the cellular RA binding protein 2 (CRABP2), critical for the transport of RA from the cytoplasm to the RARs in the nucleus [12], can indeed trigger a long-distance chromatin repression effect at loci of an entire RARa-regulated epigenetic network. We found that, not only the knock down of endogenous CRABP2 by RNAi, but simply the mere interference of RA transport into the nucleus, achieved by expressing a dominant negative CRABP2 mutant unable to enter the nucleus [13], can initiate the wave of aberrant repression at the chromatin of multiple RA-responsive genes. The wave of repression involves first RARb2, thus affecting cell growth, and next branches downstream, to involve genes that control both RA metabolism/homeostasis and morphogenesis.
In conclusion, interference with RA transport at RARa into the nucleus is sufficient to induce coordinated, heritable, chromatin repression at multiple loci of a RA-responsive gene network downstream of RARa, with pleiotropic biological outcomes.

Results
Interference with RA transport into the nucleus is sufficient to induce transcriptional repression of genes downstream of RARa RARa activation requires the transport of RA to RARa in the nucleus by CRABP2 [12,14,15]. HME1 cells, which express both RARa (Fig. 1A left) and CRABP2 (Fig. 1A, right), can properly integrate RA signal through RARa, as demonstrated by the transcriptional activation of two prototypic RA-responsive genes, RARb2, a downstream RARa target, and CRBP1, a downstream RARb2 target (Fig. 1B, left and right).
To transport RA into the nucleus, CRABP2 requires a specific nuclear localization signal (NLS) [13]. A mutant CRABP2-KRK protein, which was shown to bind RA with affinity similar to the one of the wild type CRABP2 protein, cannot transport RA into the nucleus due to critical mutations in the NLS [13]. Indeed, by using immunocytochemistry, we found that the V5-tagged CRABP2-KRK protein, transiently expressed in HME1 cells, differently from the wild type CRABP2-V5 protein, is not able to enter the nucleus after addition of RA (0.1 mM, 30 minutes) (Fig. 1C).
Next, we tested whether RA transport into the nucleus is hampered in CRABP2-KRK-positive cells. Stable expression of the CRABP2-KRK-V5 protein in HME1 cells (shown for the KRK-15 clone in Fig. 1D, left), while not affecting the expression of endogenous RARa relative to the control clone EV7 (Fig. 1D, right), clearly exerts a dominant negative effect over the endogenous CRABP2. This conclusion is based on the observation that RA-induced transcriptional activation of both RARb2 and CRBP1 is reduced in the KRK-15 clone relative to the control clone EV7 (Fig. 1E, left and right). Thus, targeting CRABP2 function prevents RARa function and affects, in a negative and irreversible fashion, the transcriptional status of RA-responsive genes downstream of RARa.
Evidence of chromatin repression at RA-responsive genes downstream of RARa consequent to CRABP2 knockdown To test whether targeting endogenous CRABP2 in HME1 cells can indeed induce heritable aberrant repression of the chromatin at both RARb2 and CRBP1, we knocked down CRABP2 by stable RNA interference with either one of two CRABP2-targeting shRNA sequences, CRABP2-A and CRABP2-C (Fig. S1A). A scrambled (mock) shRNA sequence, which should not recognize any human mRNA, was used as a control (Fig. S1A). Only the shRNAs sequences directed against CRABP2 were shown to efficiently decrease exogenous CRABP2 protein expression (Fig. S1B).
We further tested two CRABP2 knock down clones, Si-CRABP2-A6, carrying the CRABP2-A sequence, and Si-CRABP2-C6, carrying the CRABP2-C sequence, along with the control clone Mock13, carrying the scrambled sequence (Fig.  S1C). Both Si-CRABP2-A6 and Si-CRABP2-C6 displayed a significant decrease of the CRABP2 transcript ( Fig. 2A, left), while they still expressed the RARa receptor ( Fig. 2A, right). RA failed to activate the transcription of both RARb2 and CRBP1 in both knock down clones (Fig. 2B, left and right).
