Animal development requires complex mechanisms for controlling and regulating cell polarity. Fundamental polarity mechanisms define the apical and basal domains of individual cells, whereas planar cell polarity defines directionality within a sheet of cells. In many contexts, disruption of polarity can lead to aberrant growth, and is a classic hallmark of cancer metastasis. Many polarity proteins are evolutionarily conserved, even though the morphology and cellular characteristics of polarized tissues are amazingly diverse between species, and even within the same organism across different developmental stages. Many components of these pathways have been identified, but the regulatory mechanisms that determine patterns of polarity are not fully understood.
The basic mechanism of apical–basal polarity involves localization of apical and basal components to opposite sides of a cell with adherens junctions positioned in between the two domains (Macara,2004; Margolis and Borg,2005). The Par complex, composed of atypical Protein Kinase C (aPKC), Bazooka/Par3, and Par6, constitutes an evolutionarily conserved apical complex that plays a crucial role in the establishment of apical–basal polarity (Kemphues et al.,1988; Muller and Wieschaus,1996; Hutterer et al.,2004). In Drosophila embryos, this apical complex is also required for planar cell polarity in the early convergent extension movements associated with gastrulation (Zallen and Wieschaus,2004; Blankenship et al.,2006). Furthermore, the localization of this complex affects the positioning of adherens junctions in developing epithelia (Harris and Peifer,2004,2005,2007). The core components of adherens junctions are E-Cadherin, Armadillo (β-catenin, Arm), and α-catenin. Adherens junctions interact dynamically with the Actin cytoskeleton and provide adhesion between adjacent cells (Yamada et al.,2005; Weis and Nelson,2006).
Recent studies have found interactions between components of Wnt signaling pathways and a key component of the Par complex (aPKC; Etienne-Manneville and Hall,2003; Schlessinger et al.,2007; Zhang et al.,2007). In particular, noncanonical Wnts, such as Wnt11 in Xenopus and Wnt5a in migrating fibroblasts and developing hippocampal neurons have been found to play a role in organizing cell polarity. The downstream Wnt signaling pathway components play roles as well (Barrow,2006). The adaptor protein Disheveled (Dsh) adopts a polarized distribution and interacts with and activates aPKC (Zhang et al.,2007). The kinase GSK3β, also interacts with aPKC and participates in polarity determination (Etienne-Manneville and Hall,2003). Interestingly, in Xenopus embryos, Wnt/Dsh signaling appears to regulate the localized assembly of adherens junctions (Yamanaka and Nishida,2007). Furthermore, the aPKC complex is also required for adherens junction formation in mouse neuroepithelia (Imai et al.,2006) and in chicken neural progenitors (Engelsberg et al.,2008). Overall, the pathway or pathways regulating aPKC function are complex. In Drosophila, the small GTPase Cdc42 functions through aPKC to stabilize adherens junctions, and appears to do so by regulating intracellular vesicular trafficking of junction components (Georgiou et al.,2008; Harris and Tepass,2008; Leibfried et al.,2008).
Here, we investigated the mechanism of aPKC regulation and its downstream effects during embryogenesis. We report that aPKC levels are regulated downstream of GSK3β in the Drosophila epidermis. We find that aPKC is directly phosphorylated by GSK3β, and that this leads to stabilization of adherens junctions by increasing the levels of Arm. These results provide direct in vivo evidence that interactions between the Wnt signaling pathway and the apical–basal machinery play an important role during cell morphogenesis.
Posttranslational Regulation of aPKC
The epithelial cells that make up the Drosophila embryonic epidermis undergo dramatic morphological changes in late embryogenesis. All embryonic epidermal cells secrete cuticle, the outer shell that protects larvae (Hillman and Lesnick,1970; Martinez Arias,1993). In embryonic segments, half the cells secrete only cuticle and these take on an irregular, squamous shape, while the other half of cells secrete both cuticle and denticles and take on a regular, rectangular shape (Fig. 1A). Denticles are actin-based structures that form on the ventral side of the embryo and align in a stereotypic pattern (Fig. 1B). The elaborate patterning and denticle organization of this tissue (Hatini and DiNardo,2001) makes this an excellent model system for studying the signaling pathways that regulate polarity and morphogenesis (Colosimo and Tolwinski,2006; Price et al.,2006; Walters et al.,2006). For these reasons, we used this tissue to investigate potential interactions between Wnt pathway components and the apical–basal polarity proteins.
