Graphical Abstract

   graphic file with name fx1.jpg
   [47]Open in a new tab

Highlights

     * •
       L-Pro induces a reversible embryonic-stem-to-mesenchymal-like
       transition (esMT)
     * •
       The esMT is characterized by a dynamic redistribution of E-cadherin
     * •
       L-Pro acts as an epigenetic modifier remodeling H3K9 and H3K36
       methylation
     * •
       L-Pro and vitamin C regulate esMT-MesT plasticity modulating
       H3K9/H3K36 methylation
     __________________________________________________________________

   Comes et al. report that L-proline acts as an epigenetic modifier in
   embryonic stem cells (ESCs) inducing a reversible
   embryonic-stem-to-mesenchymal-like transition, which converts compacted
   ESCs into highly motile, invasive stem cells. Vitamin C counteracts
   L-proline promoting a mesenchymal-like-to-embryonic-stem-cell
   transition. These findings strengthen the role of the amino acids as
   key regulators of stem cell behavior.

Introduction

   In recent years, a number of small molecules have been identified that
   control stem cell differentiation and/or cell reprogramming.
   Surprisingly, only a few metabolites have been identified in this
   context, possibly because their relevance has been underappreciated.
   The “physiological” compounds identified so far include epigenetic
   modifiers or regulators of the cell energy/redox status acting on
   mitochondrial oxidative metabolism. For example, saturated fatty acids
   and the acylcarnitines were shown to improve embryonic stem cell (ESC)
   differentiation ([48]Yanes et al., 2010). Butyrate, a short-chain fatty
   acid, increases the efficiency of cell reprogramming by inhibiting
   histone deacetylases ([49]Mali et al., 2010). Finally, ascorbic acid
   (vitamin C), a redox controller and a cofactor for histone
   demethylases, is a key regulator of stem cell differentiation and
   reprogramming ([50]Cao et al., 2012; Esteban et al., 2010; Wang et al.,
   2011; Yang et al., 2008). However, it has recently become evident that
   also some naturally occurring amino acids can regulate stem cell
   behavior. Indeed, it has been shown that (1) ESC self-renewal depends
   on L-threonine catabolism ([51]Shyh-Chang et al., 2013; Wang et al.,
   2009) and (2) L-proline (L-Pro) forces ESCs toward an epiblast stem
   cell (EpiSC)-like state ([52]Washington et al., 2010) and regulates ESC
   metastability ([53]Casalino et al., 2011). L-Pro is particularly
   interesting because its mitochondrial oxidative catabolism is linked to
   cell survival/cell death in cancer cells ([54]Liu et al., 2012; Phang
   et al., 2012) and to life span extension in C. elegans ([55]Zarse
   et al., 2012). Here, we provide evidence that physiological
   concentrations of L-Pro are sufficient to convert ESCs into
   mesenchymal-like, highly motile, invasive pluripotent stem cells, which
   acquire metastatic potential in vivo. This previously unforeseen
   embryonic-stem-cell-to-mesenchymal-like transition (esMT) is fully
   reversible either after L-Pro withdrawal or by addition of vitamin C.
   Most remarkably, L-Pro extensively remodels both the transcriptome and
   the epigenome, thus acting as a potent signaling metabolite in
   pluripotent stem cells.

Results

L-Pro Modifies ESC Morphology Inducing Extensive Cytoskeletal Rearrangements

   Mouse ESCs grow as leukemia inhibitory factor (LIF)-dependent flat
   colonies in the presence of L-Pro ([56]Casalino et al., 2011;
   Washington et al., 2010). Examining different growth conditions, we
   found that this L-Pro-induced morphological transition markedly
   increased by lowering the cell-plating density to 50 ESCs/cm^2
   ([57]Figure S1 available online). Notably, clonal L-Pro-induced cell
   (PiC) colonies showed three distinctive zones ([58]Figure 1A): (1) a
   “central zone” or “colony core” consisting of polygonal-shaped, tightly
   adherent cells; (2) a “peripheral zone” at the external margin of the
   colony, consisting of less-cohesive spindle-shaped cells; and (3) a
   “surrounding zone,” forming a crown of mesenchymal-like cells scattered
   around the colony. These latter cells were characterized by a large
   nucleus with prominent nucleoli and a highly polarized mesenchymal
   shape with large and flat lamellipodial protrusions extending at the
   leading edge and needle-like trailing structures on the rear
   ([59]Figure 1A). Alterations of cell shape and the formation of
   lamellipodial protrusions are primarily driven by actin cytoskeleton
   remodeling ([60]Giannone et al., 2007; Ridley, 2011). Accordingly, we
   found that mesenchymal-like PiCs displayed long and polarized F-actin
   stress fibers frequently terminating in large and mature focal adhesion
   complexes. Presumably, as a consequence of this actin
   cytoskeleton-driven shape transition, also the microtubules organized
   into an extended network typical of mesenchymal motile cells
   ([61]Figures 1B and 1C). Thus, ESCs undergo a morphological
   transformation, which is reminiscent of changes associated with the
   acquisition of persistent and directional cell locomotory properties.

Figure 1.

   Figure 1
   [62]Open in a new tab

   L-Pro Induces Cytoskeletal Rearrangements and Cell Motility in ESCs

   (A and B) Representative photomicrographs of crystal violet (A) and
   Phalloidin-TRITC (B)-stained PiC and ESC colonies. Retraction fibers
   (^∗) and leading edge lamellipodia (arrow) are indicated (A). Nuclei
   were stained with Hoechst. (B) The scale bars represent 100 μm.

   (C) Fluorescence confocal photomicrographs of ESCs/PiCs labeled for
   vinculin and phalloidin (upper panels) or for β-tubulin (lower panels).
   Actin stress fibers were labeled with Phalloidin-TRITC (red; [B and
   C]). Microtubules and focal adhesions were stained with anti-β-tubulin
   (green; [C], lower panels) and anti-Vinculin (green; [C], upper panels)
   antibodies. Nuclei were counterstained with TO-PRO-3. The scale bars
   represent 50 μm.

   (D) Representative frames of time-lapse series from untreated (control)
   and L-Pro-treated ESCs. Images (20×) were captured starting from day 3
   postplating.

