Wnt won the war: Antagonistic role of Wnt over Shh controls dorso-ventral patterning of the vertebrate neural tube
Developmental Neurobiology and Regeneration Unit, Institute for Research in Biomedicine, Parc Cientific de Barcelona; Centro de Investigación Biomédica en Red sobre Enfermedades Neurodegenerativas (CIBERNED), and Department of Cell Biology, University of Barcelona, Barcelona, Spain
The vertebrate spinal cord is a paradigm of a patterned structure. At its ventral region reside neurons that govern the motor control while its dorsal region is occupied by neurons that command and organize sensory information. This dorso-ventral (D-V) organization is established early in development (Jessell,2000). The early developing spinal cord consists of a simple tube (the neural tube) whose lumen is surrounded by a pseudo-stratified layer of mitotically active neuroepithelial cells (Fig 1). When entering the differentiation pathway, postmitotic neurons migrate laterally and distribute towards the lateral region. The use of molecular markers has permitted the identification of eleven subtypes of neural progenitors arranged along the dorso-ventral axis of the spinal cord (Wilson and Maden,2005) (Fig. 1). The identity of each subtype is defined by a combinatorial code of transcription factors (Briscoe et al.,2000; Jessell,2000; Wilson and Maden,2005). Due to its simple architecture and manipulability, the developing spinal cord has largely been used as a model to investigate the mechanism by which tissues are patterned.
Spinal cord D-V patterning results from the activity of several secreted signaling molecules that provide positional information to neural progenitor cells. These include Sonic hedgehog (Shh), Wnts, Bone morphogenetic proteins (Bmp), Fibroblast Growth Factors (FGF), and Retinoic Acid (RA), although it is generally accepted that counteracting gradients of Shh, emanating from the ventral aspects of the spinal cord, and Bmps and Wnts, produced from the dorsal ones, are the main actors in spinal cord DV pattern formation (Wilson and Maden,2005) (Fig. 1). In this review, we discuss recent experimental evidence that shows novel roles of Wnts in inducing dorsal identities at the expense of ventral ones.
How patterning signals determine the identity of neural progenitors in a precise spatio-temporal order is still an issue of extensive study. Most of our knowledge about the mechanisms that account for cell specification in the neural tube comes from the characterization of Shh action, which is both necessary and sufficient for the establishment of the five ventral most neural subtypes of the spinal cord (for an excellent recent review see Dessaud et al.,2008).
SHH DIRECTS VENTRAL SPINAL CORD PATTERNING
Shh is a member of the Hh family of secreted signaling molecules, crucial in animal development and in adult tissue homeostasis (Chari and McDonnell,2007; Wang et al.,2007). Shh is initially expressed in the notochord, which is located ventrally to the neural tube (Marti et al.,1995b). Subsequently, cells at the ventral midline of the neural tube (floor plate cells) activate expression of Shh, and in turn become a new source of Shh (Marti et al.,1995a,b; Roelink et al.,1995).
Shh signaling involves two transmembrane proteins: Patched (Ptc), which is the Shh receptor, and Smoothened (Smo), which initiates the intracellular signaling. In the absence of Shh ligand, Ptc blocks Smo activity. Binding of Shh to Ptc releases Smo inhibition, which then transmits the intracellular signal (Alcedo et al.,1996; Marigo et al.,1996; Stone et al.,1996; van den Heuvel and Ingham,1996; Taipale et al2002, Bijlsma et al2006). Lately, growing evidence shows a critical role of primary cilia in the transduction of Hh signaling in vertebrates (for details, see excellent recent reviews by Rohatgi and Scott,2007; Gerdes et al.,2009). Downstream of Smo the signal culminates with the regulation of the activity of members of the Gli family of zinc-finger-containing transcription factors that can activate or repress target genes in the presence or the absence of Shh, respectively (reviewed in Huangfu and Anderson,2006; Wang et al.,2007). Three Gli proteins have been identified in vertebrates: Gli1, which has only a transcriptional activator function, and Gli2 and Gli3, which exhibit both transcriptional repressor and activator properties (reviewed in Jacob and Briscoe,2003; Ruiz i Altaba et al.,2003). The activity of Gli proteins is differentially regulated by Shh signaling. Importantly, in the absence of Shh, Gli3 is converted into a repressor transcription factor (Gli3R) by the proteolytical removal of its C-terminal domain. In this condition, Gli3R represses the pathway target genes. By contrast, Gli2 is fully degraded in the absence of the ligand. Shh inhibits Gli protein processing and, in addition, induces the expression of Gli1. The full-length Gli proteins formed in the presence of Shh act as transcriptional activators, which induce the expression of the pathway target genes (Dai et al.,1999; Ruiz i Altaba,1999; Sasaki et al.,1999; Aza-Blanc et al.,2000; Wang et al.,2000; Pan et al.,2006). All three Gli proteins are expressed in the spinal cord during progenitor subtype specification although in different D-V domains. Gli1 is expressed in a ventral to dorsal gradient. On the contrary, Gli3 expression is inhibited by Shh and shows an opposed pattern of expression to Gli1. By contrast, Gli2 expression seems not to be controlled by Shh and is expressed along the entire D-V axis of the spinal cord (Hui et al.,1994; Lee et al.,1997; Ruiz i Altaba,1998; Bai et al.,2002; Jacob and Briscoe,2003).
