LDL receptor-related proteins 5 and 6 in Wnt/β-catenin signaling:Arrows point the way

Signaling by the Wnt family of secreted growth factors has key roles in development and disease (Wodarz and Nusse, 1998; Veeman et al.,2003). Since the discovery of Wnt1 as an oncogene that causes mouse mammary tumorigenesis (Nusse and Varmus, 1982), the Wnt gene family, which includes 19 members in the human genome, has been found in all animal species examined. Wnt proteins regulate many stages of development, from patterning of the embryo and generation of tissues and cell types, to regulation of cell movements,polarity, axon guidance and synapse formation(Nusse, 2003; Packard et al., 2003; Strutt, 2003; Veeman et al., 2003). Defective Wnt signaling plays major roles in diseases such as cancer(Bienz and Clevers, 2000; Polakis, 2000) and osteoporosis (Patel and Karsenty,2002). Therefore, the investigation of Wnt signal transduction is crucial for understanding development and disease.

The interaction of Wnt proteins with their receptors on the cell surface is the first step in transducing the extracellular signal into intracellular responses. The first identified Wnt receptors were members of the Frizzled(Fz) family of seven-pass transmembrane receptors(Wodarz and Nusse, 1998), 10 of which are encoded in the human genome. In addition to Fz proteins, the canonical Wnt/β-catenin signaling pathway requires single-span transmembrane proteins that belong to a subfamily of low-density-lipoprotein(LDL) receptor related proteins (LRPs): vertebrate Lrp5 and Lrp6, and their Drosophila ortholog Arrow (Pinson et al., 2000; Tamai et al.,2000; Wehrli et al.,2000), which are the focus of this review.

We will discuss the structure and function of Arrow/Lrp5/Lrp6, their interactions with Wnt, Fz and the intracellular β-catenin signaling apparatus, their biogenesis and modulation by extracellular antagonists, and,finally, the roles of LRP5 mutations in human diseases. Because of their related biochemical properties in Wnt signaling, we will often, unless otherwise specified, use Lrp5/Lrp6 to refer Lrp5 and Lrp6 together in the discussion. For an overview of Wnt signaling, including accounts of various Wnt transduction pathways and components, readers may refer to many excellent reviews (Wodarz and Nusse,1998; Adler, 2002; Huelsken and Behrens, 2002; Nusse, 2003; Strutt, 2003; Veeman et al., 2003).

The outcome of the most intensively studied Wnt pathway – the canonical Wnt/β-catenin signaling pathway – relies to a large extent on the regulation of the stability/abundance of the β-catenin protein (Fig. 1). It is widely accepted that in this Wnt pathway β-catenin associates with, and acts as an obligatory nuclear co-activator for, the TCF/LEF (T cell factor/Lymphoid enhancer factor) family of transcription factors(Bienz and Clevers, 2003; Cong et al., 2003; Tolwinski and Wieschaus,2004). In the absence of a Wnt ligand, the level of cytosolicβ-catenin is kept low as a result of its amino-terminal phosphorylation-dependent ubiquitination/proteosome degradation. Whenβ-catenin is low, TCF/LEF is associated with transcriptional co-repressors and suppresses Wnt-responsive gene expression(Fig. 1A). Upon Wnt stimulation, β-catenin phosphorylation and degradation is inhibited, and the accumulation of β-catenin promotes its association with TCF/LEF,leading to the activation of Wnt-responsive transcription(Fig. 1B).

β-catenin phosphorylation is thus a crucial regulatory step in this Wnt pathway. β-Catenin phosphorylation involves the sequential actions of casein kinase 1 (Ck1) and glycogen synthase kinase 3 (Gsk3), and takes place in a protein complex assembled by the scaffolding protein Axin and the tumor suppressor protein Apc, the adenomatous polyposis coli gene product(Fig. 1A). Phosphorylatedβ-catenin is recognized by the F-box protein β-Trcp, which is a part of an ubiquitin-ligase complex that conjugates β-catenin with ubiquitin for proteosome degradation (Polakis,2002; Wu et al.,2004). This phosphorylation-degradation process is often perturbed in human colorectal cancers and other tumors, either by a loss of functional Apc or Axin, or by β-catenin mutations that prevent its phosphorylation and degradation, thereby resulting in constitutive Wnt/β-catenin signaling (Bienz and Clevers,2000; Polakis,2000).

How Wnt inhibits β-catenin phosphorylation is not well defined, but may be achieved by any of, or combinations of, the following(Fig. 1B): (1) degradation of the Axin protein (Willert et al.,1999; Yamamoto et al.,1999; Mao et al.,2001b; Tolwinski et al.,2003); (2) alteration of the composition of the Axin complex (such as by dissociation of Gsk3 or β-catenin from Axin)(Kishida et al., 1999a; Willert et al., 1999; Li et al., 1999; Itoh et al., 2000); (3)binding of the Gsk3 binding protein (GBP) to Gsk3(Farr et al., 2000; Salic et al., 2000); and (4)inhibition of Gsk3 (or Ck1) kinase activity(Cook et al., 1996). A recent genetic study in Drosophila(Tolwinski et al., 2003),however, suggested that Wnt/β-catenin signaling can occur in a Gsk3-independent manner (Box 1). Another cytoplasmic protein Dishevelled has a crucial but poorly understood role in the inhibition of the Axin/Gsk3 complex, and is discussed in more detail in later sections.

The Wnt-Fz ligand-receptor relationship is best characterized in Drosophila. Studies using cultured Drosophila cells indicate that two members of the Fz family, Dfz1 and Dfz2 (Fz and Fz2 – FlyBase),are Wingless (Wg, Drosophila Wnt1) receptors, which bind Wg with high affinities (K_d=10^–8 M and 10^–9 M for Dfz1 and Dfz2, respectively) (Bhanot et al., 1996; Rulifson et al.,2000). Fly mutants lacking both Dfz1 and Dfz2,but not mutants lacking either, have severely defective Wg signaling(Bhat, 1998; Kennerdell and Carthew, 1998; Bhanot et al., 1999; Chen and Struhl, 1999; Muller et al., 1999),providing unambiguous evidence that, in many contexts, Dfz1 and Dfz2 are redundant Wg receptors.