Moreover, ChIP analysis with anti-acetylated histone H4 (Ac-H4) showed significant hypoacetylation, which remained unresponsive to RA, of the chromatin regions encompassing either the RARb2-RARE or the CRBP1-RARE (Fig. 2C, top and bottom). Apparently, the chromatin at both RARb2 and CRBP1 was converted from a state poised for transcription to an exacerbated repressed state unresponsive to RA, which could be reverted only by treatment with the HDAC inhibitor Trichostatin A (TSA) ( Thus, as a consequence of CRABP2 knock down, the chromatin of two loci downstream of RARa has acquired a repressed ''state of no return'', unresponsive to RA. This state, non-permissive for transcription, differs from the poised state, responsive to RA, which is permissive for transcription. Hampering CRABP2 function in HME1 cells leads to biological phenotypes that reflect homozygosis for epigenetically silent RARb2 and CRBP1 alleles We previously demonstrated that knock down of the tumor suppressor RARb2 in HME1 cells confers resistance to RAinduced growth inhibition [7] (Fig. 3A, left). Analysis of RAresistance by colony formation in HME1-derived clones with either ectopic expression of CRABP2-KRK (CRABP2-KRK15), or CRABP2 knock down (Si-CRABP2-A6 and Si-CRABP2-C6) (Fig. 3A, right) clearly indicated loss of RARb2 function. RAresistance is expected only in association with homozygous repression of the chromatin at RARb2 alleles, which are consequently non permissive (np) for transcription (Fig. 3B). Similarly, we previously demonstrated that CRBP1 knock down in HME1 cells hampers acinar morphogenesis in 3D culture [7] (Fig. 3C). We observed aberrant acinar morphology also in HME1-derived clones with either ectopic expression of CRABP2-KRK (CRABP2-KRK15) or CRABP2 knock down (Si-CRABP2-A6 and Si-CRABP2-C6) (Fig. 3C), thus indicating loss of CRBP1 function. Loss of proper acinar morphogenesis is expected only in association with homozygous repression of the chromatin at CRBP1 alleles, which are consequently non permissive (np) for transcription (Fig. 3D).
Interference with CRABP2 function apparently induces loss of both RA-induced growth inhibition and 3D-acinar morphogenesis in a significant fraction of cells, strongly suggesting the occurrence of heritable homozygous epigenetic silencing at both RARb2 and CRBP1 loci.
Evidence of CpG hypermethylation corroborates the occurrence of heritable epigenetic silencing at both RARb2 and CRBP1 consequent to deranged CRABP2 function DNA hypermethylation is an epigenetic and heritable modification. For this reason, we tested for DNA hypermethylation at RARb2 and CRBP1 in HME1 cells with deranged CRABP2 function. First, we found that treatment of Si-CRABP2-A6 cells with the demethylating agent 5-aza-29-deoxycitidine (5-Aza) could significantly restore RA-induced RARb2 and CRBP1 transcription (Fig. 4A, left and right). Then, we tested by quantitative methylation specific PCR (qMSP) whether RARb2 and CRBP1 regulatory regions in the CRABP2 knock down clones were indeed marked by DNA hypermethylation. For the detection of RARb2 methylated (M) alleles, we used primers previously shown to recognize the RARb2 methylation epicenter [9], while for CRBP1 we used primers recognizing the two regions, M1 and M2, within the CRBP1 CpG island that we demonstrated previously to be the first undergoing aberrant DNA methylation in cells with an impaired RARa signaling [7]. This analysis clearly shows that CRABP2 knock down clones A6 and C6 have significantly more RARb2 and CRBP1 methylated (M) alleles relative to the control clone Mock13 (Fig. 4B, left and right). The finding that RARb2 and CRBP1 silencing is associated with DNA hypermethylation, a well-established hallmark of aberrantly repressed chromatin, further reinforces our conclusion that the repressed state of RARb2 and CRBP1 chromatin in cells with deranged CRABP2 function is heritable, and therefore epigenetic.

