Datasets:

chemNLP

/

ncbi-text

like 0

The ensheathment of axons by myelin is essential for the efficient propagation of nerve impulses both in the central nervous system and peripheral nervous system (PNS). In the PNS, myelin is made by Schwann cells, each cell wrapping around a single axon, creating concentric layers. The compaction of these layers results in the formation of the dense and interperiod lines typical of mature myelin (). The wrappings created by each Schwann cell are distinct and separated by the Nodes of Ranvier, structures specialized for Na and K ion exchange (; ). The major protein component of myelin in the PNS is myelin protein zero (P0), a single-pass transmembrane molecule containing one Ig-like loop in the extracellular domain and a 69–amino acid highly basic cytoplasmic domain (). P0 is found throughout peripheral myelin and is essential for normal myelination. Mutations in the gene cause peripheral neuropathy with either prominent demyelination, slowed nerve conductions and onset in childhood, or mainly axonal dysfunction, essentially normal nerve conductions, and onset as an adult (; ; , ). There are currently >95 different mutations in P0 correlating with human neuropathies (). Mice null for the gene have uncompacted myelin in the PNS and develop a severe, early onset demyelinating neuropathy, whereas heterozygotes have a later onset neuropathy with substantial amounts of inflammation (; ; ; ). Mutations in P0 are dominant, suggesting that the mutant protein interferes with the function of wild-type protein (; ). Tissue culture and knock-in experiments have borne this out; cotransfection with wild-type and mutant P0 results in the loss of adhesion function (), and transgenic mice expressing a mutant P0 in a wild-type background develop demyelinating peripheral neuropathy (). The role of P0 in maintaining compact myelin may, in part, be caused by its ability to mediate homotypic interactions (; ; ; ), and decreases in adhesion have been correlated with the severity of disease (). Crystallographic studies of the extracellular domain of P0 suggest that it interacts in cis to form homotetramers, which, in turn, interact with similar tetramers in an apposing fold of the myelin membrane, thus contributing to the formation of compact myelin (). Supporting this model, mutations in several of the residues likely to participate in cis- and trans-interactions of the homotetramers can cause inherited neuropathy and reduced cell–cell adhesion in vitro (). The cytoplasmic domain of P0 is also important for myelin compaction and adhesion. Several different nonsense or point mutations in the intracellular domain have been found in patients presenting with different neuropathies (), and truncation of the cytoplasmic domain eliminates adhesion function (; ). Our own results implicate PKCα-mediated phosphorylation of the cytoplasmic domain of P0 in the regulation of P0-mediated adhesion and potentially formation/maintenance of compact myelin. First, we showed that deletion of a 14–amino acid sequence that eliminates a putative PKCα target site (198RSTK201) as well as point mutations within this domain eliminate P0-mediated adhesion. A patient presenting with late onset Charcot Marie Tooth disease (CMT) 1B was found to have a mutation in the PKCα target site R198S. In vitro analysis of the P0 function bearing the R198S mutation revealed a deficit in P0-mediated adhesion. We further demonstrated that PKCα and the receptor for activated C kinase 1 (RACK1) are associated with the cytoplasmic domain of P0 and that the inhibition of PKCα activity also inhibits P0-mediated adhesion. Point mutations that eliminate potential phosphorylation target sites (S199 or 204A) also result in the loss of adhesion, and deletions eliminating these serine residues also result in CMT (). We have now identified a protein, p65, that interacts directly with P0 and RACK1, bringing PKCα in close proximity to its target sites in the cytoplasmic domain of P0. Deletion of the P0 domain responsible for p65 binding results in the loss of P0-associated RACK1 and PKCα and the loss of P0-mediated cell adhesion. Importantly, two CMT patients carrying a point mutation in this domain, G184R, have been identified. Recombinant P0 with this mutation does not interact with p65, and cells transfected with this mutant P0 are unable to form adhesions. These data strongly suggest that the interaction of p65 with the cytoplasmic domain of P0 provides the foundation for the attachment of RACK1 and PKCα, resulting in the phosphorylation of P0 at serines 199 and/or 204. This is substantiated by functional rescue of the G184R mutant P0 by mutation of serines 199 and 204 to glutamic acid. Collectively, these data suggest that regulation of P0 phosphorylation and adhesion, which is mediated, in part, by the binding of p65, RACK1, and PKCα to a specific sequence in the cytoplasmic domain of P0, plays an important role in myelination. To identify potential proteins that interact with the cytoplasmic tail of P0, we screened a cDNA library from P30 rat sciatic nerve, a time of peak myelin expression, using the P0 cytoplasmic domain as bait in a yeast two-hybrid assay (). 