Studies in nematodes, Xenopus and mammalian cells support the Wnt-Fz ligand-receptor relationship (Sawa et al., 1996; Yang-Snyder et al., 1996; He et al.,1997; Rocheleau et al.,1997; Thorpe et al.,1997; Hsieh et al.,1999). However, the specificity of Wnt-Fz interactions remains largely unresolved, particularly in vertebrates, because of difficulties in producing soluble Wnt proteins, the large numbers of Wnt and Fz genes, and the potential multitudes of Wnt-Fz interactions and functional redundancies. Nevertheless, recent success in purification of the first Wnt protein, mouse Wnt3a, may help to pave the way for comprehensive biochemical studies of Wnt-Fz interactions in vitro (Willert et al., 2003).

The roles of Arrow and Lrp6 in Wnt signaling were discovered via genetic studies. Drosophila mutants lacking arrow phenotypically resemble the wg mutant (Wehrli et al., 2000), and mutant mice lacking Lrp6 exhibit composite phenotypes similar to mutations of several individual Wnt genes(Pinson et al., 2000). In Xenopus embryos, dominant-negative Lrp6 blocks signaling by several Wnt proteins, whereas overexpression of Lrp6 cooperates with Wnt and Fz to activate Wnt/β-catenin signaling(Tamai et al., 2000). Furthermore, Arrow and Lrp6 are required for cells to respond to Wnt, and act upstream of known intracellular Wnt signaling components in Drosophila and Xenopus(Tamai et al., 2000; Wehrli et al., 2000),pinpointing a role for Arrow/Lrp6 in Wg/Wnt signal reception. A recent genetic study in mice indicates that Lrp5 also has a role in Wnt signaling(Kelly et al., 2004).

Unlike Fz, which is required for multiple Wnt pathways(Strutt, 2003; Veeman et al., 2003), Arrow and Lrp6 appear to be specifically required for Wnt/β-catenin signaling. In Drosophila, Dfz1, but not Dfz2, plays a central role in planar cell polarity (PCP) determination (reviewed by Adler, 2002; Strutt, 2003), whereas arrow mutants exhibit normal PCP(Wehrli et al., 2000). This indicates that Arrow is not required for Dfz1 PCP function. Similarly, in Xenopus, blocking Lrp6 function has little effect on gastrulation movements (Semenov et al.,2001), which are regulated by a Wnt11/Fz pathway analogous to Dfz1/PCP signaling (Heisenberg et al.,2000; Tada and Smith,2000; Wallingford et al.,2000).

Lrp5 and Lrp6 are highly homologous, and are widely co-expressed during embryogenesis and in adult tissues (Dong et al., 1998; Hey et al.,1998; Kim et al.,1998; Pinson et al.,2000; Houston and Wylie,2002; Kato et al.,2002; Fujino et al.,2003; Kelly et al.,2004). Lrp6^–/– mice are perinatal lethal and exhibit mid/hindbrain defects, posterior truncation and abnormal limb patterning, which resemble the defects of mice mutant for Wnt1,Wnt3a and Wnt7a (Pinson et al., 2000). Lrp5^–/– mice have normal embryogenesis, grow to adulthood and are fertile, but show osteoporosis(Kato et al., 2002) and some metabolic abnormalities (Magoori et al., 2002; Fujino et al., 2003). Thus,the Lrp6 loss-of-function phenotype is much more severe than the Lrp5 loss-of-function phenotype, indicating that Lrp6 has a more influential role than Lrp5 during embryogenesis. There is a hint of functional redundancy between Lrp5 and Lrp6, as defects in Lrp6^–/–embryos are less severe than those observed in individual Wnt mutants(Pinson et al., 2000). Indeed, Lrp5^–/–;Lrp6^–/– double mutant mice die during gastrulation(much earlier than Lrp6^–/– mice); they lack the primitive streak and nascent mesoderm(Kelly et al., 2004), and thus phenotypically resemble the Wnt3 mutant(Liu et al., 1999). Thus,genetic studies in mice indicate that Wnt3 requires either Lrp5 or Lrp6 in order to function, whereas Wnt1, Wnt3a and Wnt7a rely on Lrp6 primarily and Lrp5 to some extent. Whether signaling by other Wnt proteins requires Lrp5 and/or Lrp6 cannot be inferred at the moment due to the early lethality of double Lrp5^–/–;Lrp6^–/– mutants and the degree of redundancy between the two proteins. It seems unlikely that Lrp5 and Lrp6 are involved in signaling by distinct Wnt proteins during development (note that Lrp5^–/– mice have normal embryogenesis),although more genetic studies are needed to clarify this issue. It is also unclear whether, in Drosophila, Arrow is required for signaling by other Wnt proteins besides Wg.

An allelic series of compound mutants reveal the following order of severity of developmental abnormalities: Lrp5^+/–(normal)<Lrp6^+/–<Lrp5^–/–<Lrp5^+/–;Lrp6^+/–<Lrp5^–/–;Lrp6^+/–<Lrp6^–/–<Lrp5^+/–;Lrp6^–/–<Lrp5^–/–;Lrp6^–/– (Kelly et al., 2004). This is likely to reflect the severity of loss of Wnt/β-catenin signaling, and supports the view that Lrp5 and Lrp6 share significant overlapping functions but that Lrp6 plays a more crucial role, at least during embryogenesis. This is consistent with observations that overexpression of Lrp6 exhibits significantly stronger activity than Lrp5 in Xenopus (Tamai et al.,2000) and mammalian cells(Holmen et al., 2002). It is possible that Lrp6 has a higher affinity for Wnts or a stronger signaling efficacy than Lrp5, or both.