Derangement of CRABP2 function exerts a chromatin repression effect branching downstream of RARb2
We previously demonstrated in a mouse embryonic carcinoma cell model that an endogenous dominant negative RARa mutant, lacking part of the E domain harboring the RA-binding domain, induced concerted epigenetic repression of both RARb2 and CYP26A1, encoding for a RA hydrolase involved in RA catabolism [8]. We reproduced this finding also in human cells carrying an exogenous dominant negative RARa mutant lacking the RA-binding domain (Fig. S2). Here we show that hampering CRABP2 function leads to significant CYP26A1 chromatin repression also in HME1 cells.
First, we found that impairment of CRABP2 function in HME1 cells by either CRABP2 knock down, or expression of the mutant CRABP2-KRK, leads to significant downregulation of RAinduced CYP26A1 transcription (Fig. 5A, left and right, respectively). CYP26A1 transcription is driven by a promoter region containing a proximal RARE at 287 and seems to be enhanced by an upstream region containing a distal RARE at 21973 [16,17]. ChIP analysis with anti-acetyl histone H4 shows that CYP26A1 downregulation in Si-CRABP2-A6 and Si-CRABP2-C6 clones is marked by histone deacetylation, which is unresponsive to RA, both in the region containing the distal RARE (data not shown) and in the region containing the proximal RARE (Fig. 5B,  left). Consistently, treatment with the HDAC inhibitor TSA could efficiently restore RA-induced CYP26A1 transcription in CRABP2 knock down clones (shown here for Si-CRABP2-A6 in Fig. 5B,  right).
Second, we tested whether the CYP26A1 repressed chromatin state, consequent to CRABP2 knock down, was also marked by DNA hypermethylation. By in silico analysis of the 59 regulatory regions of human CYP26A1, we identified two canonical CpG islands: one encompassing the distal RARE, and one encompassing the proximal RARE (Fig. S3). Bisulfite sequencing of these two regions showed that the proximal CpG island is fully methylated in the CYP26A1-negative cell line MDA-MB-231, while it is fully unmethylated in two CYP26A1-positive cell lines, T47D and HME1 (Fig. 5C, left). In contrast, the methylation status of the distal CpG island did not show any significant difference between HME1 and MDA-MB-231 (data not shown). Therefore, we focused our analysis on the proximal CpG island. By using qMSP with primers able to discriminate between the different methylation status of the control cell lines HME1, T47D and MDA-MB-231 (Fig. 5C, right), we found that the CRABP2 knock down clones have significantly more CYP26A1 methylated (M) alleles relative to the control clone Mock13 (Fig. 5D, left). Consistently, treatment with the demethylating agent 5-Aza could significantly restore RA-induced CYP26A1 transcription in CRABP2 knock down cells (shown here for Si-CRABP2-A6 in Fig. 5D, right).
Finally, we asked whether CYP26A1 epigenetic downregulation is consequent to, or concomitant with, the epigenetic downregulation of CRBP1, the other RARb2 target. We found that CYP26A1 transcription is still RA-inducible in HME1 cells knocked down for CRBP1 (Si-CRBP1) (Fig. 5E). Thus, CYP26A1 transcriptional downregulation, induced by hampering CRABP2 function, is consequent to a ''long distance'' repression effect, branching downstream of RARb2, and involving both CRBP1 and CYP26A1 chromatin (Fig. 6).

Discussion
In different cell systems, and using different mechanistic approaches, we previously demonstrated that an impaired RARa signalling, due to derangement/loss of RARa itself, confers an exacerbated repressed chromatin state, marked by repressive epigenetic changes at several RA-responsive genes downstream of RARa [7][8][9]. This study shows that hampering CRABP2, a factor critical for RA transport onto nuclear RARa, in cells with a functionally intact RARa, also leads to epigenetic repression of RA-responsive genes downstream of RARa, with heritable biological outcomes.
We provide evidence that derangement of CRABP2 function is sufficient to trigger the coordinated repression of the RARa direct target RARb2, and two RARb2 downstream targets, CRBP1 and CYP26A1. Specifically, in HME1 cells with functional RARa, we observed that not only the silencing of endogenous CRABP2, but the mere interference with CRABP2-mediated RA-transport into the nucleus, achieved by expressing the CRABP2-KRK protein with a mutated nuclear localization signal, induces heritable epigenetic changes at genes of a RA-responsive gene network downstream of RARa (Fig. 6).