13 cDNA sequences encoding peptides that interact with the P0 bait were sequenced; 11 contained an identical 1.1-kb cDNA fragment. The complete cDNA codes for a 65-kD protein (p65) previously identified in the nucleus of cells undergoing meiosis (GenBank/EMBL/DDBJ accession no. ; ) and as a 55-kD nucleolar protein (GenBank/EMBL/DDBJ accession no. ; ). The selectivity of the interaction of p65 and P0 in the yeast two-hybrid system used was determined by testing other bait proteins as well as a C-terminal 28–amino acid deletion of P0 that fails to mediate cell adhesion (). Only the full-length P0 or P0 lacking the 13 C-terminal amino acids (, Δ1) that do not affect P0-mediated adhesion were able to interact (not depicted). The 1.1-kb p65 cDNA reacts with two distinct mRNA bands from rat sciatic nerve of ∼2.3 and 2 kb (). Transection of the sciatic nerve results in a progressive loss of p65 message: 4 d after transection, the transcripts are greatly reduced and are barely detectable by day 12 (). This loss parallels the loss of P0 message (, bottom), suggesting that the levels of expression of P0 and p65 are coordinately regulated. An antibody developed to a peptide corresponding to a sequence in p65 recognizes an ∼65-kD protein band in sciatic nerve, cultured mouse Schwann cells, and L cells expressing full-length P0 (). In addition, full-length P0 but not P0 lacking the 28 C-terminal amino acids (, Δ2) is associated with p65, as determined by coimmunoprecipitation (), further validating the results of the two-hybrid screen. The results of the yeast two-hybrid assay suggest that the interaction between P0 and p65 is direct. To definitively demonstrate this, we engineered a fusion peptide containing the cytoplasmic domain of P0 linked to the CaM-binding peptide (CBP) and the fragment of p65 identified in the yeast two-hybrid assay as a GST fusion. GST-p65 binds directly to CBP-P0 immobilized on CaM-coated wells in a dose-dependent manner, whereas GST alone does not (). The 1.1-kb message identified by the two-hybrid screen corresponds to a fragment of p65 lacking the N terminus (amino acids 184–431). An analysis of the primary structure of p65 shows that the fragment obtained by the yeast two- hybrid screen contains three possible tetratricopeptide repeat (TPR) motifs; these are amino acid stretches likely involved in protein–protein interactions and, therefore, are good candidates for the interaction between P0 and p65. Thus, we created three different p65 fragments, each containing one of the TPR motifs (). Using these constructs both in an in vitro binding assay () or pull-down assay (, bottom), we were able to determine that the domain responsible for interaction with P0 is located within the C-terminal 126 amino acids of p65. To more specifically define the p65-binding site on the cytoplasmic tail of P0, we created a series of P0 deletion mutants () and generated cell lines stably expressing these mutants. Lysates of the cell lines expressing P0 were then used in pull-down experiments with GST-p65. Full-length wild-type P0 specifically interacts with GST-p65, and deletion of the first 22 amino acids in the N terminus of the cytoplasmic tail (P0Δ7 and P0Δ6; see ) does not prevent the interaction (). However, P0 carrying deletion Δ5, corresponding to amino acids 173–184, is not pulled down by GST-p65 beads (), and the association of p65 with P0Δ4 (amino acids 185–194) is severely reduced (). These results suggest that p65 interacts with the region between amino acids 173 and 194. We also verified the association between endogenous p65 and P0 by coimmunoprecipitation. Immunoprecipitates of P0 bearing an HA tag from cell lysates using anti-HA antibody followed by immunoblotting with anti-p65 show association between the two proteins in cells expressing wild-type P0 or P0 mutants Δ6 and Δ7 but not P0 mutants Δ4 