In the spinal cord, Shh operates as a morphogen, the term applied to any extracellular signaling molecule that, acting at a distance from its source, induces different cell types in a concentration-dependent manner (Wolpert,1969; Dessaud et al.,2008). A three-step model has been proposed to describe how Shh action patterns the ventral spinal cord: (1) a gradient of Shh controls, in a concentration-dependent manner, the expression of a set of homeodomain (HD) and basic Helix-Loop-Helix (bHLH) transcription factors. They are grouped into class I or class II proteins according to whether they are inhibited or activated by Shh respectively, (2) cross-regulatory interactions between pairs of class I and II proteins sharpen their boundaries of expression along the D-V axis, (3) the combinatorial expression of the HDs and bHLH transcription factors is what ultimately defines the identity of the neural progenitor cells, which in turn will give rise to postmitotic neuronal subtypes (Briscoe et al.,2000) (Fig. 2). This model could be extended to the action of other patterning signals, not only in the spinal cord.
Several lines of evidence support a model where the gradient of Shh would be transduced into a gradient of intracellular Gli activity, which in turn can account for the morphogen activity of Shh (Stamataki et al.,2005). In addition, neural progenitors can elicit different responses depending on the duration of exposure to Shh/Gli signaling: longer times of exposure promote the induction of more ventral cell types (Stamataki et al.,2005; Dessaud et al.,2007). Shh −/− deficient mice lack most of the ventral cell types (Chiang et al.,1996). Some of these defects are rescued in the Shh−/− Gli3 −/ − double mutants; even though no V3 interneurons and floor plate cells are specified, other ventral cell types like motor neurons are induced in the double mutants, although in an intermingled distribution (Litingtung and Chiang,2000). Since most of the patterning depending on Gli3 is attributable to its repressor activity (Persson et al.,2002; Meyer and Roelink,2003), it is conceivable that Shh function not only induces Gli activator but also restricts repressor Gli3 activities. Consistently, Shh−/−; Gli3 −/− phenotype is similar to that found in Gli2−/−; Gli3−/− double mutants, where all the Gli activity is abolished (Bai et al.,2004). In addition, these phenotypes suggest the existence of additional signal(s), different from Shh/Gli, able to direct neural cell specification even at the most ventral regions of the spinal cord. The identity of these signal(s) has not been yet firmly defined, although dorsally expressed signaling molecules like Bmps or Wnts are the most obvious candidates.