Arrow/Lrp5/Lrp6 is a subfamily of the LDL receptor (LDLR) family(Fig. 2), which has diverse roles in metabolism and development (Herz and Bock, 2002). Human LRP5 was isolated through its homology to LDLR (Dong et al., 1998; Hey et al., 1998; Kim et al., 1998). Human LRP6 was identified by its homology to LRP5(Brown et al., 1998). Arrow/LRP5/LRP6 are type I single-span transmembrane proteins with 1678, 1615 and 1613 amino acid residues, respectively. LRP5 and LRP6 share 73% and 64%identity in extracellular and intracellular domains, respectively, whereas Arrow is equally related (40% identical) to LRP5 and LRP6(Box 2, Box 3, Fig. S1). Indeed, LRP6 substitutes for Arrow during Wg signaling in cultured Drosophila cells (Schweizer and Varmus, 2003), and constitutively activated Arrow (discussed later) activates Wnt/β-catenin signaling in mammalian cells and Xenopus embryos (Tamai et al.,2004).

The simplest model to account for the role of Arrow/Lrp5/Lrp6 in Wnt signaling is that Arrow/Lrp5/Lrp6 are Wnt coreceptors. Nevertheless, a few issues concerning this view remain:

1. Do Arrow/Lrp5/Lrp6 bind Wnt? Co-immunoprecipitation (co-IP) experiments suggest that Lrp5/Lrp6 can bind Wnt1, Wnt3a, Wnt4 and Xenopus Wnt8(Xwnt8; Table 1)(Tamai et al., 2000; Mao et al., 2001b; Kato et al., 2002; Itasaki et al., 2003; Liu et al., 2003), although Wnt-Lrp5/Lrp6 binding affinities have not been reported, perhaps because of a weaker binding between Wnt and Lrp5/Lrp6 than that between Wnt and Fz. The first two YWTD β-propeller plus EGF-like domains of Lrp6(Fig. 2; see also Fig. 6) are involved in binding Xwnt8 (Itasaki et al., 2003),and, in mammalian cells, these domains are required for Lrp6 to cooperate functionally with Wnt/Fz (Mao et al.,2001a). Additionally, an artificially-created secreted Lrp5 extracellular domain blocks Wnt signaling, presumably by competing away Wnt ligands (Mao et al., 2001b; Gong et al., 2001)(Fig. 3). We note that Wnt proteins are palmitoylated (i.e. covalently modified by a lipid moiety) and are, therefore, lipoproteins (Willert et al., 2003; Nusse, 2004). It may make sense intuitively that they are ligands for Lrp5/Lrp6, which, as members of the LDLR family, may be specialized in binding lipoproteins. However, in Drosophila, Wg binding to Arrow has not been demonstrated(Wu and Nusse, 2002)(Table 1).

2. Does Arrow/Lrp5/Lrp6 associate with Fz? Some evidence suggests so. mFz8CRD,which is the cysteine-rich domain of mouse Fz8, and which binds Wnt (Hsieh et al.,1999), exhibits a Wnt1-dependent association with both the Lrp5 and Lrp6 extracellular domains in vitro(Tamai et al., 2000; Semenov et al., 2001),leading to the proposal that Wnt induces a complex formation between Fz and Lrp5/Lrp6 (Fig. 1). Indeed,forced proximity of Dfz2 and Arrow, by fusing the Arrow intracellular domain with the Dfz2 cytoplasmic tail (Fig. 3), activates β-catenin signaling in a Wg-independent manner(Tolwinski et al., 2003),implying that Wg brings Arrow and Dfz2 together. However, attempts to detect a complex of Wg, Dfz2CRD and the Arrow extracellular domain have not so far been successful (Wu and Nusse,2002).

Further experiments will be needed to substantiate any Fz-Lrp5/Lrp6 association. First, can Wnt-Lrp5/Lrp6 affinities be measured? Second, can an in vitro Wnt-Fz-Lrp5/Lrp6 (extracellular domains) complex be observed for Wnt and Fz proteins other than just Wnt1 and Fz8? Third, can a complex between wild-type Fz and Lrp5/Lrp6 be detected at the plasma membrane? Finally,although the simplest version of the coreceptor model is that Lrp5/Lrp6 and Fz bind independently to Wnt, it remains to be examined whether Fz binding to Wnt enhances the Wnt-Lrp5/Lrp6 interaction, and/or vice versa. Answering these questions may not be simple, however, because Wnt-Fz specificity itself is not well understood, and, with the possible exception of a few Wnts, it is not clear which of the 19 Wnts and 10 Fzs engage Lrp5/Lrp6-dependent Wnt/β-catenin signaling.

Given that Wnt/β-catenin signaling requires both Fz and Arrow/Lrp5/Lrp6, an obvious question to ask is what role do these two distinct receptors play in Wnt signal transduction? A mutant Lrp6 protein lacking the intracellular domain is completely inactive, and in fact blocks Wnt and Fz signaling in a dominant-negative fashion(Tamai et al., 2000)(Fig. 3). Conversely, mutant Arrow/Lrp5/Lrp6 proteins that lack the extracellular domain (but are anchored on the membrane), referred to here as Arrow/Lrp5/Lrp6ΔN(Fig. 3), activateβ-catenin signaling constitutively in mammalian cells(Mao et al., 2001a; Mao et al., 2001b; Liu et al., 2003) and in Xenopus embryos (Tamai et al.,2004), suggesting that Arrow/Lrp5/Lrp6 have a signaling capacity that is normally suppressed by the extracellular domain. The Lrp5 intracellular domain anchored to the plasma membrane via myristylation (a covalent lipid modification that targets proteins to the plasma membrane) is also constitutively active (Fig. 3) (Mao et al.,2001b). Thus, Arrow/Lrp5/Lrp6 is a key signaling receptor for the Wnt/β-catenin pathway.