We began by looking at the protein expression pattern of aPKC during late embryogenesis. While the expression of aPKC is uniform throughout early stages of embryonic development (Wodarz et al.,2000), we found that aPKC protein is up-regulated in anterior–posterior stripes in later stages (stage 16, Fig. 1D). Armadillo (Arm) expression marks cell outlines (Fig. 1C), and costaining reveals that aPKC is specifically up-regulated in the rectangular, denticle-producing cells (Fig. 1E). This pattern appears to be specific to the cells of the ventral epidermis during denticle formation stages. On the dorsal side of the embryo, all cells produce actin-based protrusions, the majority of which are morphologically distinct from ventral denticles. p-Tyr staining (Fig. 1F) reveals the precursors to these actin structures, and aPKC staining reveals a mild up-regulation in a single row of cells (Fig. 1G–H, arrowhead), but the dramatic stripes seen on the ventral side were not observed.
We hypothesized that posttranslational regulation could control aPKC levels. One candidate for posttranslational regulation is the kinase GSK3β, because of its known links to aPKC (Etienne-Manneville and Hall,2003), and its function in targeting proteins for degradation in various signaling pathways (Doble and Woodgett,2003). We therefore investigated aPKC protein expression in GSK3β maternal/zygotic (M/Z, all mutants referred to in this study are maternal and zygotic mutants, see the Experimental Procedures section) mutant embryos. We found that aPKC was not present in stripes, but instead was present at equal levels in all cells of the ventral epidermis (Fig. 1J). Arm expression was even throughout this tissue (Fig. 1I). Costaining of Arm and aPKC showed even distribution of aPKC at the cell membrane (Fig. 1K).
A recent study has suggested that Dsh can activate aPKC, although the mechanism remains unclear (Zhang et al.,2007). In contrast, Dsh and GSK3β have well-established roles in the canonical Wnt pathway where Dsh inhibits GSK3β activity. As shown above, GSK3β appears to be regulating aPKC; therefore, we wondered whether the lack of GSK3β inhibition in dsh mutants would also lead to uniform aPKC. Specifically, we would expect that up-regulation of aPKC expression in anterior–posterior stripes would not occur. To test this hypothesis, we investigated aPKC expression in dsh mutants and found that aPKC no longer localized in anterior–posterior stripes (Fig. 1M). Arm expression was uniform (Fig. 1L), and costaining showed that some aPKC was present at the cell membrane (Fig. 1N). Altogether, these data indicate that disrupting the function of either Dsh or GSK3β leads to uniform aPKC protein levels across the ventral epidermis.
Loss of GSK3β Leads to Increased aPKC Activity
Previous work in migrating astrocytes and developing hippocampal neurons showed that aPKC and GSK3β interact to regulate cell polarity (Etienne-Manneville and Hall,2001,2003). In our experiments, loss of GSK3β leads to a loss of aPKC stripes in the late embryo. We therefore examined the genetic relationship between GSK3β and aPKC. We hypothesized that, because inhibition of GSK3β may lead to an increase in aPKC levels, aPKC activity should be increased as well.
To determine whether loss of GSK3β enhances aPKC activity, we compared the effects of ectopically expressed aPKC transgenes in wild-type embryos vs. embryos harboring an amorphic allele of GSK3β. An activated form of aPKC contains an N-terminal deletion (aPKCΔN) and can no longer bind to Par-6, which normally restricts aPKC to the apical domain of cells (Betschinger et al.,2003). Therefore, aPKCΔN is free to act throughout cells. Wild-type embryos display a stereotypic pattern of denticles (Fig. 2D). Embryos expressing aPKCΔN do not display substantial defects in cuticle formation and patterning (Fig. 2G). Although GSK3β mutants develop entirely naked cuticles due to hyperactivation of canonical Wnt signaling (Siegfried et al.,1990), intact cuticles do form in these mutants (Fig. 2A,J). When aPKCΔN was expressed in GSK3β mutants, we saw a severe disruption of apical–basal polarity in (Fig. 2B,M). This severe cuticular disruption caused by apical–basal polarity defects is quite similar to the phenotypes seen in lgl, dlg, or scrib (M/Z) mutants (Fig. 2C; Tanentzapf and Tepass,2003). These results suggest that loss of GSK3β enhances the phenotype of aPKCΔN, leading to disruption of apical–basal polarity in the epidermis.