   (E) Average number of ESCs/PiCs migrating in 15% FBS (10.5 ± 1.5 ESCs
   versus 33 ± 3 PiCs) or toward a FBS gradient 1%–15% (13 ± 2 ESCs versus
   110 ± 13 PiCs).

   (F) Average number of PiCs migrating in 1% FBS either alone (control)
   or toward EGF (50 ng/ml), insulin (5 μg/ml), SDF-1 (100 ng/ml), or
   Cyr61 (200 ng/ml). Data are mean ± SD (n = 3); ^∗p < 0.01; ^∗∗p <
   0.005.

   See also [63]Figure S1 and [64]Movies S1, [65]S2, and [66]S3.

L-Pro Induces a Motile Phenotype in ESCs

   To test directly the possibility that L-Pro is sufficient to induce
   changes in adhesive and motile properties of stem cells, the ESC-to-PiC
   phenotypic transition was monitored by time-lapse video microscopy
   ([67]Figure 1D; [68]Movies S1, [69]S2, and [70]S3). Control ESCs formed
   compact colonies from which, occasionally, individual cells in direct
   contact with the substrate extended transient and short-lived
   cytoplasmic protrusions that appeared to direct the collective mode of
   locomotion of the entire colony ([71]Movie S1). Cell divisions were
   visualized as transient swellings or “bubbles” that sporadically
   protruded from the colony. Within 1–3 days of plating, L-Pro-treated
   ESCs formed, as expected, spherical colonies; however, by day 3 onward,
   single, highly motile cells started to scatter, detaching from the
   colony margins through the extension of highly dynamic and prominent
   protrusions ([72]Figure 1D; [73]Movies S2 and [74]S3). Importantly,
   these now freely motile, single PiCs maintained the ability to divide
   and frequently re-established dynamic and reversible cell-cell contacts
   with cells in the colony core. Consistent with the acquisition of
   directed cell motility, PiCs, but not ESCs, were able to migrate
   efficiently through porous membranes in transwell migration assays in
   response to gradients of fetal bovine serum (FBS) or toward different
   chemoattractants, including epidermal growth factor (EGF), insulin,
   stromal cell-derived factor-1 (SDF-1), and cysteine-rich angiogenic
   inducer 61 (Cyr61) ([75]Figures 1E and 1F). Thus, L-Pro is sufficient
   to promote the transition of ESCs toward a motile mesenchymal-like
   phenotype.

L-Pro Induces Remodeling of the ESC Transcriptome

   To elucidate whether a diverse transcriptional program is the basis of
   the morphological and motility differences between ESCs and PiCs, we
   compared their gene expression profiles. We found ∼1,500
   L-Pro-dependent, differentially regulated protein-coding genes
   ([76]Figures 2A and 2B; [77]Table S1). Gene ontology analysis revealed
   that L-Pro-regulated genes were functionally very diverse
   ([78]Figure 2C), but the pluripotency-associated factors (Oct4, Nanog,
   and Sox2) remained, as expected, unchanged. Interestingly, whereas
   genes involved in L-Pro uptake (Slc38a2), synthesis (Aldh18a and P5cr),
   or catabolism (Prodh) were not affected ([79]Table S1), both
   procollagen C-endopeptidase enhancer (Pcolce) and dipeptidyl peptidase
   4 (Dpp4/CD26), which regulate L-Pro recycling, were significantly
   downregulated. DPP4, which cleaves X-Pro dipeptides generating free
   L-Pro, was the most L-Pro-repressed protease ([80]Table S1). In
   contrast, expression of the DPP4 inhibitor, glypican 3 (Gpc3), was
   induced. Thus, ESCs respond to increased extracellular levels of free
   L-Pro by repressing genes involved in L-Pro recycling. Remarkably, PiCs
   and EpiSCs ([81]Brons et al., 2007; Tesar et al., 2007) shared 33% of
   differentially expressed protein-coding genes with respect to ESCs
   ([82]Figure 2D). However, Cer1, Asb4, Foxa2, Gata6, Sox17, and Stra8,
   which are among the most upregulated genes in EpiSCs, did not change in
   PiCs. Similarly, the developmental pluripotency-associated genes Dppa3,
   Dppa4, and Dppa5 that are repressed in EpiSCs were, instead, expressed
   at comparable levels in PiCs and ESCs ([83]Brons et al., 2007; Tesar
   et al., 2007). Moreover, the transcription factor Rex1 (Zfp42), which
   is silenced in epiblast cells just after implantation and is strongly
   repressed in EpiSCs, was only weakly downregulated in PiCs ([84]Table
   S1). Notably, immunofluorescence analysis revealed that PiCs, like
   ESCs, were REX1 positive (≥95%) and that REX1 expression persisted in
   the mesenchymal-like PiCs scattered around the colony ([85]Figures S2A
   and S2B). These data indicate that the morphological changes of PiCs
   are likely driven by a specific transcriptional program that is
   significantly different from the program induced in EpiSCs. Indeed, a
   significant set of genes regulated by L-Pro were either transcription
   factors or cofactors (∼12%; [86]Figure 2C; [87]Table S1), among them
   several Krüppel-like factors (Klfs), which are key regulators of cell
   differentiation and reprogramming. Of note, whereas Klf2, Klf4, and
   Klf5 were all downregulated by L-Pro, the expression of Klf6 and Klf7
   increased. Interestingly, PiCs and Klf2/Klf4/Klf5 triple knockdown ESCs
   ([88]Jiang et al., 2008) shared ∼29% of differentially expressed genes
   ([89]Figure 2D), thus suggesting that modulation of Klfs might be
   relevant for PiC specification.

Figure 2.

   Figure 2
   [90]Open in a new tab

   L-Pro Remodels the ESC Transcriptome

   (A) Heat map diagram of differentially expressed genes in PiCs versus
   ESCs.

   (B) Volcano plot showing the magnitude of differential expression
   [log[2](fold change), x axis] compared to the measure of statistical
   significance [−log[10](p value), y axis]. Genes with a magnitude of
   differential expression ≥5-fold are indicated in red.