An important insight into the function of Lrp5 came from the finding that the Lrp5 intracellular domain binds Axin in both yeast two-hybrid and co-IP assays (Mao et al., 2001b), an observation that has been extended to Arrow(Tolwinski et al., 2003) and Lrp6 (Liu et al., 2003; Tamai et al., 2004). As mentioned previously, Axin is a scaffolding protein that contains binding sites for Apc, β-catenin, Gsk3, Ck1 and possibly other proteins(Polakis, 2002; Kikuchi, 1999)(Fig. 4). Axin nucleates this Axin complex, resulting in β-catenin phosphorylation and degradation(Fig. 1). Thus, the binding between the Arrow/Lrp5/Lrp6 intracellular domain and Axin permits the Wnt co-receptors to directly control β-catenin phosphorylation and degradation. The domain of Axin involved in binding Arrow/Lrp5 has only been mapped via the yeast two-hybrid assay and remains poorly defined(Fig. 4). The DIX domain of Axin is necessary, but not sufficient, for the Axin-Arrow/Lrp5 interaction,whereas the RGS domain may be inhibitory to it(Mao et al., 2001b; Tolwinski et al., 2003).

Wnt stimulation of mammalian cells induces the Lrp5-Axin association within minutes (Mao et al., 2001b). In Drosophila embryos, Axin-GFP is recruited to the plasma membrane only in Wg-responsive cells, and this may occur through Arrow, although direct evidence is lacking (Cliffe et al.,2003). How does Wnt promote the Arrow/Lrp5-Axin interaction?

A recent study discovered that a PPP(S/T)P motif, which is reiterated five times in Arrow/Lrp5/Lrp6 intracellular domains(Box 3, see Fig. S1), is the minimal module that is necessary and sufficient for Lrp6 signaling function in mammalian cells and Xenopus embryos(Tamai et al., 2004). When a single PPPSP motif is transferred artificially to a truncated LDLR protein(which has no role in Wnt/β-catenin signaling) it becomes phosphorylated and can fully activate Wnt/β-catenin signaling(Tamai et al., 2004)(Fig. 3). Importantly, Axin preferentially binds to the phosphorylated PPPSP motif, whose phosphorylation in Lrp6 is rapidly induced by Wnt (Tamai et al., 2004). It was thus proposed that Wnt activates Lrp6 signaling by inducing Lrp6 phosphorylation at the PPP(S/T)P motifs, which serve as inducible docking sites for Axin, thereby recruiting Axin to the plasma membrane (Fig. 5). This model is also likely to apply to Arrow and Lrp5, which share conserved PPP(S/T)P motifs (see Fig. S1). The phosphorylation-dependent activation of Lrp6 and its inducible recruitment of Axin is reminiscent of other types of transmembrane signaling, such as that by tyrosine kinase receptors and cytokine receptors(Pawson and Scott, 1997).

These results suggest the existence of at least two unidentified components that control Wnt activation of Arrow/Lrp5/Lrp6: a kinase that phosphorylates the PPPSP motif and a phosphatase that dephosphorylates it. As Wnt induces Lrp6 phosphorylation (Tamai et al.,2004), Wnt may activate the PPPSP kinase or inhibit the PPPSP phosphatase. Alternatively, Wnt may regulate the access of Lrp6 to this kinase or phosphatase, perhaps by inducing a `conformational change' in Lrp6. Indeed,a recent study suggests that the oligomerization of Lrp6 may keep Lrp6 in an inactive state, and Wnt binding to Lrp6 may induce a conformational change that activates the Lrp6 oligomer (Liu et al., 2003).

In mammalian cells, Gsk3 overexpression enhances the Lrp5/Lrp6-Axin interaction in co-IP experiments (Mao et al., 2001b; Liu et al.,2003). This finding is puzzling as Gsk3 antagonizes Wnt/β-catenin signaling. In Drosophila mutant embryos lacking Gsk3/zw3, Wg recruitment of Axin to the plasma membrane and Wg signaling to Axin can occur (Cliffe et al., 2003; Tolwinski et al.,2003), implying that the Arrow-Axin association is not defective in the absence of Gsk3. Then why does Gsk3 overexpression enhance Lrp5/Lrp6-Axin interaction in mammalian cells? One possibility may be that Lrp5/Lrp6 signaling destabilizes Axin (see below), whereas Gsk3 can phosphorylate and stabilize Axin (Willert et al., 1999; Yamamoto et al., 1999). An increase in the level of Axin following Gsk3 overexpression may explain `enhanced' Lrp5/Lrp6-Axin binding. Another possibility is that Lrp5/Lrp6 preferentially interact with Axin phosphorylated by or complexed with Gsk3 (Mao et al.,2001b). Additional explanations may include that Gsk3 overexpression mimicks the action of the PPPSP kinase.

How does Arrow/Lrp5/Lrp6 activation and binding to Axin initiateβ-catenin signaling? One possibility is that Arrow/Lrp5/Lrp6 binding to Axin promotes Axin degradation. Indeed, Wnt stimulation(Willert et al., 1999; Yamamoto et al., 1999), Wg overexpression (Tolwinski et al.,2003), and Lrp5ΔN overexpression(Mao et al., 2001b), all induce Axin degradation in mammalian and Drosophila cells (see also Cliffe et al., 2003). These studies demonstrated reductions in Axin protein level after 2-4 hours of Wnt stimulation (or longer when transfection or transgenic experiments are involved). However, β-catenin stabilization is detectable within 30 minutes of Wnt stimulation and can thus occur before an obvious reduction in levels of Axin (Willert et al.,1999). One explanation, according to a recent theoretical and experimental analysis (Lee et al.,2003), is that a slight decrease in Axin protein level may have a significant effect on β-catenin phosphorylation and degradation.