and Δ5 (). p65 was originally identified as a nuclear (SC65) or nucleolar protein (No55); however, the predicted localization of SC65 using the pSort program (ExPASy Tools) is cytoplasmic. The 55-kD nucleolar protein No55 has an additional 34 amino acids at the N terminus, and the first 18 amino acids have the property of a signal peptide. No55 lacks C-terminal amino acids present in SC65 and has no nuclear localization signal. The predicted localization for No55 is 97% cytoplasmic, suggesting that it may be transported to its final destination via association with another molecule. The association of p65 with P0 suggests a nonnuclear localization. We used two different cell fractionation procedures to analyze the relative localization of p65 in control L (LCo) cells and L cells expressing P0 (LP0; ). p65 is found in the nuclear, cytoplasmic, and membrane fractions in both LCo and LP0 cells; however, more p65 is detected in the membrane and cytoplasmic fraction of LP0 than LCo cells (). In contrast, the amount of p65 detected in the nuclear fraction of LCo cells is about double that of LP0 cells (). These results clearly indicate that the presence of P0 affects a redistribution of p65, reducing its nuclear presence and increasing its cytoplasmic and membrane localization. This is consistent with a dual function for this molecule. A missense mutation found in the p65-binding region of P0, G184R, correlates with a mild form of CMT with variable penetrance (unpublished data). G184 is at the interface of the two deletions that affect p65 binding to P0, prompting us to investigate the possibility that the G184R point mutation may result in a reduced or weakened interaction between P0 and p65. Thus, we generated cell lines expressing P0 carrying this mutation; coimmunoprecipitation assays using anti-p65 antibody show that the interaction between P0 and p65 is much reduced in these mutants as compared with wild-type P0 (). The same results are seen in pull-down assays using GST-p65 (). The results presented so far suggest that p65 interacts with an extended domain in the cytoplasmic tail of P0 and that residue G184 is necessary but not sufficient for the interaction. However, the limits of this domain remain ill defined. In particular, the C-terminal extent of this region overlaps deletion Δ2, a region of the cytoplasmic domain we previously showed was critical for interaction with RACK1 and PKCα. Deletion of this domain also abrogates p65 binding (). Thus, to better define the p65–P0 interaction domain, we designed a series of peptides to use as competitors in the GST-p65 pull-down assay. The sequence and position of the peptides are shown in . Peptide 1 mimics the functionally important RSTKAAS motif. Peptide 2 contains the C-terminal fragment of Δ4, and peptides 3 and 4 include the important G184 residue; however, peptide 4 contains the human G184R mutation. GST-p65 pull-down assays were performed in the presence of increasing concentrations of each peptide. The amount of P0 interacting with p65 was determined by immunoblotting with anti-HA antibody. The immunoblots were scanned, and the density of the immunoprecipitated bands was compared with that of the total amount of P0 in the lysate. The histogram in shows comparisons for each peptide at plateau values. Peptide 1 has no effect; thus, the functionally important RSTKAAS motif is not critical for p65 binding. Peptide 2 reduces the interaction between p65 and P0 by ∼50%, which is consistent with the attenuation of p65 binding in the Δ4 deletion and in the previously analyzed deletion that included the RSTKAAS motif () but extended into the Δ4 region. Surprisingly, peptide 3 containing G184 results in an increased interaction between p65 and P0 (). The effect of this peptide is, in fact, dependent on G184, as peptide 4, which is identical to peptide 3 except that it contains the mutation G184R, neither promotes nor inhibits the interaction between P0 and p65 in pull-down assays (). The same results are obtained when peptides 3 and 4 are used to perturb the binding of GST-p65 fusion protein to the cytoplasmic domain of P0 in an in vitro binding assay: peptide 3 results in an ∼50% increase in binding over control, whereas peptide 4 has no effect (). One possible interpretation of these results is that the region represented in peptides 3 and 4 stabilizes the interaction of P0 and p65 and that G184 is essential for this stabilization effect. The region of P0 defined by peptide 2 is dominant, as a combination of peptides 2 and 