   (C) Gene ontology (GO) analysis ([91]http://david.abcc.ncifcrf.gov).
   Pie chart showing the most represented GO term in the microarray
   analysis comparing PiCs versus ESCs. The numbers inside each pie slice
   show the fraction (%) of differentially expressed protein-coding genes
   ([92]Table S1) in each category.

   (D) Venn diagrams describing the overlap between genes differentially
   expressed in PiCs versus EpiSCs or Klf2/Klf4/Klf5 (Klf2,4,5) triple
   knockdown ESCs compared with control ESCs.

   (E) KEGG pathway enrichment analysis; pathways related to
   cytoskeleton/cell motility/adhesion or cancer are highlighted in green
   and blue, respectively.

   See also [93]Figure S2 and [94]Table S1.

   Gene ontology analysis of the PiC expression profile denoted a
   consistent overrepresentation of genes in six Kyoto Encyclopedia of
   Genes and Genomes (KEGG) pathways, all related to cell shape and
   motility ([95]Figure 2E). In keeping with the altered cytoskeleton
   architecture, 9% of L-Pro-regulated genes were structurally and/or
   functionally related to the cytoskeleton, whereas 12.8% were related to
   cell adhesion, cell junction, or cell motility ([96]Figure 2C;
   [97]Table S1). In line with the idea that cell motility/invasion is
   associated with extracellular matrix (ECM) remodeling, ∼11% of the
   L-Pro-regulated genes were extracellular proteins, including secreted
   proteases, as well as modulators of their proteolytic activity
   ([98]Table S1). These significant sets of genes regulated by L-Pro
   likely define the molecular signature that is required to direct and
   sustain the acquisition of the PiC motile phenotype. The PiC expression
   profile was also enriched in genes associated with six cancer-related
   KEGG pathways ([99]Figure 2E; [100]Table S1), among which there are
   members of the Janus kinase/signal transducers and activator
   (JAK/STAT), phosphatidylinositol 3-kinase (PI3K), and mitogen-activated
   protein kinase (MAPK) pathways, as well as the transforming growth
   factor (TGF)-beta pathway ([101]Table S1). Collectively, these findings
   support the possibility that L-Pro may promote the acquisition of
   features typical of invasive cancer cells.

PiCs Are Invasive and Metastatic Pluripotent Stem Cells

   To directly test PiC invasiveness, we first evaluated the effect of
   L-Pro on cell invasion in transwell assays. PiCs, but not ESCs, were
   able to migrate through Matrigel-coated transwells. More relevantly,
   >65% of migrating PiCs efficiently invaded a three-dimensional-Matrigel
   plug ([102]Figure 3A). Interestingly, the migratory and invasive
   properties of PiCs at 500 cells/cm^2 (PiCs^500) were strongly reduced
   compared to PiCs^50 generated at clonal density, i.e., 50 cells/cm^2
   (≥70%; [103]Figure S1B). This latter finding supports the key role of
   the seeding density in defining the motile PiC phenotype. Moreover,
   PiCs generated from the feeder-independent E14Tg2a ESCs showed a highly
   motile/invasive phenotype, thus ruling out the possibility of an ESC
   line-specific effect of L-Pro ([104]Figure S3).

Figure 3.

   Figure 3
   [105]Open in a new tab

   Invasive and Metastatic Potential of L-Pro-Treated ESCs

   (A) Parallel migration (gray bars) and cell invasion (black bars)
   transwell assays measuring migration/invasion of ESCs and PiCs toward a
   FBS gradient (1%–15%) ± SDF-1 (100 ng/ml). The average number of
   migrating and invading PiCs is indicated. Data are mean ± SD; n = 3;
   ^∗p < 0.05; ^∗∗p < 0.001.

   (B) Schematic representation of in vivo metastasis assay. Nonobese
   diabetic (NOD)/SCID mice were injected intravenously with either
   EGFP-marked ESCs, PiCs (1 × 10^6 cells), or B16-BL6 melanoma cells (2 ×
   10^5).

   (C) Lungs (left panels) and hematoxylin and eosin-stained lung tissue
   sections (right panels) from ESC/PiC-injected mice, showing large
   teratomas in two PiC-injected mice; n = 5/group.

   (D) Representative photomicrographs of lung tissue sections showing
   EGFP^+ PiC-derived teratomas (white stars, upper panel) or a B16-BL6
   melanoma cell-derived tumor used as a control (white star, lower
   panel).

   Scale bars represent 100 μm (C) and 500 μm (D). See also
   [106]Figure S3.

   We then asked if the migratory and invasive phenotype of PiCs
   correlated with metastatic potential in vivo. To this end, enhanced GFP
   (EGFP)-labeled ESCs or PiCs were injected intravenously into
   immunodeficient mice. The metastatic B16-BL6 melanoma cells were used
   as a positive control. Mice were sacrificed 4 weeks after injection,
   and the histology of liver and lung tissues was examined for
   macroscopic abnormalities. Whereas the ESC-injected mice showed
   morphologically normal tissues with no evidence of tumor masses, large
   teratomas were clearly visible in the lungs of PiC-injected mice
   ([107]Figure 3C). Moreover, teratomas were EGFP-positive confirming
   that they originated from the exogenous cells ([108]Figure 3D). These
   data provide direct evidence that PiCs have acquired metastatic
   properties and indicate that L-Pro-induced stem cells acquire two of
   the most decisive pathological traits of the so-called “migrating
   cancer stem cells,” i.e., stemness and invasiveness.