It is also possible that Arrow/Lrp5/Lrp6 binding inhibits the activity of the Axin complex, by altering its component composition. This may be important in the early phase of Wnt signaling when Axin degradation is insignificant. In any event, one should keep in mind that, in mammalian cells,Wnt/β-catenin signaling induces the expression of an Axin homolog, Axin2(also known as Axil/Conductin) (Behrens et al., 1998; Yamamoto et al.,1998; Yan et al.,2001; Jho et al.,2002; Leung et al.,2002; Lustig et al.,2002). Thus, a reduction in the Axin protein level is likely to be accompanied by an increase in the level of Axin2 during Wnt signaling, thereby complicating the Axin degradation scenario. How Arrow/Lrp5/Lrp6 promotes Axin degradation or inhibits Axin function is unknown, but their recruitment of Axin to near the plasma membrane appears essential. Indeed, the Lrp5 intracellular domain, although capable of binding Axin, is incapable of signaling in mammalian cells unless it is anchored to the plasma membrane(Mao et al., 2001b).

Overexpression of Arrow/Lrp5/Lrp6ΔN, or even a single PPPSP motif(tethered to the LDLR; Fig. 3),constitutively activates the β-catenin pathway(Mao et al., 2001a; Mao et al., 2001b; Liu et al., 2003; Tamai et al., 2004), probably in a Wnt- and Fz-independent manner. However, this is difficult to verify as a cell completely lacking Fz proteins may not exist. Wnt and Fz might normally function to activate the signaling activity of Arrow/Lrp5/Lrp6. Given the activated nature of the truncated Arrow/Lrp5/Lrp6ΔN, a scenario in which Wnt/Fz induces post-translational cleavage of Arrow/Lrp5/Lrp6 is attractive yet lacks experimental evidence. If Arrow/Lrp5/Lrp6 were to be activated in this way, the cleavage would have to occur extracellularly because an Lrp5/Lrp6 intracellular domain that is not anchored to the membrane is inactive (Mao et al.,2001b).

Fz function remains a mystery. Fz is thought to have a signaling role because its intracellular regions are required for Wnt/β-catenin signaling (Umbhauer et al.,2000). Fz proteins also play a key part inβ-catenin-independent signaling, such as in the PCP pathway, in Ca²⁺/PKC (protein kinase C) signaling, and perhaps in other pathways (Adler, 2002; Strutt, 2003; Veeman et al., 2003). Although some Fz functions, such as PKC activation, can be blocked by pharmacological inhibitors of the trimeric G proteins, whether Fz function during Wnt/β-catenin signaling relies on G proteins remains debatable(reviewed by Malbon et al.,2001). A protein that is required for most, if not all, Fz functions is Dishevelled (Dsh in Drosophila and Xenopus, and Dvl1-3 in mammals), another mysterious protein that is genetically defined downstream of Fz in both Wnt/β-catenin and PCP pathways(Boutros and Mlodzik, 1999),and that may also be required for Fz activation of PKC(Sheldahl et al., 2003). Dsh/Dvl is a modular scaffolding protein that contains a DIX domain (which also exists in Axin), a PDZ domain (a domain discovered in PSD, Discs-large, and ZO1 proteins) and a DEP domain (a domain discovered in Dishevelled, Egl-10, and Pleckstrin proteins) (Boutros and Mlodzik, 1999; Wharton,2003). Dsh/Dvl is recruited to the plasma membrane upon overexpression of a number of different Fz proteins(Axelrod et al., 1998; Boutros et al., 2000; Rothbacher et al., 2000; Umbhauer et al., 2000), and may bind directly to the Fz carboxyl-terminal region via the PDZ domain(Chen et al., 2003; Wong et al., 2003). However,Fz recruitment of Dsh/Dvl to the plasma membrane does not correlate fully with the activation of Wnt/β-catenin signaling(Axelrod et al., 1998; Rothbacher et al., 2000; Umbhauer et al., 2000), and Dsh is not localized near the plasma membrane in Wg-responsive cells in fly embryos (Axelrod et al., 1998; Axelrod, 2001; Cliffe et al., 2003).

Thus, it remains unclear how Fz and Dsh/Dvl fit into the scenario in which Arrow/Lrp5/Lrp6 binding to Axin initiates β-catenin signaling. Three models can be proposed. Dsh appears to be epistatic to, or downstream of,Arrow, because Dsh overexpression activates β-catenin signaling in arrow mutants (Wehrli et al.,2000) and the constitutively active Dfz2-Arrow fusion protein is inactive in dsh mutants(Tolwinski et al., 2003). In addition, Dsh/Dvl can associate with and inhibit Axin(Fagotto et al., 1999; Kishida et al., 1999b; Li et al., 1999; Smalley et al., 1999; Salic et al., 2000). Therefore, one scenario, referred to here as the `co-recruitment' model, is that Fz and Arrow/Lrp5/Lrp6 recruit Dsh/Dvl and Axin into the coreceptor complex, respectively, thereby bringing Dsh/Dvl and Axin into proximity for effective Axin inhibition or degradation(Fig. 5A). However, a lack of correlation between Dsh/Dvl plasma membrane localization and Wg/Wnt signaling poses difficulties for this model, although it is possible that a small fraction of Dvl/Dsh recruited to the membrane, albeit undetectable, is sufficient for signaling. This model implies that Dvl/Dsh has a key role in Axin inhibition or degradation, and could account for the finding that overexpression of Dsh activates β-catenin signaling in arrowmutant flies (i.e. via inhibiting or degrading Axin in the cytoplasm).

The second scenario, referred to here as the `vesicle-transport' model(Cliffe et al., 2003)(Fig. 5B), is based on the observations that Dsh/Dvl and Axin, upon overexpression, are co-localized in intracellular `dots' that may represent `vesicles'(Axelrod et al., 1998; Fagotto et al., 1999; Kishida et al., 1999; Smalley et al.,1999; Axelrod,2001; Cliffe et al.,2003), and that Axin recruitment to the plasma membrane requires Dsh (Cliffe et al., 2003). This model proposes that Dsh/Dvl, through association with vesicles and Axin,shuttles Axin to the plasma membrane, where it becomes associated with Arrow/Lrp5/Lrp6. This view is consistent with the observation that the Dvl DIX domain, which is essential for Dsh/Dvl function in β-catenin signaling,harbors phospholipid-binding activity and mediates vesicle association(Capelluto et al., 2002), but it does not easily explain how, in Drosophila, Dsh overexpression activates β-catenin signaling in arrow mutants.