3 used in optimal doses is an effective competitor (). Although the interaction of RACK1 and PKCα and the phosphorylation of the PKCα target motif are essential for P0 adhesion function (), RACK1 and PKCα do not interact directly with P0 (unpublished data). Thus, we considered the possibility that p65 acts as an adaptor, linking P0 with RACK1 and, consequently, PKCα. GST-p65 is able to pull down RACK1 independently of P0 expression (, Co cells) and interacts directly with RACK1 in in vitro binding assays (). We next determined whether p65 plays a role in the interaction between RACK1 and P0 in intact cells. The lysates of cells expressing P0 and P0 deletion mutants immunoprecipitated with anti-HA antibody and immunoblotted with anti-p65 antibody shown in were also immunoblotted with anti-RACK1 antibodies (). RACK1 is detected in the P0 immunoprecipitates whenever p65 is present but not in the absence of p65 binding (compare with 6 C). Together, these results indicate that p65 interacts directly with RACK1 and, as p65 interacts directly with P0, further indicate that p65 acts as a bridge between RACK1 and P0, allowing activated PKCα to phosphorylate P0. To further correlate the loss of p65 binding with the loss of interaction between P0 and PKCα, we compared the phosphorylation of serine residues on P0 wild-type and P0 mutants Δ5 and G184R. Confluent cell layers were lysed and immunoprecipitated with agarose-bound anti-HA followed by immunoblotting with a PKC-specific antiphosphoserine antibody. As shown in , the mutant P0 forms are hypophosphorylated when compared with the wild-type control. The data thus far presented imply that p65 acts as a bridge bringing RACK1 and thus PKCα to the cytoplasmic domain of P0. Because the phosphorylation of serine residues 199 and 204 by PKCα is essential for P0 adhesion function (), loss of the P0–p65 interaction should also result in the loss of P0-mediated adhesion. This is indeed the case: L cells expressing P0 deletion mutants Δ4 and Δ5 as well as the G184R mutation, all of which compromise p65 binding to P0, show much reduced P0-mediated cell–cell adhesion when compared with cells expressing equal levels of wild-type P0 (). To ensure that the loss of adhesion is not caused by altered cell surface expression, intact cells were biotinylated using a cell-impermeable biotinylation reagent followed by lysis and immunoprecipitation with anti-HA antibody. All three mutant P0s are found at the cell surface (). The importance of the G184 residue is further reflected in the fact that cells expressing this mutant do not form an adhesive interface as do cells expressing wild-type P0 (). We have previously shown that serine residues 199 and 204 are essential for P0 adhesion function, presumably acting as substrates for PKCα (). If the function of p65 is indeed to position PKCα so it can phosphorylate P0, a mutation at serine residues 199 and 204 that mimics phosphorylation should abolish the need for p65. Thus, we introduced the S199, 204E double mutation into the P0 cDNA constructs containing the Δ4 and Δ5 deletions as well as the G184R point mutation, which are all mutations that abolish p65 binding. As predicted, the ability of cells expressing the P0 deletion mutants and the G184R mutation to form adhesions is rescued by the presence of the double mutation S199, 204E (; compare G184R with GR/SE). The importance of the cytoplasmic region of P0 in the formation and/or stabilization of P0-mediated adhesion is well established (; ). The studies presented here were targeted at understanding the mechanistic basis for the role of the cytoplasmic domain. We previously demonstrated that PKCα-mediated phosphorylation of serines 199 and 204 is essential for function and that RACK1 played a role, possibly in mediating the binding of PKCα with P0 (). We now demonstrate the requirement for an adaptor protein, p65, that links RACK1/PKCα to P0. Our functional analysis shows that p65 interacts directly with a well-defined region of P0; that, in the absence of binding, RACK1 and PKCα are absent from the cytoplasmic domain of P0; and that P0 is hypophosphorylated and is unable to mediate the formation of cell–cell adhesions in spite of the fact that it still expressed at the cell surface. p65 has been previously reported, but in an entirely different context. It was originally cloned from an expression library using an antibody to the rat synaptonemal complex antibody () and was subsequently cloned from a human expression library and localized to the interphase