L-Pro Induces an Embryonic-Stem-Cell-to-Mesenchymal-like Transition

   Our data are also consistent with the possibility that L-Pro-induced
   morphological and motility phenotypes contribute to the acquisition of
   cell plasticity, likewise canonical epithelial-to-mesenchymal
   transition (EMT) ([109]Yang and Weinberg, 2008). Thus, we hypothesized
   that L-Pro might trigger an EMT-like process in ESCs, namely an esMT.
   EMT is characterized by the induction of mesenchymal markers, such as
   N-cadherin (N-cad), vimentin (Vim), and matrix metallopeptidase 2
   (Mmp2), and the concurrent downregulation of E-cadherin (E-cad), a
   major component of cell adherens junctions ([110]Kalluri, 2009; Thiery
   et al., 2009). N-cad, Vim, and Mmp2 were all upregulated in
   L-Pro-treated ESCs ([111]Figure 4A). The transcription factor brachyury
   (T) was also strongly induced ([112]Figures 4A and 4D). Notably, the
   expression of T has recently been associated with EMT during metastasis
   ([113]Fernando et al., 2010; Sarkar et al., 2012). N-cad transcripts
   were detected from day 2 onward during the esMT, whereas T transcripts
   were induced at day 3 ([114]Figure 4B), in a timeframe that correlated
   with the onset of the motile phenotype ([115]Figure 1D). At the protein
   level, both N-CAD and T accumulated later on ([116]Figure 4C). E-CAD,
   instead, was expressed at comparable levels in ESCs and PiCs
   ([117]Figures 4E and 4F). Accordingly, the expression of the E-cad
   transcriptional repressors, Snail and Slug, were not significantly
   altered ([118]Figure 4E). Nevertheless, altered trafficking and cell
   surface distribution of the E-CAD protein can also contribute to its
   functional downregulation ([119]Cavallaro and Christofori, 2004). In
   line with this notion, we observed a reduction of E-CAD protein at
   cell-cell junctions, accompanied by an accumulation of E-CAD in
   perinuclear vesicular-like compartments from day 3 of the esMT
   ([120]Figures 4G and 4H).

Figure 4.

   Figure 4
   [121]Open in a new tab

   L-Pro Induces an esMT

   (A) qPCR analysis of the expression of EMT markers in L-Pro-treated
   versus untreated ESCs (day 5).

   (B and C) Time course of N-cad and T expression measured by qPCR (B)
   and western blot (C) analysis. GAPDH, glyceraldehyde 3-phosphate
   dehydrogenase.

   (D) Microphotographs of T immunostaining (left panel) and phase
   contrast (right panel) of PiCs. The scale bar represents 100 μm.

   (E) qPCR analysis of E-cad, Snail, and Slug expression in L-Pro-treated
   versus untreated ESCs (day 5).

   (F) Time course of E-CAD expression in L-Pro-treated ESCs by western
   blot analysis. Densitometric analysis of E-CAD protein levels relative
   to untreated ESCs after normalization to GAPDH.

   (G) Images of E-CAD immunostaining of control and L-Pro-treated ESCs
   using conventional (upper panels) or confocal (lower panels)
   fluorescence microscopy. White arrow (inset) indicates the vesicular
   perinuclear localization of E-CAD. The scale bar represents 50 μm.

   (H) Fluorescence confocal photomicrographs showing colocalization
   (white arrows) of E-CAD with the Golgi marker GM130. Nuclei were
   counterstained with Hoechst. The scale bar represents 20 μm.

   (I) Time course of Pkdcc expression in L-Pro-treated ESCs by qPCR. (A,
   B, E, and I). Data are fold change compared to control ESCs after
   normalization to Gapdh.

   (J) Downregulation of Pkdcc expression by siRNAs (left panel) in
   L-Pro-treated ESCs (day 5). Data are fold change in gene expression
   compared to control (nontargeting siRNA). Effect of Pkdcc-silencing on
   the colony-type frequency (right panel); data are mean ± SD, ∼100
   colonies scored; n = 3; ^∗p < 0.01.

   See also [122]Figure S4.

   To assess if L-Pro-induced esMT might impact on the differentiation
   potential, we compared both the cardiac and neuronal differentiation
   kinetics of PiCs and ESCs. Despite the fact that PiCs, but not ESCs,
   expressed high levels of the panmesodermal marker brachyury, the
   kinetics of cardiac markers expression were comparable in the two lines
   ([123]Figure S4A). On the other hand, whereas neuronal and glia markers
   displayed similar levels in PiCs and ESCs, their kinetic of expression
   was delayed in PiCs ([124]Figure S4B). These data suggest that
   L-Pro-induced esMT does not promote or anticipate cardiac
   differentiation, whereas it slightly delays neural differentiation.
   Finally, to identify critical factors involved in the L-Pro-induced
   esMT, we focused our attention on the protein kinase domain-containing,
   cytoplasmic (PKDCC) kinase. Pkdcc was the most highly upregulated
   protein kinase-coding gene in L-Pro-treated ESCs and was induced from
   day 2 onward ([125]Figure 4I; [126]Table S1). Pkdcc is required for the
   round-to-flat shape transition of chondrocytes occurring during bone
   development ([127]Imuta et al., 2009) and is involved in the control of
   protein export from the Golgi ([128]Kinoshita et al., 2009). Of note,
   E-CAD accumulated in the Golgi complex in PiCs ([129]Figure 4H).
   Interestingly, we observed that transient Pkdcc-silencing inhibited the
   acquisition of the flat morphology in L-Pro-treated cells
   ([130]Figure 4J). Thus, L-Pro is sufficient to promote the esMT that is
   dependent, at least in part, on Pkdcc.

L-Pro Alters the Epigenetic Signature of ESCs

   To gain clues as to the mechanism through which the simple exposure to
   a nonessential amino acid induces dramatic changes in gene
   transcription and ESC behavior, we hypothesized that L-Pro might act as
   an epigenetic modifier. Consistent with this idea, the silencing of
   Jmjd1a, a H3K9 demethylase, forces ESCs toward a PiC-like phenotype,
   i.e., flat morphology, sensitivity to trypsin digestion and
   upregulation of Fgf5 and T ([131]Loh et al., 2007). We therefore tested
   whether L-Pro is sufficient to alter methylation levels of various
   critical histone lysine residues in ESCs. L-Pro augmented the global
   levels of H3K9me2 and H3K9me3, as well as the levels of H3K36me3
   ([132]Figure 5A, left panel). In contrast, the levels of H3K4me3 and
   H3K27me3 were unaltered ([133]Figure 5A, right panel). More relevantly,
   L-Pro triggered a genome-wide reprogramming of H3K9 and H3K36
   methylation status. H3K9 methylation was significantly altered at
   16,621 sites, with the highest increase in methylation occurring at
   noncoding intergenic regions of the genome ([134]Figures 5B, [135]S5A,
   and S5B). Genes showing PiCs-enriched H3K9me3 occupancy at proximal and
   distal regulatory regions are reported in [136]Figure S5C. H3K9
   methylation status changed in a significant fraction (27%) of the 1,521
   L-Pro-deregulated genes (p < 0.05; [137]Figure 5C, upper panel).
   Moreover, those genes displaying a modified H3K9 methylation status
   were enriched in pathways relevant for the esMT process
   ([138]Figure 5C, lower panel). Consistently, genes downregulated in
   PiCs displayed an increase in H3K9me3 at regulatory regions
   ([139]Figure 5D). Noteworthy, repetitive elements were frequently
   detected at sites displaying increased H3K9me3, suggesting an induced
   local heterochromatin spreading into nearby genes that influences their
   expression ([140]Figure 5D). Remarkably, chromatin immunoprecipitation
   sequencing (ChIP-seq) showed the highest H3K9me3 enrichment at
   constitutive heterochromatin, i.e., pericentromeres and gene deserts,
   indicating that L-Pro induces heterochromatin reorganization toward
   more dense structures in the nucleus ([141]Figures S5B and S5D).