Wnt family of secreted palmitoylated glycoproteins are, like Hhs, key factors during embryonic development and adult life (Logan and Nusse,2004; Clevers,2006). Currently, three separate molecular pathways have been identified that transduce Wnts signaling in cells (the canonical, planar cell polarity, and Ca2+ pathways) (Komiya and Habas,2008). In the canonical Wnt pathway, the transcriptional co-activator β-catenin is stabilized and translocates to the nucleus where it regulates transcription by interacting with members of the TCF/Lef family of DNA binding proteins (Behrens et al.,1996; Molenaar et al.,1996; van de Wetering et al.,1997). In the absence of Wnt signaling, cytosolic β-catenin is degraded by the proteosome via a mechanism that requires phosphorylation of β-catenin by GSK3β. In these conditions, Tcf proteins bind to Groucho co-repressor resulting in the inhibition of target genes. Binding of a secreted Wnt ligand to cell surface receptors of the Frizzled family and LRP5/6 initiates signaling, which is transduced through Dishevelled, a cytoplasmic component of the Wnt signaling pathway, to inhibit GSK-3β activity and thus stabilize β-catenin (Bhanot et al.,1996; Aberle et al.,1997; Tamai et al.,2000; Komiya and Habas,2008). Multiple Wnts are expressed both in the dorsal and ventral regions of the neural tube (Parr et al.,1993; Hollyday et al.,1995; Megason and McMahon,2002). Additionally, expression of Tcf1, Tcf3, and Tcf-4 proteins but not Lef1 has been detected in the developing spinal cord (Alvarez-Medina et al.,2008). Megason and McMahon (2002), using in ovo chick electroporation assays, demonstrated a mitogenic activity of Wnt1 and Wnt3a on neural progenitors, but not of other Wnts analysed. Wnt1 and Wnt3a, the expression of which is restricted to the dorsal midline of the spinal cord, transduce mainly through the β-catenin canonical pathway. Several lines of evidence coming from gain or loss of function in chicken, mice, and zebrafish confirmed a proliferative role of Wnt/β-catenin signaling in neural tissues, mainly attributable to Wnt1 and Wnt3a ligands (Dickinson et al.,1994; Megason and McMahon,2002; Zechner et al.,2003; Chesnutt et al.,2004; Ille et al.,2007; Bonner et al.,2008). In contrast to their proven activity in proliferation, manipulations of Wnt signaling have not consistently ascertained a role in the definition of cell identities (Megason and McMahon,2002; Chesnutt et al.,2004; Ille et al.,2007). However, some solid recent experimental data indicate that the Wnt canonical pathway is indeed implicated in the specification of neural progenitor cells' identity.
WNT/β-CATENIN AND SHH HAVE OPPOSITE EFFECTS ON NEURAL TUBE PATTERNING
Several lines of evidence show that Wnts are involved in DV patterning. Wnt1/Wnt3a double mutant mice show reduced numbers of dI1-3 dorsal-most interneurons and increased intermediate interneurons (Muroyama et al.,2002). In addition, conditional mice mutant embryos lacking β-catenin in neural progenitors lose expression of Olig3, a bHLH transcription factor key for dP1-3 dorsal progenitors, while embryos expressing a constitutively active form of β-catenin along the entire spinal cord axis show ectopic expression of Olig3 (Zechner et al.,2007). In a zebrafish heat-shock transgenic model, expression of the secreted Wnt inhibitor Dickkopf1 resulted in the loss or reduction of dorsal progenitor cell types (Bonner et al.,2008). In chick electroporation assays, the ectopic activation of the Wnt/β-catenin pathway resulted in the expansion of dorsal genes like Pax6 and Pax7 and the concomitant inhibition of ventral genes like Olig2 and Nkx2.2 (Alvarez-Medina et al.,2008). On the other hand, inhibition of Wnt/β-catenin pathway by means of the dominant-negative forms of Tcfs (DN-Tcf) resulted in the dorsal expansion of ventral markers and the suppression of dorsal identities. These changes in the patterning of neural progenitors had a correlative impact in the resulting postmitotic neurons (Alvarez-Medina et al.,2008).
Using elegant mice genetic tools, Yu et al. (2008) induce the expression of a constitutively active form of β-catenin in patches of the ventral neural tube. Consistent with the results from chick in ovo electroporation assays, the ventral expression of a stabilized form of β-catenin cell autonomously induced dorsal identities and inhibited ventral ones in a time-dependent manner. Altogether these results suggest a model where Wnt1 and Wnt3a, emanating from the dorsal aspects of the neural tube, induce dorsal while inhibiting ventral identities, and suggest that the patterning activity of Wnt/β-catenin antagonizes that of the Shh pathway (Fig. 3). In addition to its action in the spinal cord, Wnt canonical signaling induces neural dorsal identities in other regions of developing Central Nervous System such as the forebrain and the eye (Backman et al.,2005; Veien et al.,2008; Solberg et al.,2008). Different manipulations of Wnt signaling during the development of the spinal cord have not consistently reported the same results. The reasons for these discrepancies are not clear. Some evidence suggests that small differences in the timing or the antero-posterior position where the manipulations and data analysis are performed may dramatically impact on the conclusions obtained, as indicated in Bonner et al. (2008). In order to illustrate this, Bonner et al. (2008) conclude that in zebrafish embryos Wnt is required for proliferation between 12 and 18 hpf. Blocking Wnt signaling after 18 hpf has no consequences on progenitor proliferation but, in contrast, still affects D-V patterning. These suggested that careful and detailed analysis of timely controlled experimental manipulations are needed in order to allow conclusive statements.