nucleolus (). These previous results were quite surprising; however, it is not unusual to find proteins or alternate transcripts fulfilling very different functions in the cell. This dual functionality is consistent with the redistribution we find when parental L cells are compared with L cells expressing P0. The two putative functions (one nuclear and related to cell division and one cell surface and related to cell–cell interactions) are completely separate temporally and spatially, as Schwann cells in the process of ensheathment are in the final stages of a terminal differentiation program. Deletion analyses of the P0 cytoplasmic domain in conjunction with competition assays using peptides mimicking specific regions of the cytoplasmic domain were used to map the p65-binding region. The P0 site to which p65 binds spans ∼18 amino acids (residues 179–197). The use of peptides mimicking regions of the cytoplasmic domain of P0 as competitors suggests that the p65-binding site has two component parts. This is based on the fact that peptides mimicking the N-terminal half of the binding region (amino acids 179–189) when used in pull-down or in vitro binding assays enhance binding, appearing to prime p65 for stable binding or to stabilize binding to P0. In contrast, peptides mimicking the C-terminal half of the binding region (amino acids 190–199) inhibit binding. Furthermore, in the presence of both peptides, the binding of p65 to P0 is inhibited, suggesting that the C-terminal half site is dominant. However, both regions are essential for binding, as the deletion of either region abrogates binding. Our interpretation of these data is that binding of p65 is stabilized by a change in configuration that requires interaction with the N-terminal half site of P0. Human mutations in the cytoplasmic domain of P0 give rise to CMT with variable severity (). Truncations of the cytoplasmic domain and point mutations are among those associated with the disease. Therefore, it is not surprising to find that the p65-binding region is the site of a point mutation, G184R, giving rise to the disease. What is quite interesting is that this mutation is within the N-terminal half of the p65-binding site, the site we suggest is essential for priming or stabilizing p65 binding. A peptide spanning the N-terminal half of the P0 p65-binding site but containing the G184R mutation no longer is able to inhibit or stabilize binding. The neuropathy resulting from the G184R mutation is late onset and extremely variable in penetrance (unpublished data). Because mutations in P0 are dominant, variable penetrance may well be the result of the relative ratios of the normal versus mutant P0 synthesized. Additionally, because P0 function is suggested to be based on cis-tetramers (), this particular mutation may be compensated by wild-type cis-partners carrying the p65–RACK1–PKCα complex that phosphorylate mutant partners. This is consistent with the fact that mutation of serines 198 and 204 to glutamic acid restores wild-type function to P0s carrying deletions or point mutations that compromise the binding of p65. The machinery we have identified to be associated with the cytoplasmic domain of P0 suggests that like members of the cadherin (; ; ) and integrin () families of adhesion molecules, P0 function is modulated through a set of associated cytoplasmic components designed to regulate phosphorylation (M. ). This further suggests that dephosphorylation also plays a critical role in regulating P0 function (; ; ). Regulated phosphorylation may play a role in the early stages of myelination, as phospho-P0 is most prevalent during the period of maximal myelination (). Possibly, P0 adhesion is turned off and on during this time to regulate the rate of compaction. Additionally, as suggested by , phosphorylation may be a component of the machinery that is critical for the regulation of trans-intracellular membrane interactions essential for normal myelin compaction. In this scenario, phosphorylation would serve to retard maturation by inhibiting interactions of the cytoplasmic domain with phospholipids on the opposing membrane; thus, during maximal periods of myelin formation, intercellular adhesion predominates, whereas maturation is prevented. After completion and/or stabilization of wrapping, phosphorylation is decreased, and compaction ensues. Although the phosphorylation of P0 is clearly essential for adhesion function, it may also play a role in other aspects of myelination. We (; W. ) and others () have