Apparently, the abrogation of RARa function, be it due to RARa silencing/genetic mutations, or derangement of a factor upstream of RARa (e.g. CRABP2), results in the conversion of the chromatin of RARa-regulated genes from an inactive, yet poised, state permissive for transcription into an exacerbated repressed state that is non permissive for transcription. We refer to the latter state as the ''state of no return'', because it is marked by repressive epigenetic modifications, which remain unresponsive to RA [9]. This exacerbated, repressed state is distinct from the physiological repressed poised state, which is still responsive to RA. We still do not know what molecular mechanism(s) is capable of ''invoking'' the recruitment of chromatin repressive activities at RA-responsive genes downstream of RARa, once RARa function is impaired.
As a result of derangement of CRABP2 function, and consequent impairment of RARa function, we found evidence that cells develop homozygosis for epigenetically silent genes that are either RA-receptors (RARb2) or RA-responsive genes involved in both RA metabolism and morphogenesis (CRBP1 and CYP26A1). Specifically, we demonstrated that the homozygous epigenotypes for these repressed genes are heritable based on the analysis of biological and morphological phenotypes in HME1 cells either carrying the dominant negative mutant CRABP2 protein, or knocked down for CRABP2. Even when RARa was still expressed, we observed in a significant fraction of cells both RA resistance, indicative of loss of RARb2 function, and aberrant acinar morphogenesis, indicative of loss of CRBP1 function. The RA-resistant phenotype and the aberrant acinar morphology is expected to reflect only epigenotypes homozygous for repressed, non-permissive RARb2 and CRBP1 alleles, respectively. Consistently, in the same cells, we found evidence of aberrant CpG methylation, an epigenetic hallmark of repressed chromatin, at both RARb2 and CRBP1. The repressive repercussion due to derangement of CRABP2 function affects also the chromatin of CYP26A1, another RA-responsive gene downstream of both RARa and RARb2. Downregulation of CYP26A1 transcription is marked by both hypoacetylation unresponsive to RA and hypermethylation of the CpG island containing the proximal CYP26A1 RARE. Apparently, RARs and genes of the RA metabolism (CRBP1 and CYP26A1), are part of the same network. This RARa-regulated gene network is clearly implicated also in cell growth and cell morphogenesis. Further, this gene network can undergo concerted epigenetic repression as a consequence of derangement of factor(s) capable of interfering with RARa function.
In conclusion, this study reinforces the supposition that epigenetic repression in cancer cells may result from an ordered, rather than random, re-programming of the chromatin in response to intrinsic and extrinsic cues; which mirrors the order that underlies development [18].

Cell cultures
Cells. The human immortalized, non-transformed breast epithelial cell strain hTERT-HME1, here referred to as HME1, was grown in Mammary Epithelial Growth Medium (MEGM) plus bovine pituitary extract as per manufacturer's instructions (Lonza, Walkersville, MD). The HME1-derived clones knock down for RARb2 and CRBP1 have been described in [7]. The human breast cancer cell lines T47D and MDA-MB-231 (ATCC, Manassas, VA), and the T47D-derived clones DNC8 and LXC5, carrying the dominant negative RARa 403, or the cognate control vector, respectively [9], were cultured in Dulbecco's modified Eagle's medium (DMEM, Invitrogen), supplemented with 5% FBS (Invitrogen, Carlsbad, CA). Cells were all maintained at 37uC in 5% CO 2 and 85% humidity.