WNT/β-CATENIN ANTAGONIZES SHH SIGNALING BY INDUCING GLI3 EXPRESSION
How does Wnt canonical signaling antagonize Shh patterning activity? Three general models, not necessarily exclusive, could account for this: (1) Wnt/β-catenin affects other signals, like Bmp/Smad or Shh/Gli, having therefore indirect consequences on patterning, (2) Wnt/β-catenin promotes dorsal identities, in parallel to BMPs, by the direct regulation of sets of patterning determinants opposite to those of Shh/Gli, and (3) Wnt/β-catenin activity mediates BMP patterning effects. As discussed below, evidence supporting the first and second models is available, while the third needs to be evaluated.
Early studies show that roof plate Bmps antagonize Shh activity in the neural tube and that Bmps are required for the correct specification of dorsal cell types (Liem et al.,1995,2000; Barth et al.,1999; Mekki-Dauriac et al.,2002; Wine-Lee et al.,2004; reviewed in Chizhikov and Millen,2005; Liu and Niswander,2005). In addition, Wnt canonical pathway induces Bmp7 expression (Alvarez-Medina et al.,2008). Altogether, this suggests a model where Wnts would induce Bmps expression, which, in turn, would specify dorsal identities. However, this seems not to be the case since overexpression of Noggin, a soluble Bmp inhibitor (Gazzerro and Canalis,2006), did not affect the capacity of the ectopic Wnt/β-catenin to induce dorsal markers like Pax7 (Alvarez-Medina et al.,2008). Moreover, in mice embryos expressing stabilized β-catenin no increase of phosphorylated Smad1/5/8, which is an indicator of Bmp signaling activity, has been observed (Yu et al.,2008). In sum, evidence supports that Wnts patterning effects are not mediated by Bmp/Smad signaling. Due to cross-talk events between Wnt and Bmp, signals have been reported in several systems including the developing spinal cord (Guo and Wang,2009; Chesnutt et al.,2004; Wine-Lee et al.,2004; Ille et al.,2007), more studies are required to define the precise relation between these signals during the D-V patterning process, as discussed below.
Based on in ovo luciferase analysis using the Gli binding sites, firefly luciferase reporter (GBS-FF), Alvarez-Medina et al. (2008) determined that Wnt/β-catenin activity inhibits Shh/Gli pathway in the developing chick spinal cord. Thus, Wnt1/3a inhibited the transcriptional activity of Gli3. Conversely, a dominant-negative form of Tcf strongly activated the GBS-FF reporter. In order to evaluate whether Wnts patterning effects were dependent on Shh signaling, Alvarez-Medina et al. (2008) used the DNA binding domain of Gli3 (Gli3-Znf) to block any endogenous activator or repressor Gli activities. In the presence of Gli3-Znf, the forced expression of Wnt1/3a did not expand dorsal markers like Pax6 or Pax7 to the ventral regions, and, conversely, the co-electroporation of DN-Tcf with Gli3-Znf did not inhibit Pax7 expression (Alvarez-Medina et al.,2008). Altogether, these data indicate that Wnt/β-catenin patterning activity was dependent, at least partially, on Gli repressor activity inhibitory of Shh signaling.