reported that transcription of myelin-associated genes is deregulated in the absence of P0, and, thus, phosphorylation may play an important role in downstream signaling processes. These are among the questions that we will be pursuing in the near future. Antibodies used in this study are as follows: anti-GST (Ab-3; Calbiochem), anti-FLAG M2 (Stratagene), anti-HA (Roche Applied Science), anti–maltose-binding protein (MBP; New England Biolabs, Inc.), anti-RACK1, anti-PKCα, and antiflotillin (BD Biosciences), antiphosphoserine antibody specific for PKC (Cell Signaling Technology), HRP-conjugated secondary antibodies (Invitrogen or Jackson ImmunoResearch Laboratories), and agarose-conjugated anti-HA (Bethyl Laboratories). The anti-p65 antibody was prepared in rabbits using amino acid sequence 104–118 as antigen. Antibody AH6 (nucleolar antigen; Developmental Studies Hybridoma Bank at the University of Iowa) was used as a nuclear marker. The assay used was based on the interaction mating method (). cDNA encoding the cytoplasmic domain of P0 was introduced into yeast (strain RF231/pSH18-34) as a fusion with LexA. mRNA from the sciatic nerve of 20–30-d-old rats was use to create a cDNA library fused to a transcription activation domain. The library was then introduced in yeast to create the prey strains. Bait and prey strains grown on selective medium were mated and grown on indicator plates. The cytoplasmic domain of myelin-associated glycoprotein, β1-integrin, and TGFβ type I receptor were used as bait protein controls. p65 and P0 probes were obtained by PCR using the p65 fragment identified in the yeast two-hybrid screen and cDNA for the P0 cytoplasmic domain as templates. The probes were labeled with [P]dATP and [P]dGTP (to normalize for GC content) using random priming and Klenow DNA polymerase. The blots were hybridized and washed using the methods described in . Sciatic nerve transection and regeneration were performed as described previously (). Full-length wild-type P0 cDNA was used as a template for PCR to create P0 cytoplasmic construct (C-terminal 69 residues). The PCR fragment was cloned into the BamHI–SalI cloning site of the pCAL-N-FLAG expression vector (Stratagene) to generate a fusion with CBP. The constructs were confirmed by sequencing. CBP–wild-type P0 was prepared and purified according to the manufacturer's protocol. Purity of the peptide was assessed by Coomassie staining and Western blotting. The 1.1-kb cDNA identified in the yeast two-hybrid assay was used as a template to generate four truncated p65 constructs: p65-1 (residues 178–431), p65-2 (residues 178–210), p65-3 (residues 266–298), and p65-4 (residues 299–431). p65-1, -2, and -3 constructs were ligated into the XbaI–HindIII cloning site of the pGEX 2T expression vector (GE Healthcare), and p65-4 was ligated into the BamHI–SalI cloning site of pGEX 4T2. All constructs were verified by restriction enzyme cleavage and sequencing. Peptides were purified by glutathione affinity chromatography and assayed by Coomassie staining and Western blotting. MBP–full-length human RACK1 was provided by D. Mochly-Rosen (Stanford University, Palo Alto, CA). The plasmid was transformed into strain BL21, and MBP-RACK1 protein was purified from the bacterial cleared lysate by maltose affinity chromatography (New England Biolabs, Inc.). The purity of fusion peptide was confirmed by Coomassie staining after SDS-PAGE. GST-p65 peptides bound to glutathione-Sepharose were equilibrated in binding buffer (20 mM Hepes-KOH, pH 7.9, 50 mM KCl, 2.5 mM MgCl 10% glycerol, 0.02% NP-40, 1.5% goat serum, 2 mM sodium-ortho-vanadate, and protease inhibitor cocktail) for 1 h at 4°C. L cells stably expressing P0 were lysed in mild lysis buffer (20 mM Hepes, pH 7.2, 150 mM NaCl, 3 mM KCl, 2.5% NP-40, 1 mM sodium-ortho-vanadate, 10 μg/ml DNase, and protease inhibitor cocktail [Sigma-Aldrich]) and cleared by centrifugation at 16,000 for 20 min. The supernatants were incubated with the glutathione-Sepharose–bound GST-p65 peptides overnight at 4°C. Beads with bound protein complexes were collected by centrifugation at 500 and washed three times with 10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.2% NP-40, 2 mM Na-o-vanadate, and protease inhibitor cocktail (Sigma-Aldrich). Beads were resuspended in SDS sample buffer and analyzed by SDS-PAGE and Western blotting. High protein binding wells (Pierce Chemical Co.) were incubated with 50 μl of 20 μg/ml CaM in 50 mM bicarbonate buffer overnight at 4°C. The wells were washed in PBA (PBS