The third scenario, which perhaps can be referred to as a `parallel signaling' model (Fig. 5C),implies that Fz-Dsh/Dvl-Axin and Arrow/Lrp5/Lrp6-Axin represent two parallel branches: overactivation of either branch is sufficient to activateβ-catenin signaling, whereas simultaneous activation of both is required under physiological conditions. This model, which can explain why Dsh overexpression bypasses Arrow function, is based on observations that Lrp5/Lrp6ΔN signaling does not seem to be affected by depletion of Dvl/Dsh proteins in mammalian and Drosophila cells [from short interfering RNA (siRNA) or RNA interference (RNAi) experiments](Li et al., 2002; Schweizer and Varmus, 2003). These observations apparently contradict the finding in Drosophilaembryos that signaling by the constitutively active Dfz2-Arrow fusion protein(Fig. 3) requires Dsh(Tolwinski et al., 2003),although it is possible that the mechanisms by which Lrp5/Lrp6ΔN and the Dfz2-Arrow fusion protein become constitutively active may be different. However, depletion of the three Dvl proteins or Dsh through siRNA/RNAi is unlikely to be complete, rendering the interpretation less straightforward. Whether ArrowΔN, which is constitutively active in mammalian cells and Xenopus embryos (Tamai et al.,2004), can activate β-catenin signaling in dshmutant flies will be a key test. Given the possible Fz-Lrp5/Lrp6 and Dsh/Dvl-Axin interactions, these two parallel branches, if they exist, may nonetheless operate in physical proximity.

Finally, because Wnt activates Lrp6 signaling by inducing Lrp6 phosphorylation at the PPP(S/T)P motif(Tamai et al., 2004), it could be possible that Fz/Dsh signaling acts by activating or recruiting the PPPSP kinase to phosphorylate Lrp6. We consider this scenario to be less likely as Dsh overexpression can activate β-catenin signaling in the absence of Arrow function (Wehrli et al.,2000).

Wnt signaling is essential for many aspects of nematode development. However, some Wnt pathways in worms are organized differently to those in Drosophila and vertebrates(Korswagen, 2002). Nonetheless, a canonical Wnt/β-catenin signaling pathway controlling neuronal migration was discovered in worms that involves Wnt, Fz, Dsh, Axin,Apc, Gsk3 and β-catenin in a similar way as in flies and vertebrates(Korswagen et al., 2002). Perplexingly however, no Arrow/Lrp5/Lrp6 homologs have been identified in the worm genome, although other Lrp genes (such as Lrp1) exist. Thus either a functional homolog of Arrow/Lrp5/Lrp6 has yet to be discovered, or nematodes use other means for Wnt/Fz/Dsh to activate β-catenin signaling in the absence of Arrow/Lrp5/Lrp6-Axin interaction. This latter possibility shares some resemblance to the `parallel signaling' model discussed above.

Lrp5/Lrp6 are subjected to modulation by secreted antagonistic/modulatory ligands in vertebrates and by other types of regulations(Box 4). Two families of such ligands have been identified: the Dickkopf (Dkk) family and the Wise family,which antagonize Wnt/β-catenin signaling through interactions with Lrp5/Lrp6. Dkk and Wise homologs have not been found in invertebrate genomes.

Wnt signaling is required for posterior patterning in vertebrates; thus,inhibition of Wnt signaling permits anterior development(Niehrs, 1999). Xenopus Dkk1 was isolated as a head-inducing molecule and behaves as an antagonist for Wnt signaling (Glinka et al., 1998). Genetic analysis of Dkk1^–/– mice, which lack head formation, is consistent with this view (Mukhopadhyay et al., 2001). Distinct from several families of secreted Wnt antagonists that bind Wnts, including the sFRP (secreted Frizzled-related protein) family, Wif1 (Wnt inhibitory factor 1) and Xenopus Cerberus(Semenov and He, 2003), Dkk1 does not bind Wnt but is a high affinity ligand for Lrp6(K_d=0.3-0.5 nM) and Lrp5 (Bafico et al., 2001; Mao et al.,2001a; Semenov et al.,2001). Dkk1 disrupts the Fz-Lrp5/Lrp6 complex formation induced by Wnt1 in vitro (Semenov et al.,2001), suggesting that Dkk1 inhibits Wnt signaling by preventing Fz-Lrp5/Lrp6 complex formation. The Dkk1-Lrp5/Lrp6 antagonistic relationship is supported by mouse genetic studies(MacDonald et al., 2004). Thus, reducing the dosage of Lrp5 or Lrp6 can significantly rescue phenotypes associated with a loss of Dkk1 function, and vise versa. For example, while Dkk1^–/– mutant mice lack head formation and die during embryogenesis, Dkk1^–/–;Lrp6^+/– mice have extensive head development and can survive to postnatal stages (MacDonald et al., 2004).

By inhibiting Lrp5/Lrp6, Dkk1 appears to be a specific antagonist for Wnt/β-catenin signaling (Semenov et al., 2001), and is thus distinct from sFRPs, Wif1 and Cerberus,which may antagonize multiple Wnt pathways(Semenov and He, 2003). Of the two conserved cysteine-rich domains of Dkk1 (see Fig. S2),the carboxyl one is essential for its binding to Lrp6 and its antagonization of Wnt signaling, whereas the amino terminal one may exert some undefined regulatory roles (Brott and Sokol,2002; Li et al.,2002; Mao and Niehrs,2003). Dkk1 may interact with a region encompassing the third and fourth YWTD β-propeller-EGF-like domains of Lrp6, which is distinct from the Wnt-binding region (Mao et al.,2001a; Itasaki et al.,2003) (Fig. 6).