Three dimensional (3D)-cultures. HME1 cells and derived clones were grown on reconstituted basement membrane (Matrigel) to induce breast epithelial differentiation into acinilike structures, essentially as described [19]. Briefly, single cells were induced to form acini on chamber slides coated with growth factor-reduced Matrigel (BD Biosciences, San Jose, CA) in medium plus 2% matrigel for 10-15 days. After fixation with 4% paraformaldehyde for 20 min, permeabilization with Phosphate Buffered Saline (PBS) plus 0.1 % Triton X100 for 10 minutes, and blocking with PBS plus 1% BSA, 1% goat serum, 0.05 Tween 20 for 2 h, cells were incubated over night with both an antibody specific for the Golgi apparatus (anti-GM 130, 1:400, BD Biosciences), and an antibody for integrin (anti-CD49f, 1:200, Chemicon, Temecula, CA), followed by detection with goat antimouse Alexa Fluor 546 (1:400, Molecular Probes) and goat antirat Alexa Fluor 488 (1:400, Molecular Probes, Eugene, OR). Nuclei were counterstained with 300 nM DAPI (Sigma, St. Louis, MO). 30 acini, or more, per each clone were analyzed by confocal microscopy (SP2 Spectral Confocal Microscope, Leica, Wetzlar, DE) to inspect for the presence of a hollow lumen and apicobasal polarization. The morphology observed in 70% or more of the acini was considered to be the prevalent phenotype.
Colony formation assay. Exponentially growing cells were seeded at 3610 2 cells/well in 6-well plates and allowed to attach for 48 h. After treatment with either 0.1 mM RA or vehicle (ethanol) for 24 h, the medium was replaced with drug-free medium and cells were allowed to grow until the appearance of colonies was observed (10-14 days). Colonies fixed with methanol and stained with Giemsa were analyzed with Image J software (http://rsbweb.nih.gov/ij/) to establish the percentage of growth compared to the non-treated control (colony formation index). Statistical significance was calculated by Student's t-test on three independent determinations; p values at least ,0.05 were considered as significant.
Stable RNA interference (RNAi). The sequences CRABP2-A (59-CTG ACC AAC GAT GGG GAA C-39), CRABP2-C (59-GGT TGT CCC TGG ACT TGT C-39) (Gene Bank NM_001878, nucleotides 477-495, and 9-27 respectively) targeting CRABP2 mRNA, and the control mock sequence (59-ACG TAC GTA CGT AGT GGG G-39), which does not recognize any human mRNA, were cloned into the pSUPER-retro vector according to the manufacturer's instructions (Oligoengine, Seattle, WA). The silencing efficiency of the short hairpin RNAs (shRNAs) produced by these constructs was preliminary tested on exogenous CRABP2 transiently cotransfected with the shRNAs in COS cells as previously described [7]. The pSuper-CRABP2-A, pSuper-CRABP2-C, and pSuper-Mock constructs were stably transfected in HME1 cells by using Lipofectamine Plus (Invitrogen). Single stable clones were selected in puromycin 1 mg/ml, tested for the presence of the correct construct by PCR and sequencing, and further analyzed for the level of endogenous CRABP2 transcript by Real Time RT-PCR.

Drugs and treatments
All-trans-retinoic acid (RA) (Sigma, St. Louis, MO), 5-aza-29deoxycitidine (5-Aza) (Sigma), Trichostatin A (TSA) (Sigma), and a specific RARa antagonist ER50891 (a kind gift of Kouichi Kikuchi, Discovery Research Laboratories, Ibaraki, Japan [20]) were dissolved and stored as described previously [9]. Drugs were diluted in MEGM for HME1 cells and derived clones, or DMEM plus 5% charcoal-stripped FBS (Invitrogen) for T47D cells and derived clones. Cells were allowed to attach over night and treated in the dark with different drug combinations as indicated in the Results section. RA-treatment was performed for 24 h for colony formation assays, and for 72 h for transcription assays, adding fresh RA every 24 h. ER50891 treatment was for 24 h, while TSA and 5-Aza treatments were for 72 h.
antisense 59-CGC GCC GCG ACC TCC CGC GC-39). The PCR signal from the M alleles was normalized to the signal from a control CYP26A1 region amplified by using primers that do not recognize any CpG (P774 sense: 59-TTA GTG AAG GTT GTT TTG GGT-39 and 59-AAT ACA AAT CCC AAA ACT TAA-39). Statistical significance was calculated by Student's t-test on three independent determinations; p values at least ,0.05 were considered as significant.