with 0.25% BSA and 0.02% NaN), blocked for 1 h at 37°C in 5% nonfat milk in PBS, washed in PBA three times, and incubated with a previously determined saturating concentration of CBP-P0 cytoplasmic domain in PBA for 1 h at 37°C. After washing three times in PBA, the wells were incubated with increasing concentrations of GST alone or GST–p65-1 for 2 h at room temperature followed by several washes with PBA. The wells were then incubated for 1 h with 1:1,000 anti-GST antibody in PBS followed by washing and incubation with HRP-conjugated goat anti–mouse IgG (1:1,000) in PBS with 0.25% BSA. The wells were thoroughly rinsed with PBS and incubated with -phenylene diamine substrate for 30 min. The reaction was stopped by adding 2 M HSO, and absorbance was determined at 450 nm in a spectrophotometer (Spectramax Plus; Molecular Devices). All samples were assayed in triplicate, and the experiment was repeated three times. Statistics and the dose-response graph were computed with Kaleidagraph software (Synergy). Confluent cell layers were washed in PBS and lysed in buffer containing 20 mM Hepes, pH 7.9, 50 mM KCl, 2.5 mM MgCl, 10% glycerol, 1% Triton X-100, 5 mM NaF, and protease inhibitor cocktail (1 ml/10-cm plate; Sigma-Aldrich) for 10 min at 4°C. The lysates were cleared by centrifugation at 14,000 , and the supernatant was incubated with anti-HA or control antibody for 2 h at 4°C, with rotation followed by 1 h with anti–rat IgG covalently attached to magnetic beads. The beads were then collected using a magnetic stand, washed extensively with PBS containing 0.5% Triton X-100, and eluted with SDS sample buffer. Eluted material was fractionated on SDS-PAGE, transferred to polyvinylidene difluoride, and assayed by immunoblotting with the appropriate antibody. Single cells were prepared from semiconfluent cell layers: layers were washed with PBS, incubated for ∼2 min with 0.002% trypsin in PBS, and collected in complete medium (DME with 5% FBS) with 5 μg/ml of added antipain and 10 μg/ml DNAase. Cells were collected by centrifugation at 1,000 , resuspended in DME (wih 20 mM Hepes, pH 7.4, antipain, and DNAase), and counted in a hemocytometer (AO-Spencer Brightline; Reichert Scientific Instruments). The cell suspensions were diluted to a concentration of ∼10/ml, and 1 ml was added to 30-mm Petri plates containing 2 ml DME/Hepes. Dishes were rotated at 70 rpm in a humidified chamber at 37°C. After ∼4 h, cells were observed and photographed under a microscope (Axiovert 25CFL; Carl Zeiss MicroImaging, Inc.). Several fields of cells were used to quantify cell numbers using ImagePro Plus (Media Cybernetics). Results were tabulated as ratios of cells in aggregates versus total cell number for a minimum of five fields. For biotin labeling of cell surface P0, cell layers were washed in PBS and incubated with the membrane-impermeable biotinylation reagent sulfo- -hydroxysuccinimide-SS-biotin (Pierce Chemical Co.) at a concentration of 1 mg/ml in PBS for 30 min at room temperature. Cell layers were washed three times with ice-cold PBS, pH 8.0, and lysed as described in the Coimmunoprecipitation assays section. Cleared cell lysates were incubated with streptavidin-conjugated magnetic beads (Roche Applied Sciences) for 1 h, the beads were extensively washed, and bound material was eluted with SDS sample buffer and analyzed by immunoblotting with anti-HA antibody. For cellular fractionation, confluent layers of control L cells or cells expressing wild-type P0 (LP0) were washed in ice-cold PBS and scraped in 0.25 M sucrose in Hepes buffer, pH 7.9, containing protease inhibitor cocktail (Sigma-Aldrich). Cells were homogenized, and the nuclear and membrane fractions were separated using Optiprep (Sigma-Aldrich) according to the manufacturer's directions. Alternatively, L control and LP0 cells were fractionated into nuclear and cytoplasmic fractions using the Ne-PER kit (Pierce Chemical Co.). L cells expressing P0 or the P0 mutant G184R were grown on poly--lysine–coated coverslips, washed free of serum, fixed in 4% PFA for 20 min at room temperature, and permeabilized in 0.1% Triton X-100 for 5 min. After washing in PBS, the coverslips were incubated with rat anti-HA for 1 h at room temperature. The coverslips were washed extensively and incubated for another hour in AlexaFluor488 anti–rat antibody diluted in PBS with 5% goat serum. After several washes, the coverslips were mounted on glass slides, and images were captured using an inverted laser-scanning confocal microscope (TCS SP2 AOBS; Leica).