To further analyze the disruption of apical–basal polarity in GSK3β mutants expressing aPKCΔN, we performed immunofluorescence of apical–basal markers. In wild-type embryos, E-Cadherin (E-Cad) is restricted to the adherens junctions and expressed in a punctate pattern (Fig. 2E), while the Crumbs protein is localized to the apical domain of cells (Fig. 2F). However, expression of aPKCΔN led to expansion of E-Cad and Crb along the lateral membrane (Fig. 2H,I). In GSK3β mutants, E-Cad and Crb are also expanded along the lateral membrane (Fig. 2K,L), similar to embryos with increased levels of aPKC activity (Fig. 2H,I). In addition, there are ectopic junctional puncta in GSK3β mutants (Fig. 2K, arrowheads). When aPKCΔN is expressed in the absence of GSK3β, E-Cad is no longer restricted to apical puncta (Fig. 2N) and also appears in ectopic puncta (Fig. 2N, arrowheads). Crb is expanded along the lateral membrane (Fig. 2O). In addition to ectopic E-Cad expression, the epidermis of these embryos is multilayered, a classic result of loss of apical–basal polarity (Fig. 2N, arrows). These data indicate that expression of activated aPKC in GSK3β mutants leads to severe disruption of apical–basal polarity.
Thus, our data are consistent with the hypothesis that loss of GSK3β appears to increase aPKC activity, suggesting that GSK3β is upstream of aPKC. To further investigate this relationship, we analyzed aPKC protein levels in GSK3β mutants. We found that GSK3β mutants displayed significantly increased levels of aPKC protein compared with wild-type embryos (Fig. 3A, top panel). Additionally, we compared the levels of aPKC in dsh vs. GSK3β mutants and observed a large increase in aPKC protein in GSK3β mutants, but not in dsh mutants (Fig. 3B). This suggested the possibility that aPKC protein levels were regulated similarly to Arm protein levels (Fig. 3A,B, second panels from top) by GSK3β-mediated degradation.
To test if aPKC is a direct phosphorylation target of GSK3β, we performed in vitro kinase assays and found that GSK3β can directly phosphorylate purified human aPKC protein (Fig. 4A) and Drosophila His-Myc-aPKCΔN purified from embryonic lysates (Fig. 4B). In addition, our kinase assays show aPKC can phosphorylate GSK3β (Fig. 4A,B) as previously described (Etienne-Manneville and Hall,2003).
Because a well-known role of GSK3β phosphorylation is to target proteins for proteosome-mediated degradation, we tested if aPKC is indeed conjugated to ubiquitin by performing ubiquitination assays (Wang et al.,2007). Ubiquitin-conjugated proteins exist briefly, and can only be observed when the proteasome is inhibited. We cotransfected HA-tagged ubiquitin with V5-tagged aPKC, and found that blocking protein degradation with the proteosome inhibitor MG132 led to HA-ubiquitin conjugation to aPKC in a characteristic poly-ubiquitin smear (Fig. 4C). These results indicate that aPKC is conjugated to ubiquitin as has been observed for other PKC isoforms (Chen et al.,2007). Taken together, these results show that GSK3β is an important regulator of aPKC activity along the apical–basal axis, and may regulate aPKC levels by targeting aPKC for proteosomal degradation.
In Drosophila embryos, reducing Cdc42 function disrupts adherens junctions specifically in the ventral epidermis (Goldstein and Macara,2007). More recent studies have found that Cdc42 and aPKC regulate trafficking of the apical protein Crumbs and as well as E-Cad (Georgiou et al.,2008; Harris and Tepass,2008; Leibfried et al.,2008), and this is critical for adherens junction stability. aPKC appears to be downstream of Cdc42 in endocytosis, because expressing an activated form of aPKC (aPKCΔN) can partially rescue the defects in adherens junction integrity when Cdc42 function is reduced. However, the full breadth of molecular targets of Cdc42 and aPKC in regulating junctional stability remain to be identified. Because phosphorylation of Arm is a major form of adherens junction regulation (for review, see Nelson and Nusse,2004; Daugherty and Gottardi,2007), and we have previously identified phosphorylation sites on Arm that are important for adhesion (Colosimo and Tolwinski,2006), we hypothesized that Arm is a downstream target of aPKC.