Figure 5.

   Figure 5
   [142]Open in a new tab

   L-Pro Modifies the Epigenomic Signature of ESCs

   (A) Western blot analysis of global H3K9 mono-, di-, and trimethylation
   (left) and H3K4, K27, and K36 trimethylation (right) levels in PiCs
   compared to ESCs. Densitometric analysis is relative to control ESCs
   after normalization to total histone H3. Data are mean ± SD (n = 3);
   ^∗p < 0.05; ^∗∗p < 0.001.

   (B) Genome-wide H3K9me3 peaks (upper panel) and SICER-based
   differential distribution (lower panel) of H3K9me3 binding sites at
   genic and intergenic regions in PiCs versus ESCs.

   (C) Venn diagram correlating microarray gene expression and ChIP-seq
   data (p = 0.03; Fisher’s exact test; upper panel). KEGG pathway
   enrichment analysis of 439 loci that are H3K9me3 differentially
   enriched and deregulated (lower panel).

   (D) Snapshot of H3K9me3 profile at Smarcad1, Apod, and Ceacam1 genes.

   (E) Genome-wide H3K36me3 peaks (upper panel) and SICER-based
   distribution (lower panel) of differential H3K36me3 binding sites at
   genic and intergenic regions in PiCs versus ESCs.

   (F) Venn diagram correlating microarray gene expression and ChIP-seq
   data (p = 0.03; Fisher’s exact test; upper panel) and KEGG pathways
   enrichment analysis of 628 loci that are H3K36me3 differentially
   enriched and deregulated (lower panel).

   (G) Snapshot of H3K36me3 profile at Pitx2, Fgf5, and Pkdcc genes.

   See also [143]Figure S5 and [144]Table S2.

   Differentially expressed genes that were not altered in the H3K9me3
   profile might be regulated by additional epigenetic marks.
   Consistently, ChIP-PCR analysis revealed a significant enrichment of
   the H3K27me3-repressive mark at promoters of L-Pro-downregulated genes
   ([145]Figure S5E). Conversely, L-Pro-upregulated genes showed reduced
   levels of H3K27me3 and increased levels of the H3K4me3-permissive mark.
   Consistent with a global H3K9 methylation gain, we detected an overall
   increase of H3K9 methylation at both promoter and gene body regions
   ([146]Figure S5E).

   We also investigated the whole-genome profile of H3K36me3 and observed
   8,648 significant peaks altered by L-Pro treatment ([147]Figure 5E).
   Intragenic regions of differentially methylated clusters of loci were
   H3K36me3-increased in PiCs relative to ESCs ([148]Figure S5F), and a
   significant fraction (41%; p < 0.05) of PiC-specific genes showed
   significant H3K36me3 changes according to the transcriptional status
   ([149]Figures 5F and 5G). Notably, L-Pro induced H3K36me3 enrichment at
   noncoding regions, in line with recent evidence that H3K36me3 might
   promote heterochromatin formation ([150]Figure S5G; [151]Chantalat
   et al., 2011). Collectively, our findings support a role of L-Pro as a
   genome-wide inducer of H3K9 and H3K36 methylation in pluripotent stem
   cells.