Gli3 transcription factor has largely been recognized as a potent inhibitor of the Shh pathway (Wang et al.,2000; Jacob and Briscoe,2003). In the spinal cord, Gli3 is expressed in a dorsal to ventral gradient suggesting that Gli3 activity could account for the Wnt/β-catenin patterning effects. Full-length Gli3 overexpression in chick neural tubes partially mimicked Wnt inhibitory ventral programme activity (Alvarez-Medina et al.,2008). Confirming these findings, in a mice model where stabilized β-catenin was ventrally expressed in the Olig1 domain, the ectopic induction of V2 identities at the expense of motor neurons was partially rescued with the removal of Gli3, but not with the elimination of Gli2 (Yu et al.,2008). Thus, Wnts patterning effects are dependent on Gli3 and presumably on its repressor form. This leads to a model where Wnt/β-catenin signaling would control Gli3 expression, which in turn antagonizes Shh activity, a model further supported by the fact that ectopic Wnt/β-catenin had the ability to induce Gli3 expression in chick and mouse (Alvarez-Medina et al.,2008; Yu et al.,2008). Conversely, Gli3 expression was drastically reduced in the spinal cord of Wnt1/3a double mutant mice and in the chick neural tubes electroporated with a DN-Tcf form (Alvarez-Medina et al.,2008). Supporting the idea that Wnt/β-catenin directly controls Gli3 expression, several Tcf-binding consensus sequences have been identified within highly conserved non-coding regions (HCNR) in the human Gli3 locus (Abbasi et al.,2007; Alvarez-Medina et al.,2008); two of these Tcf-binding sites containing HCNRs were responsive to Wnt signaling manipulation in neural cells and contained sufficient information to drive Gli3 expression to the dorsal NT (Alvarez-Medina et al.,2008). Nevertheless, in order to fully confirm a direct control of Wnt-β-catenin signaling on the expression of Gli3, further analysis of the binding of Tcf-β-catenin complexes to Gli3 HCNR should be performed.
In sum, these data reveal that the Wnt canonical pathway restricts Shh signaling, and influences DV patterning, by directly regulating Gli3 expression (Fig. 3). It is not clear at the moment if Wnt/β-catenin can additionally influence the processing of Gli3, a key issue that needs to be further explored. Other levels of cross-talk between Shh and Wnt signals are, of course, possible. Thus, it has been reported that Gli3 repressor forms can directly interact and inhibit β-catenin activity (Ulloa et al.,2007). To complicate the scenario, the fact that repressor Gli3 can physically interact with several Smads proteins (Liu et al.,1998), opens the possibility of the existence of an intricate regulatory network between Wnt-Bmps-Shh pathways, details of which, including their impact on D-V patterning, remain to be discovered.
Remarkably, experimental evidence indicates the existence of a Wnt-β-catenin cell specification activity independent on Gli3 (Yu et al.,2008). In support of this is the fact that loss of Gli3 does not totally reproduce the DV patterning defects seen in the Wnt/β-catenin loss of function models. For instance, the absence of dI1–3 interneurons in Wnt1/Wnt3a double mutant mice has not been reported in Gli3-deficient mice. The contribution of the Gli3-dependent versus the Gli3-independent Wnt signaling action on patterning is not clear at the moment. It is possible that the specification of the most dorsal aspects of the developing spinal cord is defined mostly by Gli3-independent Wnt signaling, while both Gli3-dependent and -independent mechanisms mediate Wnt action at intermediate and ventral regions. The precise mechanisms by which Wnts pattern the neural tube in a Gli3-independent manner are unknown although as mentioned earlier, one possibility is that Wnts directly control the expression of HD and bHLH proteins and act in parallel to Bmps (Yu et al.,2008). Other possibilities that need to be further explored are the existence of cross-talk events between Wnt signaling and other patterning signals like retinoic acid, and that some of the Bmp activity would be mediated by Wnts.