We investigated if activated aPKC could influence the epithelial integrity defect in arm loss-of-function mutants. Embryos that are maternally and zygotically mutant for arm043A01, a truncation allele of arm, typically fall apart by the late stages of embryogenesis due to a lack of cell–cell adhesion (Fig. 5A; Muller and Wieschaus,1996; Harris and Peifer,2004; Tolwinski and Wieschaus,2004a). Surprisingly, we found that increasing the amounts of aPKC mitigated the epithelial integrity defect in arm043A01 mutants, leading to embryos with small patches of intact cuticle that developed denticles (Fig. 5B). As the activity of ectopic aPKC increased, the percentage of embryos that displayed patches of cuticle and denticles increased (Table 1).
Table 1. Summary of Cuticle Phenotypes Caused by Altering aPKC Activity in Various Genotypes
Cuticle phenotype (maternal/zygotic mutants)
Cuticle phenotype (maternal mutants)
zw3M11-1; armGAL4; UAS-aPKCCAAX WT
No denticles 52/95 (50% expected)
Mild denticle loss 43/95 (50% expected)
zw3M11-1; armGAL4; UAS-aPKCCAAX K293W
No denticles 69/114 (50% expected)
Mild denticle loss 55/114 (50% expected)
zw3M11-1; armGAL4; UAS-aPKCΔN
Apical expansion 21/102 (25% Expected)
Mild denticle loss 58/102 (50% expected)
armO43A01; armGAL4; UAS-aPKCCAAX K293W
Loss of cell adhesion 80/184 (8/184 weak rescue expected 25%)
Arm protein expression revealed that, by late stages of development, only very small clumps of epidermis were left in arm043A01 mutants (Fig. 5C). In some cases, Actin still coalesced into apical foci, indicating that early denticle precursors may sometimes form in these embryos (Fig. 5D). However, intact denticles are never seen in cuticle preparations of these mutants (Tolwinski and Wieschaus,2001,2004a). Costaining reveals that the apical Actin foci are randomly placed within the cells (Fig. 5E, inset). We then looked at Arm expression in arm043A01 mutants expressing activated aPKC. We found that the aPKCΔN transgene led to large patches of epidermal cells with intact adherens junctions (Fig. 5F). In addition, denticle precursors were well formed in these cells, as revealed by Actin expression (Fig. 5G). Costaining of Arm and Actin revealed that the denticle precursors were randomly oriented within the cells (Fig. 5H, inset). Importantly, this rescue effect on adherens junctions and denticle development in arm043A01 mutants was only observed by increasing aPKC activity, and not with gain-of-function or dominant-negative Cdc42 alleles (Table 1).
Because increasing the levels of aPKC correlated with an increase in embryos with intact patches of cuticle, we then considered whether aPKC might stabilize adherens junctions by stabilizing Arm through phosphorylation. As shown in the Western blot in Figure 5I, embryos with increased levels of aPKC do show increased levels of Arm protein, compared with wild-type controls, although whole embryo lysates do not discriminate between junctional and signaling pools of Arm. Furthermore, aPKC and Arm can be coprecipitated from embryonic lysates, although attempts to find a direct protein–protein interaction using recombinant aPKC and Arm were unsuccessful (data not shown). Finally, and unexpectedly, we found that expression of a kinase-dead version of aPKC in arm043A01 mutants induced a small percentage of embryos displaying small patches of cuticle (∼5%, Table 1). These results raise the possibility that the mitigation of epithelial integrity in arm043A01 mutants by aPKC is not mediated through its kinase activity. The kinase-dead version of aPKC used above also contained a membrane-targeting CAAX sequence, and so may be acting as a scaffold to recruit another protein to the membrane. Nevertheless, these results indicate that increasing the levels of aPKC stabilizes epithelial cell–cell adhesion by increasing the levels of Arm.