Vitamin C Reverts the PiC Phenotype and Induces H3K9 and H3K36 Demethylation

   Increasing evidence indicates that demethylation of H3K36 and H3K9 is
   required for an efficient somatic/pre-induced pluripotent stem cell
   (iPSC) to iPSC transition ([152]Chen et al., 2013; Wang et al., 2011).
   This crucial step in cell reprogramming is controlled by vitamin C,
   which is a cofactor of H3K9 jumonji demethylases ([153]Wang et al.,
   2011) and a key regulator of stem cell differentiation and somatic cell
   reprogramming ([154]Cao et al., 2012; Esteban et al., 2010; Yang
   et al., 2008). We thus tested the effect of vitamin C on the PiC
   phenotype. Remarkably, vitamin C almost completely reverted the esMT at
   the morphological level, promoting
   a mesenchymal-like-to-embryonic-stem-cell transition (MesT), even in
   the presence of L-Pro. Remarkably, vitamin C mimicked the phenotype
   caused by L-Pro removal ([155]Figure 6A). To rule out the possibility
   that vitamin C induced selection of ESCs rather than PiC-to-ESC
   reversion, freshly generated PiCs (P[0]) were directly plated at low
   density and the frequency of domed and flat colonies was assessed in
   the different culture conditions. When PiCs were plated in the presence
   of L-Pro, about 10% of the colonies showed a domed phenotype (ESCs;
   [156]Figures S6A and S6B), which corresponds to the degree of
   heterogeneity observed in freshly generated PiCs. Indeed, PiCs culture
   consistently display 10%–15% of domed (ESCs) and 85%–90% of flat (PiCs)
   colonies ([157]Casalino et al., 2011). Remarkably, when PiCs were
   plated either in the absence of L-Pro or in the presence of L-Pro plus
   vitamin C ([158]Figures S6A and S6B), the percentage of domed colonies
   (ESCs) increased up to 50%. Importantly, the total number of colonies
   generated in the different culture conditions was comparable. Moreover,
   the migratory and invasive properties of the vitamin C-treated PiCs
   (ESCs^Rev) resembled the one of ESCs ([159]Figure 6B). Finally, both
   the frequency and the timing of appearance of flat colonies were
   comparable in L-Pro-treated ESCs and ESCs^Rev ([160]Figures 6C and 6D),
   thus ruling out the possibility that ESCs^Rev were primed for PiC
   formation. All together these findings provide evidence that vitamin C
   reverts both the morphological and the functional phenotype of PiCs
   toward that of ESCs. At the molecular level, vitamin C repressed
   L-Pro-induced T expression and reverted L-Pro-induced downregulation of
   Inhbb expression ([161]Figure 6E). Furthermore, vitamin C inhibited the
   L-Pro-induced increase in H3K9me3 and H3K36me3 levels to an extent that
   was comparable to that observed following L-Pro withdrawal. Thus,
   vitamin C induces the MesT by remodeling the epigenetic signature of
   PiCs ([162]Figure 6F). To get insight into the epigenetic mechanism
   underlying vitamin C-induced reversion of the PiC phenotype, we knocked
   down Jmjd1a. To this end, PiCs were transfected with either shRNA
   constructs targeting different regions of the transcript (Jmjd1a-
   shRNA1 and shRNA2) or control shRNAs (Jmjd1a- shRNA1mut and Luciferase
   shRNA), as previously described ([163]Loh et al., 2007). Of note,
   neither the morphology nor the expression of pluripotency markers was
   altered in Jmjd1a KD (≥70% gene silencing; [164]Figure S6D) compared to
   controls PiCs (data not shown). We thus plated Jmjd1a KD and control
   PiCs in the presence or the absence of vitamin C and analyzed the
   frequency of flat and domed colonies. The frequency of appearance of
   domed colonies was nearly identical in vitamin C-treated Jmjd1a KD and
   control PiCs ([165]Figure S6E). Notably, the effect of vitamin C
   mimicked the one achieved by L-Pro removal ([166]Figure S6E). These
   data indicate that downregulation of Jmjd1a is not sufficient to
   prevent vitamin-C-dependent PiC-to-ESC reversion. Remarkably, neither
   glutathione nor N-acetyl-L-cysteine were able to revert PiCs toward an
   ESC phenotype ([167]Figure S6C), thus excluding the possibility that
   vitamin C acts in this process through its antioxidant activity.

Figure 6.

   Figure 6
   [168]Open in a new tab

   Vitamin C Reverts the PiC Phenotype Inducing H3K9 and H3K36
   Demethylation

   (A) Schematic representation of the experimental strategy. Freshly
   generated PiCs (P[0]) were passaged twice (P[1] and P[2]) with L-Pro
   (control), without L-Pro, or with L-Pro + VitC. Cells were then plated
   at 500 cells/cm^2, grown for 5 days, and processed for colony phenotype
   assay (domed versus flat; right panel). Data represent the percentage
   of colony-type frequency and are mean ± SD; ∼100 colonies scored; n =
   3.

   (B) Migration (white bars) and cell invasion (gray bars) transwell
   assays of ESCs, PiCs, and vitamin C-treated PiCs (ESCs^Rev) toward a
   FBS gradient (1%–15%). The average number of migrating and invading
   cells is indicated. Data are mean ± SD; n = 3; ^∗p < 0.01; ^∗∗p <
   0.0001.

   (C and D) Comparative analysis of L-Pro-induced phenotypic transition
   in ESCs^Rev and ESCs. (C) Flat colony frequency of control (untreated)
   and L-Pro-treated ESCs and ESCs^Rev at days 4 and 5 after plating. Data
   are mean ± SD (n = 3; ∼100 colonies were scored in each experiment).
   (D) Time course analysis of PiC formation in L-Pro-treated ESCs versus
   ESCs^Rev. Representative photomicrographs of crystal violet-stained,
   L-Pro-treated ESCs and ESCs^Rev colonies at the indicated time points.
   The scale bars represent 50 μm.

   (E) qPCR analysis of T and Inhbb expression. Data are fold change in
   gene expression compared to control (+L-Pro) after normalization to
   Gapdh.

   (F) Western blot analysis of H3K9me3 and H3K36me3 levels (lower
   panels). Densitometric analysis (upper panels) is relative to control
   ESCs after normalization to total histone H3. Data are mean ± SD (n =
   2); ^∗p < 0.05; ^∗∗p < 0.01.

   See also [169]Figure S6.

Discussion

   How metabolites regulate stem cell biology represents a major challenge
   and is still poorly understood. Here, we uncover an unexpected role of
   L-Pro as a key regulator of mouse ESC migratory and invasive phenotype,
   revealing that L-Pro is a potent epigenetic modifier. L-Pro, and in
   particular its mitochondrial catabolism, has become an important area
   of study being at the crossroads of key metabolic pathways, including
   citric acid cycle, polyamines biosynthesis, and urea cycle ([170]Phang
   et al., 2012). Despite that L-Pro is a nonessential amino acid
   ([171]Phang et al., 2008), we demonstrate that the addition of L-Pro in
   ESC culture medium is sufficient to induce a previously unforeseen
   esMT, which converts compacted/clustered ESCs into scattered, freely
   motile pluripotent stem cells. Of note, this phenotype was not observed
   in previous studies ([172]Casalino et al., 2011; Washington et al.,
   2010), most likely because it markedly increased when ESCs are seeded
   at clonal density in the presence of L-Pro. Accordingly, PiCs generated
   at higher density (PiCs^500) showed poor migratory and invasive
   properties compared to PiCs generated at clonal density (PiCs^50)
   indicating that seeding density strongly influences the EMT-like
   phenotype of PiCs.

   The L-Pro-induced esMT is accompanied by a global remodeling of the ESC
   transcriptome, which nicely correlates with the phenotypic variation
   observed. Indeed, genes involved in the control of cell
   cytoarchitecture, adhesion, and migration/invasiveness are largely
   affected by L-Pro. Notably, despite this extensive remodeling of the
   transcriptome, expression levels of the pluripotency-associated factors
   (Oct4, Nanog, and Sox2) and the developmental pluripotency-associated
   gene Dppa3 (Stella) are comparable in PiCs and ESCs. Furthermore, PiCs,
   like ESCs, are REX1 (Zfp42)-positive (≥95%) and show similar in vitro
   differentiation kinetics. This molecular signature and the fact that
   PiCs strictly depend on LIF for self-renewal ([173]Casalino et al.,
   2011) suggest that they define a population of inner cell mass-like
   cells that have acquired a motile and invasive phenotype.