In Xenopus and mammals, the Dkk family includes Dkk1, Dkk2, Dkk3 and Dkk4 (see Fig. S2),which exhibit distinct and dynamic expression patterns(Glinka et al., 1998; Monaghan et al., 1999) and may have distinct properties. Dkk1 and Dkk4 are antagonists for Wnt signaling(Krupnik et al., 1999; Brott and Sokol, 2002; Mao and Niehrs, 2003), whereas Dkk2 can, paradoxically, inhibit or activate (albeit weakly) β-catenin signaling, depending on the experimental assays employed(Wu et al., 2000; Brott and Sokol, 2002; Li et al., 2002). Whether Dkk2 can function as a Wnt agonist in vivo remains to be seen. Dkk3 neither binds Lrp5 or Lrp6, nor affects Wnt signaling(Krupnik et al., 1999; Mao and Niehrs, 2003).

Dkk1 also binds to vertebrate Kremen (Krm) 1 and Krm2, two related single-pass transmembrane proteins (Mao et al., 2002) (see Fig. S2). In mammalian cells, either Krm1 or Krm2 can cooperate with Dkk1 in the inhibition of Wnt-Fz-Lrp6 function (Mao et al., 2002). Drosophila has no Dkk or Krm homologs,although ectopic expression of vertebrate Dkk1 and Krm2together, but not either of these genes alone, results in inhibition of Wg signaling (Mao et al., 2002). In addition, antisense knockdown of both Krm1 and Krm2 (but not either individually) in Xenopus results in deficient head development,similar to phenotypes of embryos with no or reduced Dkk1 function(Davidson et al., 2002). Thus,Krm1 and Krm2 appear to have redundant roles in Dkk1 function. Because Dkk1 can stimulate Lrp6 internalization upon Krm2 overexpression(Mao et al., 2002), it was proposed that Dkk1, by binding both Lrp6 and Krm, induces Lrp6 internalization from the cell surface, thereby attenuating Wnt signaling. Perplexingly however, the Krm intracellular domain is neither conserved nor required for any of these functions (Mao et al.,2002), which raises the question of how can Krm have a key function in Lrp6 internalization? This internalization model is different,although not mutually exclusive, from a model in which Dkk1 functions by preventing Fz-Lrp6 complex formation(Semenov et al., 2001).

Wise was identified in Xenopus embryo assays as a secreted molecule with dual properties somewhat similar to Dkk2. Wise is an antagonist for Xwnt8/β-catenin signaling, but on its own can weakly activateβ-catenin signaling (Itasaki et al.,2003). Wise binds to the Lrp6 extracellular domain in co-IP experiments, in particular to the first two YWTD β-propeller-EGF-like domains (Fig. 6), the same region that Wnt appears to bind, and can compete with Xwnt8 for Lrp6 binding(Itasaki et al., 2003). Wise belongs to a large family of secreted `cysteine-knot' domain-containing proteins (see Fig. S2),which include members that bind and antagonize BMPs (bone morphogenetic proteins) (Hsu et al., 1998; Pearce et al., 1999; Piccolo et al., 1999). Indeed,Wise was also isolated as a BMP inhibitor(Laurikkala et al., 2003). Thus, Wise appears to be a multifunctional inhibitor for both Wnt/β-catenin and BMP signaling. This property is somewhat similar to the Xenopus protein Cerberus, which antagonizes signaling by Wnt, BMP and Nodal (Piccolo et al.,1999).

WNT signaling is not only essential for embryogenesis, but also for postnatal development and tissue homeostasis. This is illustrated by LRP5 mutations that underlie familial osteoporosis, high bone density syndromes and ocular disorders (Gong et al.,2001; Boyden et al.,2002; Little et al.,2002). Children with autosomal-recessive osteoporosis-pseudoglioma syndrome (OPPG) have low bone mass and are prone to bone fractures(Gong et al., 2001). Most of these children suffer a loss of LRP5 function due to nonsense or frame-shift mutations in the LRP5 extracellular domain(Gong et al., 2001)(Fig. 6). Remarkably, several groups of autosomal-dominant bone disorders, characterized by high bone density traits, are also associated with LRP5 mutations, which are missense in nature and clustered in the first β-propeller region of LRP5(Boyden et al., 2002; Little et al., 2002; Van Wesenbeeck et al., 2003)(Fig. 6). This is reflected in mice. Lrp5^–/– mice exhibit low bone density and frequent bone fractures reminiscent of OPPG patients(Kato et al., 2002), and transgenic mice expressing LRP5 (G171V), a mutation from high bone density patients, had increased bone mass (Babij et al., 2003). Thus, loss-of-function mutations of LRP5 lead to low bone densities whereas `gain-of-function' mutations cause high bone mass. These studies identify LRP5 as a central player and an ideal therapeutic target in bone mass regulation and in associated diseases such as osteoporosis(Patel and Karsenty, 2002). Nevertheless, several key issues remain unresolved:

1. Does WNT signaling regulate bone mass during development? LRP5 is required for the proliferation of osteoblasts (Gong et al., 2001; Kato et al.,2002), and may inhibit osteoblast apoptosis(Babij et al., 2003). Wnt/β-catenin signaling can promote mouse osteoblastic development in vitro (Gong et al., 2001), and several Wnt genes, including Wnt1, are expressed during bone accrual(Kato et al., 2002). Therefore, LRP5 probably mediates WNT signaling in bone mass regulation, but whether LRP5 mediates other types of extracellular signals during bone growth has not been determined.

2. Does LRP6 regulate bone mass? As discussed earlier, LRP6 is the more crucial WNT co-receptor and shares significant overlapping function with LRP5 during embryogenesis (Kelly et al.,2004). It is therefore somewhat surprising that even heterozygosity of LRP5 reduces bone density in mice and men(Gong et al., 2001; Kato et al., 2002), and that LRP6 does not compensate for this LRP5 function. However, it is possible that LRP6 is not expressed in osteoblasts during postnatal stages, or that the ligand for bone accrual binds preferentially to LRP5.