GSK3β Regulates aPKC Levels
aPKC is a major regulator of cell polarity in a wide array of developmental contexts (for review, see Goldstein and Macara,2007). Recently, aPKC has been shown to be regulated by Wnt signaling during polarized migration (Etienne-Manneville and Hall,2003). Our data uncover a novel signaling mechanism that links Dsh and GSK3β, known components of the Wnt pathway, as regulators of aPKC. Although aPKC mRNA was shown to be uniformly expressed in early Drosophila embryos (Wodarz et al.,2000), in late stages of development, we observed aPKC protein in anterior–posterior stripes, suggesting posttranslational regulation. In embryos lacking GSK3β or its upstream regulator dsh, aPKC is no longer in stripes and appears to be uniform in all cells. Western blotting experiments support these immunofluorescence data and indicate high levels of aPKC in GSK3β mutants, and conversely lower levels in dsh mutants. Therefore, there appears to be a regulatory pathway that determines the distribution of aPKC across this epithelial tissue, where GSK3β directly phosphorylates aPKC and likely regulates aPKC protein levels through ubiquitin-mediated degradation (Doble and Woodgett,2003).
aPKC Affects Cellular Adhesion
The second part of our studies investigated the effects of inappropriately activating aPKC activity on adhesion. Previously, it had been shown that polarized aPKC plays an important role in asymmetrically regulating microtubule polarization during the development of polarized cells (Etienne-Manneville and Hall,2003). Furthermore, aPKC has been shown to function downstream of Cdc42 in the trafficking of adherens junction components to the plasma membrane (Georgiou et al.,2008; Harris and Tepass,2008; Leibfried et al.,2008). Our data argue that aPKC acts downstream of GSK3β to affect the stability of cell–cell adherens junctions. Importantly, we found that aPKC activation could mitigate weak cellular junctions brought about by mutations in arm (arm043A01), and result in the rescue of epithelial integrity. However, in our experiments we did not observe a similar rescue effect with activated Cdc42, a known positive regulator of aPKC. These results suggest that there is a separate, parallel pathway through GSK3β that does not require Cdc42, or that activated Cdc42 has other effects that mask the epithelial integrity rescue caused by increased aPKC activity.
Our biochemical data suggest that aPKC leads to an increase of Arm protein, and the immnohistochemistry and genetic results suggest that this specifically affects the membrane pool of junctional Arm. Our data do not indicate an effect on the signaling pool of Arm protein, because patterning of the ventral epidermis does not appear to be affected by aPKC overexpression (Fig. 2B; Gottardi and Gumbiner,2004; Tolwinski and Wieschaus,2004a,b), although we have not determined if aPKC affects Arm directly or indirectly. This is consistent with the finding that trafficking of adherens junction components is increased by activated aPKC (Harris and Tepass,2008).
This study shows a link between the regulation of adhesion and cell polarity, processes important in determining the invasiveness and metastatic potential of tumors (Pagliarini and Xu,2003; Hariharan and Bilder,2006). Of interest, the only known tumor suppressor genes in Drosophila are involved in maintaining polarity. The phenotypes we observed by inappropriately activating aPKC provide the first examples of phenotypes similar to mutations in the tumor suppressor genes discs large and lethal giant larvae (Tanentzapf and Tepass,2003). Loss of these genes leads to overgrowth phenotypes, and when combined with misregulation of an oncogene, also leads to metastatic behavior (Pagliarini and Xu,2003). Our results suggest that Wnt pathway mutations have oncogenic potential not just because cell proliferation mechanisms are activated, but also because Wnt pathway mutations lead to misregulation of cell polarity that bypasses the contact inhibition mechanisms that maintain polarity (Pagliarini and Xu,2003; Nelson and Nusse,2004; Reya and Clevers,2005; Hariharan and Bilder,2006). Future work on this pathway should provide insight into how cells behave when their polarity is disrupted, and how polarity may be coupled to proliferation.
Crosses and Expression of UAS Constructs
Maternally mutant eggs were generated by the dominant female sterile technique (Chou and Perrimon,1992). Oregon R was used as the wild-type strain. Please see Flybase for details on mutants used (flybase.bio.indiana.edu). All mutants used were complete loss-of-function except for the strong hypomorph arm043A01 that contains a stop codon at the end of repeat nine, deleting the entire C-terminal region and the final three repeats, which severely limits its function in both nuclear signaling and cell adhesion (Tolwinski and Wieschaus,2004a). For aPKC expression experiments, the armGAL4 driver was used. All X-chromosome mutants use FRT 101 except for dshV26, which has FRT 18E. zw3 is the Drosophila homolog of GSK3β. The following crosses were conducted: zw3M11-1 FRT101/ovoD1 FRT101; armGAL4/+ females × UAS-aPKCΔN males; arm043A01 FRT101/ovoD1 FRT101; armGAL4/+ females × UAS-aPKCΔN males; arm043A01 FRT101/ovoD1 FRT101; armGAL4/+ females × UAS-Cdc42V12 males; and arm043A01 FRT101/ovoD1 FRT101; armGAL4/+ females × UAS-Cdc42N17 males.