WNT/β-CATENIN DETERMINES CELL FATES AT THE VENTRAL SPINAL CORD
In order to evaluate whether endogenous Wnt canonical signaling affects cell identities at the ventral regions of the spinal cord, Yu et al. (2008) analysed the phenotype of mice where conditionally β-catenin has been inactivated within the Olig1 locus, having as a consequence the disruption of the Wnt-β-catenin signal specifically in the p3 and the pMN domains. Disruption of Wnt-β-catenin signaling in the ventral spinal cord resulted in the dorsal expansion of ventral markers like Nkx2.2. Remarkably, β-catenin deletion did not affect the production or reception of Shh, indicating that these effects may not be produced, at least not totally, as an indirect consequence of affecting the Shh/Gli pathway. Consistent with these data, Wnt signaling is active in the regions where ventral progenitors are being specified (Yu et al.,2008). In addition, other lines of evidence indicating that Wnt signaling influences cell identities at ventral regions of the spinal cord have been provided. Thus, Lei et al. (2006) have shown that the secreted Wnt inhibitor sFRP2 regulates the expression of Nkx2.2. The characterization of an Nkx2.2 enhancer revealed a region sufficient enough to account for the appropriate Nkx2.2 ventral expression. This region contains one Gli- and two Tcf-binding sites. A reporter of this enhancer with its GliBS mutated was not expressed in mice embryos. By contrast, the same reporter with mutated Tcf-binding sites showed a pattern of expression expanded dorsally relative to the control. Altogether, these data indicate that the Gli-binding site is required for Nkx2.2 expression while the Tcf sites are required to limit correctly its dorsal boundary of expression. Presumably, a repressor Tcf4 activity would be responsible for defining the dorsal boundary of Nkx2.2 expression. In agreement with this idea, the overexpression in chick neural tubes of a Tcf4 dominant-negative form, which lacked its N-terminal β-catenin binding domain, inhibited the expression of Nkx2.2 induced by a constitutively active form of Gli2 (Lei et al.,2006). This result contrasted with those reported in Alvarez-Medina et al. (2008) where the overexpression of an equivalent dominant-negative form of Tcf4 induced Nkx2.2 expression. Additionally, according to Lei et al. (2006), it is expected that positive Wnt signaling, which converts Tcf4 transcription factor from a repressor to an activator, would induce Nkx2.2 expression. A dorsal expansion of Nkx2.2 domain has been observed by Megason and McMahon (2002) in chick embryo neural tubes electroporated with a constitutively active form of β-catenin. However, as mentioned earlier, the results obtained by Alvarez-Medina et al. (2008) and Yu et al. (2008) actually indicate the opposite. An explanation to justify these contradictory observations is still not available. Conclusions from experiments using DN-tcfs should be taken with caution as they may have unspecific effects on the regulation of target genes. As discussed earlier, manipulation of Wnt signaling can result in different outcomes depending on small differences in the timing and place in which these manipulations are carried out. In this context, it is worth pointing out that members of the Wnt pathway, such as β-catenin and Tcfs, may participate in processes unrelated to Wnt canonical signaling (Gribble et al.,2009; Jin et al.,2008). Thus, the misexpression of their dominant active or negative forms may have consequences that do not necessarily correspond to the patterning action of Wnt signaling in vivo. In addition, it should not be overlooked that different Tcfs elicit specific responses, so differences in the tools employed could explain these discrepant observations as well. Supporting this last idea, Bonner et al. (2008) concluded, by morpholino analysis in zebrafish, that tcf3 is required for the proliferation while tcf7 is required for the patterning activities of Wnt signaling. In sum, this indicates that our understanding of how the spinal cord is patterned dorso-ventrally is far from complete.
CONCLUDING REMARKS AND PERSPECTIVES
The analysis of the spinal cord patterning has been very valuable in the dissection of the mechanisms by which morphogen signals like Shh induce different cell identities. The D-V spinal cord patterning is achieved by the action of several signaling molecules, mainly from the counteracting gradients of Bmp and Wnts and Shh. Recent lines of evidences indicate that Wnt canonical signaling restricts Shh patterning activity by inducing Gli3 expression. Since one of the main functions of Shh is to restrict Gli3 repressor activity, the regulation of Gli3 expression and processing appears to be pivotal in D-V patterning. However, Gli3 does not totally account for Shh and Wnt patterning functions. In the case of Shh, it is clear that Gli activator inputs are required for the specification of V3 y floor plate identities. The determination of the extension of the Wnt/β-catenin Gli3-independent patterning effects is an issue to be clarified. In addition, the real contributions to patterning of Wnts versus the Bmps, which also induce dorsal identities, need to be elucidated. Thus, it is possible that some of the effects ascribed to Bmps would be mediated by Wnt/β-catenin signaling. Moreover, although manipulation of Bmp signaling confirms an action in inducing the most dorsal spinal cord cell types, it does not result in the expected long-range restriction of Shh patterning activity in a similar way to that reported with Wnts (Garcia-Campmany and Marti,2007; Garcia-Campmany and Marti, unpublished results).
The different pathways that define the spinal cord D-V patterning show considerable levels of cross-talk between them. To complicate the scenario, their action is not restricted just to cell specification but it is extended to processes such as the control of proliferation, neurogenesis, and axon guidance. Determining how individual cells integrate all this signal information is one of the most challenging questions to address in the future.
The authors thank Lidia Garcia-Campamany and Roberto Alvarez-Medina for sharing unpublished results and for valuable discussions. Work in the EM lab is supported by the Spanish Ministry of Education, grant BFU2007-60487/BMC.