   Interestingly, L-Pro-induced esMT is accompanied by a significant
   increase in H3K9 and H3K36 methylation levels as well as by a global
   remodeling of the H3K9me3 and H3K36me3 marks. This L-Pro-induced esMT
   is fully reversible either after L-Pro withdrawal or by addition of
   ascorbic acid (vitamin C), promoting a MesT and inducing the reduction
   of H3K9 and H3K36 methylation levels.

   To the best of our knowledge, only three metabolites, L-threonine,
   butyrate, and vitamin C, have been identified so far that act as
   epigenetic modulators of ESC self-renewal and/or somatic cell
   reprogramming ([174]Mali et al., 2010; Shyh-Chang et al., 2013; Wang
   et al., 2009). Interestingly, L-threonine, which is the only amino acid
   among these metabolites, influences H3K4me3 by controlling the
   abundance of S-adenosylmethionine, leaving other lysine residues
   unaltered ([175]Shyh-Chang et al., 2013). This intriguing specificity
   is also exerted by L-Pro that conversely influences the global level of
   trimethylation of lysines K9 and K36, which would likely exert a wider
   effect on gene expression. This is consistent with the extensive
   molecular and phenotypic variations induced by L-Pro on ESCs, as, for
   instance, the striking acquisition of motility and invasiveness. Most
   remarkably, L-Pro-induced global increase of H3K9 and H3K36
   trimethylation preferentially occurs at noncoding intergenic regions
   and constitutive heterochromatin. Whereas H3K9me3 is a well-known
   repressive heterochromatic mark, the H3K36me3 deposition within
   heterochromatin domains has only recently emerged. H3K36me3 might thus
   contribute to the establishment of a transcriptionally unfavorable
   chromatin conformation by recruiting repressive complexes
   ([176]Chantalat et al., 2011). Our data suggest that L-Pro can induce
   heterochromatization, increasing H3K9me3 and H3K36me3 especially at
   pericentromeric and gene desert regions, and imply that there is a
   hierarchy of target chromatin regions and associated noncoding elements
   that are more sensitive to the effect of L-Pro in ESCs.

   Thus, our findings add to the emerging evidence that metabolites are
   key mediators of epigenetic regulation of stem cell fate and strengthen
   the role of the amino acids in this process. Yet how metabolites
   control chromatin modifiers in space and time, translating a dynamic
   metabolic state into a histone code, is still a key unresolved question
   ([177]Folmes et al., 2012; Phang et al., 2013; Sassone-Corsi, 2013;
   Zhang et al., 2012). Interestingly, it has been recently proposed that
   the concentration of metabolites and their biosynthetic enzymes may
   localize to chromatin subdomains promoting the clustering of relevant
   DNA and histone modifications at specific genomic loci ([178]Katada
   et al., 2012). It is thus tempting to speculate that specific
   histone-modifying enzymes might sense the presence of L-Pro and
   eventually translate this signal into different histone maps.
   Interestingly, we have shown that vitamin C, which is a cofactor of the
   Jumonji-family histone demethylases, antagonizes the effect of L-Pro on
   H3K9me3 and H3K36me3 levels. Of note, knockdown of Jmjd1a (H3K9
   demethylase) does not prevent vitamin C-induced PiC-to-ESC reversion.
   However, because many other vitamin C-regulated histone modifier
   enzymes might be involved in this process, further studies are needed
   to reveal its molecular nature.

   We believe that our findings have several implications for stem cell
   biology. First, it has been shown that the reprogramming process
   initiates through a mesenchymal-to-epithelial transition (MET) ([179]Li
   et al., 2010; Samavarchi-Tehrani et al., 2010) and that vitamin
   C-dependent decrease of H3K9/H3K36 methylation levels promote
   somatic/pre-iPSC to iPSC transition ([180]Chen et al., 2013; Wang
   et al., 2011). Thus L-Pro and vitamin C exert opposite effects on H3K9
   and H3K36 methylation status, regulating antagonistic cellular
   processes relevant for cell reprogramming. Recently, it was reported
   that vitamin C also promotes Tet-mediated DNA demethylation
   ([181]Blaschke et al., 2013). It will be thus interesting to
   investigate the effect of L-Pro on DNA methylation.

   Second, the L-Pro/vitamin C-regulated esMT-MesT switch is reminiscent
   of the transient EMT-MET switch that has been proposed as a driving
   force of metastasis, largely influenced by microenvironment
   ([182]Brabletz et al., 2005). Indeed, it is proposed that induction of
   EMT confers migrating properties to cancer stem cell ([183]Brabletz,
   2012). How this occurs is still poorly defined. Interestingly, the ECM
   is mostly composed of collagen (80%) and is considered as a reservoir
   of mobilizable L-Pro ([184]Phang et al., 2008). Here, we suggest that
   L-Pro, which is released from ECM degradation during tumor invasion,
   may act as environmental cue of the EMT to control tumor dissemination.
   Moreover, L-Pro metabolism is dysregulated in cancer ([185]Liu et al.,
   2012; Phang et al., 2012) and L-Pro synthesis is associated with tumor
   progression and aggressiveness ([186]Scott et al., 2011), yet the
   mechanisms remain largely unsolved. Although much work remains to be
   carried out to assess the in vivo relevance of L-Pro as signal
   molecule, our findings contribute to elucidating its role in regulating
   stem cell behavior and provide insights into the mechanisms by which
   metabolites control stem cell fate.