3. How do LRP5 mutations associated with high bone density traits, such as G171V, cause LRP5 hyperactivation? A study in mammalian cells suggested that the G171V mutation does not affect WNT-LRP5 signaling, but instead prevents DKK1 inhibition of WNT-LRP5 signaling(Boyden et al., 2002),implying that loss of DKK inhibition of LRP5 may underlie increased LRP5 signaling and high bone densities. A few caveats remain. First, whether DKK1 or any other DKK is expressed during bone accrual has not yet been examined. Second, whether LRP5 high bone density mutations affect WNT or DKK binding has not been examined. These mutations are all clustered in the first YWTDβ-propeller domain, which appears to be involved in binding to WNT but not DKK (Fig. 6). Third, it remains unknown whether G171V and other high bone density LRP5 mutations,which are in different blades of the propeller, affect LRP5 activity by the same mechanism.

It is of interest to discuss here sclerosteosis, another rare autosomal-recessive bone disorder characterized by skeletal overgrowth and high bone density. Sclerosterosis is a progressive bone dysplasia associated with loss-of-function mutations of a secreted protein referred to as SOST(Balemans et al., 2001; Brunkow et al., 2001). Intriguingly, SOST is most related to WISE (38% identical; see Fig. S2),which antagonizes LRP6 via binding to the region containing the first and second YWTD β-propeller domains (Fig. 6) (Itasaki et al.,2003). Although SOST, like Wise, can also bind and antagonize BMP(Laurikkala et al., 2003; Kusu et al., 2003), it is tempting to speculate that SOST binds and antagonizes LRP5 in bone growth regulation, and that LRP5 mutations associated with high bone densities prevent/reduce SOST-LRP5 interaction.

OPPG patients also suffer, in addition to low bone mass, from severe disruption of ocular structures due to a failure of regression of the primary vitreal vasculature (the temporary capillary networks that normally regress during development) (Gong et al.,2001). This phenotype is recapitulated in Lrp5^–/– mice, possibly because of a lack of capillary endothelial apoptosis in the eye(Kato et al., 2002). This may be due to defects in ocular macrophages, which express Lrp5 and are required for the induction of capillary cell death(Kato et al., 2002).

Another hereditary ocular disorder, the autosomal-dominant form of familial exudative vitreoretinopathy (FEVR), which is characterized by the premature arrest of retinal angiogenesis/vasculogenesis, is also associated with LRP5 mutations (Toomes et al.,2004). Loss-of-function mutations in one LRP5 chromosomal copy are associated with this disease (Fig. 6), presumably due to haploinsufficiency(Toomes et al., 2004). These patients, like the obligate OPPG carriers (parents of OPPG patients)(Gong et al., 2001), also exhibit low bone mass (Toomes et al.,2004). Interestingly, some autosomal-dominant FEVR families harbor loss-of-function mutations in the frizzled 4 (FZD4) gene(Robitaille et al., 2002),which, like LRP5, is located in the chromosomal 11q13 region. Therefore, FEVR is associated with a deficiency in either LRP5 or FZD4 function, providing genetic evidence outside Drosophila that LRP5 and FZD cooperate in the same signaling pathway. In summary, LRP5 (and FZD)function is important for multiple stages of retinal angiogenesis and associated diseases, presumably because it mediates signaling by a WNT (or non-WNT) ligand.

LRP5 is also involved in lipid metabolism. LRP5 binds apolipoprotein E(APOE), and LRP5 expression is upregulated in the liver of Ldlr^–/– mice(Kim et al., 1998). Indeed, Lrp5^–/– mice fed on a high-fat diet exhibit increased plasma cholesterol levels relative to normal mice(Fujino et al., 2003), and mutants for both ApoE and Lrp5(ApoE^–/–;Lrp5^–/–) show hypercholesterolemia, impaired fat tolerance and advanced atherosclerosis(Magoori et al., 2003). The role of LRP5 in cholesterol metabolism appears analogous to the classical LDLR function and is probably WNT independent(Magoori et al., 2003).

Although the role of Arrow/Lrp5/Lrp6 in Wnt/β-catenin signaling during development is established, the underlying molecular mechanism remains relatively poorly defined. Unresolved questions concern how Wnt molecules interact, functionally and/or physically, with Fz and Arrow/Lrp5/Lrp6, how Wnt activates Arrow/Lrp5/Lrp6 by phopshorylation to regulate the Axin complex, and how Fz receptors and Dsh/Dvl proteins operate. Such studies are likely to identify additional signaling components at the plasma membrane or intracellularly that will bridge major gaps in our understanding of transmembrane signaling by Wnt proteins.

Although Dkk and Wise families of Lrp5/Lrp6 ligands/antagonists are only found in vertebrates, the high similarity between the entire Arrow and Lrp5/Lrp6 extracellular domains implies that other evolutionarily conserved ligands may yet be discovered. An area that requires more investigation is the cell biological aspect of Wnt signaling, including the biogenesis,trafficking, localization and endocytosis of Fz and Arrow/Lrp5/Lrp6 during Wnt signaling. On the translational side, the specific function of LRP5 in bone mass regulation provides an ideal therapeutic target for treatment of bone disorders, but this will critically rely on our understanding of LRP5, its putative ligand (a WNT or non-WNT), and antagonists such as DKK, WISE or others during bone accrual. In addition, the roles of LRP5 in lipid metabolism and other physiological regulations, such as in enhancing insulin secretion(Fujino et al., 2003), may have significant medical implications. In this regard, molecules that can stimulate LRP5 expression or activity may be used as therapeutics for osteoporosis, high cholesterol associated diseases and diabetes.

We thank R. Korswagen, N. Tolwinski and S. Blacklow for discussion, B. Skarnes for communication before publication, and referees for constructive suggestions.