UAS Transgenes and GAL4 Driver Lines
Two ubiquitous drivers were used for expression of transgenes: armGAL4 and daughterless-GAL4 (Brand and Perrimon,1993). UAS constructs were as follows: UAS-aPKCΔN and UAS-aPKCWT (Betschinger et al.,2003), aPKCCAAX and aPKCCAAX kinase inactive (K293W; Sotillos et al.,2004), UAS-Cdc42V12, and UAS-Cdc42N17 (Luo et al.,1994). For this study, we also cloned full-length aPKCWT and aPKCΔN into a 6xHis, 3xMyc tagged vector (pUASt-HM, Hsu and McCabe, unpublished) using Gateway technology (Invitrogen).
Antibodies and Immunofluorescence
Embryos were fixed with Heat-Methanol treatment (Muller and Wieschaus,1996) or with heptane/4% formaldehyde in phosphate buffer (0.1 M Na3PO4 pH 7.4; Tolwinski et al.,2003). The antibodies used were as follows: anti-Actin (mAb JLA-20, DSHB [Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by The University of Iowa, Department of Biological Sciences, Iowa City, IA]); anti–E-Cadherin (ratAb ECAD2, DSHB); anti-Crb (mAb Cq4, DSHB); anti-Armadillo (mAb N2 7A1, DSHB); rabbit anti-Armadillo (Riggleman et al.,1990); rabbit anti-cMyc, rabbit and goat anti-aPKCζ, mouse phosphotyrosine pY99 (Santa Cruz Biotechnology); anti–β-tubulin (E7, DSHB); rat anti-cMyc (Genetix); rabbit anti-GSK3β (Willert et al.,1999a); and anti-Sexlethal (mAb M-14, DSHB). Staining, detection, and image processing was as described in Colosimo and Tolwinski (2006).
Embryos were lysed in extract buffer (50 mM Tris pH 7.5, 150 mM NaCl, 1 % NP-40, 1 mM ethylenediaminetetraacetic acid, 10% Glycerol, Complete Mini Protease, Sigma) or RIPA lysis buffer (Santa Cruz Biotechnology, Inc.). The extracts were separated by 7.5% sodium dodecyl sulfate-polyacrylamide gel electrophoresis, and blotted as described in Peifer et al. (1994). Extracts were normalized using the BCA assay (Novagen). Overnight embryo collections were used to make extracts for Western blots, as posttranslational regulation of aPKC appears to be in late stage embryos (stage 16).
These experiments were performed as described in (Willert et al.,1999b). Extracts were made from embryos expressing a His-Myc-tagged aPKC. His-Myc-aPKC was immunoprecipitated with a rabbit c-myc polyclonal antibody or pulled down using NiNTA beads (Qiagen). Kinase assays were performed on both recombinant proteins and proteins purified from embryonic extracts. Recombinant proteins were either purchased from Cell Signaling Technologies (GSK3β and PKCζ) or prepared from bacterial lysates (Shapiro et al.,1998). aPKC phosphorylation was assayed as described in Djiane et al. (2005).
293T cells were transfected with HA-Ubiquitin and VR-aPKC (pDEST42, Invitrogen). Proteosome activity was inhibited with MG132 as in Wang et al. (2007).
We thank Alan Hall and Xiaojian Xu for help with protein expression, Xuejun Jiang and Xinjiang Wang for help with the ubiquitination assays. We thank Kathryn Anderson, Mary Baylies, Alan Hall, and Jennifer Zallen for comments on the manuscript, Todd Blankenship and the Zallen lab for discussions, and Ling-Yu Chiu for technical assistance. We thank Chris Doe, Brian McCabe, Andreas Wodarz, Jurgen Knoblich, and Sonsoles Campuzano for sharing reagents. This work was supported by the Frank A. Howard Fellows Program at the Sloan-Kettering Institute. P.F.C. was supported by an NIH-NRSA postdoctoral fellowship.