Experimental Procedures

Cell Cultures

   Wild-type TBV2 (129/SvP), β-actin/EGFP TBV2 and E14Tg2a (E14) mouse
   ESCs were cultured as described ([187]Casalino et al., 2011). PiCs were
   generated as previously described ([188]Casalino et al., 2011).
   Briefly, ESCs were seeded at low density (50–500 cells/cm^2) and grown
   in the presence of L-Pro (50–500 μM) for 5 days. Freshly generated PiCs
   (P[0]) consistently displayed 10%–15% of domed (ESCs) and 85%–90% of
   flat (PiCs) colonies. ESC treatment details are reported in the
   [189]Supplemental Experimental Procedures. B16-BL6 melanoma cells were
   cultured in RPMI, 10% FBS, 2 mM glutamine, and 100 U/ml
   penicillin/streptomycin.

In Vitro Differentiation

   For cardiomyocyte differentiation, ESCs/PiCs were allowed to
   differentiate through embryoid bodies in hanging drop (500 cells/drop)
   as described ([190]Parisi et al., 2003). Neuronal differentiation was
   performed as described ([191]Fico et al., 2008).

Pkdcc Silencing

   ESCs were plated at 50 cells/cm^2 in the presence of L-Pro (150 μM) and
   transfected twice, at 30 and 54 hr after plating, with Pkdcc siRNAs or
   control (150 nM; Silencer Select siRNA; Ambion) using Lipofectamine
   RNAiMAX (Invitrogen). tBLOCK-iT Alexa Fluor Red (50 nM; Ambion) was
   used to assess the transfection efficiency.

Jmjd1a Silencing

   PiCs were transfected with 2 μg of DNA plasmids ([192]Loh et al., 2007)
   using Lipofectamine 2000 (Invitrogen). Puromycin (1 μg/ml; Sigma)
   selection started 1 day after transfection and was maintained for
   3 days. Puromycin-resistant cell pools were either harvested for RNA
   extraction or plated at low density without L-Pro or with L-Pro ±
   vitamin C (VitC) (500 μM) and the resulting colonies analyzed at day 5
   after plating.

RNA Preparation and Quantitative RT-PCR

   Total RNAs were isolated using RNeasy mini kit and analyzed at the Ohio
   State University microarray facility. QuantiTect Reverse Transcription
   kit (QIAGEN) and SYBR Green PCR master mix (FluoCycle II SYBR;
   EuroClone) were used for quantitative PCR (qPCR) analysis. Primers are
   in the [193]Supplemental Experimental Procedures.

Cell Migration/Invasion and Metastasis Assays

   Migration assays were performed using polycarbonate (5 μm pore; Costar)
   or polyester membrane transwells (8 μm pore; BD Biosciences). ESCs were
   plated at 50 or 500 cells/cm^2 in complete medium ± L-Pro. At day 5,
   cells were detached with Accutase (Sigma) and 2 × 10^4 ESCs/PiCs were
   incubated for 20–22 hr. EGF, insulin, SDF-1 (Sigma-Aldrich), and Cyr61
   (R&D Systems) were added in 1% FBS. Invasion assays were performed
   using BD BioCoat Growth Factor Reduced MATRIGEL Invasion Chambers
   ([194]Supplemental Experimental Procedures).

   EGFP-marked ESCs were plated at 50 cells/cm^2 and grown in complete
   medium ± L-Pro for 5 days. ESCs/PiCs (1 × 10^6 cells) and B16-BL6 (2 ×
   10^5) were injected into the tail veins of 7-/8-week-old severe
   combined immunodeficiency (SCID) mice (five mice/group; Charles River
   Laboratories). ESCs/PiCs were also injected subcutaneously as control.
   After 4 weeks, mice were sacrificed and paraformaldehyde-perfused
   tissues were explanted and photographed (Leica MZ16 FA
   stereomicroscope) before embedding in optimal cutting temperature. The
   care and husbandry of mice and metastasis experimental procedures were
   in accordance with European Directives No. 86/609 and with Italian D.L.
   116 and approved by the Institute of Genetics and Biophysics (IGB)
   veterinarian.

Time-Lapse Experiments

   ESCs were seeded on gelatin-coated plates at 50 cells/cm^2 and grown in
   complete medium ± L-Pro (150 μM). Images were captured (20×) every
   5 min for about 24 hr using Nikon Eclipse TE2000-E inverted microscope.

Immunofluorescence Analysis

   Cells were fixed (4% paraformaldehyde) and permeabilized (0.1% Triton
   X-100) at room temperature. After incubation with the primary
   antibodies ([195]Supplemental Experimental Procedures), cells were
   incubated with the appropriate secondary antibodies (Alexa Fluor 1:200;
   Molecular Probes). Phalloidin-TRITC (1:300; Sigma) was used for actin
   staining. For REX1 immunofluorescence, 1–1.5 × 10^4 ESCs/PiCs were
   centrifuged at 900 rpm for 15 min (Cytocentrifuge; Thermo Scientific).
   Cytospin samples were processed as above. Images were taken using
   inverted (DMI6000B; Leica Microsystems) and confocal (TCS-SP2-AOBS;
   Leica Microsystems) microscopes. Leica FW4000 and AF6000 software were
   used for image acquisition/elaboration.

Western Blot Analysis

   Cell lysates were prepared with ice-cold radio immunoprecipitation
   assay (RIPA) lysis buffer. Cell pellets were resuspended in Triton
   buffer for histone extraction. Primary antibodies are reported in the
   [196]Supplemental Experimental Procedures. Detection was performed with
   enhanced chemiluminescence reagents (Amersham Biosciences).
   Densitometric analysis was carried out using the software GelEval 1.35.

ChIP-qPCR and -Sequencing

   Chromatin immunoprecipitation was performed as previously described
   ([197]De Bonis et al., 2006). Suitable amount of chromatin was
   incubated with 5–10 μg of the indicated antibodies ([198]Supplemental
   Experimental Procedures). Immunoprecipitated complexes were recovered
   with protein A sepharose (Pharmacia), washed with low and high salt
   buffers, reverse-crosslinked, and purified. Primers are reported in the
   [199]Supplemental Experimental Procedures. The statistical significance
   of the qPCR results was assessed using the Student’s t test.

   H3K9me3 and H3K36me3 immunoprecipitated samples were prepared for SOLiD
   System 4.0 (Applied Biosystems)-based sequencing. Mapping to the mouse
   reference genome and peak-finding software are reported in the
   [200]Supplemental Experimental Procedures.

Acknowledgments