published one of the first papers showing that the cross-linking of a glycosylphosphatidylinositol-anchored receptor (GPI-AR), such as CD59 and decay accelerating factor (DAF), induces the formation of molecular complexes with Src-family kinases (SFKs) and their activation by phosphorylation (; ). However, the GPI-ARs are located on the extracellular surface and are only anchored to the outer leaflet of the plasma membrane, and the SFKs are located in the cytoplasm or in/on the cytoplasmic leaflet of the cell membrane. Therefore, the mechanism by which GPI-ARs form complexes with SFKs and activate them has been a long-standing enigma. Because GPI-ARs and many SFKs tend to partition into the detergent-resistant membrane (DRM) fractions, the involvement of so-called raft microdomains in GPI-AR–mediated signal transduction has been proposed (; ; ; ; ). Furthermore, using immunofluorescence staining, found that when placental alkaline phosphatase (PLAP), a GPI-AR, was cross-linked, an SFK, Fyn (another putative raft-associating molecule, with myristoyl and palmitoyl anchoring chains), was recruited to the cross-linked PLAP in fibroblastic cells (the term recruit is used here for brevity, to indicate that the cytoplasmic molecules come and stay right beneath the cross-linked GPI-ARs). In addition, found that CD59 clustering in the outer leaflet induces the colocalization of CFP, anchored to the inner leaflet via two saturated chains in patches, in a cholesterol-dependent manner. Meanwhile, by using immunoprecipitation, discovered the association of GPI-ARs (CD59, CD48, and Thy-1) with Gαi2 and Gαi3 in lymphocytes. Therefore, how Gαis and SFKs on/in the cytoplasmic leaflet are recruited at cross-linked GPI-ARs in the outer leaflet, and how Gαis and SFKs become activated, is one of the key issues in signal transduction studies of GPI-ARs. These two lines of research converge because the activation of SFKs by the binding of Gαs and Gαi has been established. reported that Gαs and Gαi, but not Gαq, Gα12, or Gβγ, are the major activators for SFKs. With regard to SFK activation, there are several steps or different pathways. Dephosphorylation at the tyrosine residue near the C terminus (Y508 for Lyn) is usually necessary for the activation, and further activation is achieved by the phosphorylation of another tyrosine residue (Y397 for Lyn and Y418 for c-Src) in the activation loop, perhaps via autophosphorylation. However, showed another pathway in vitro: the binding of Gαs or Gαi to SFKs induces the activation of SFKs, without the need for dephosphorylation of the tyrosine near the C terminus. Therefore, we performed research based on a working hypothesis in which Gαi and SFKs are coincidentally recruited to GPI-AR clusters to form a complex, leading to SFK activation (; ). This working hypothesis is also consistent with the results of pull-down assays, showing the coupling of folate receptor Gαi3-Lyn or Gαi-Src (; ). In this research, we mostly focused on CD59 as our model GPI-AR, but DAF and PLAP gave essentially the same results. These pioneering studies collectively raise two further key questions. The first question is the recruitment dynamics of signaling molecules: how are Gαi and SFK molecules recruited to a CD59 cluster, and how long do they stay with the cluster for each recruitment event? We approached this problem by investigating the recruitment of cytoplasmic signaling molecules to GPI-AR clusters at the level of single molecules in living cells. By tracking single molecules in living cells, we should be able to learn how molecular interactions and recruitment take place in a spatiotemporally organized way. Conventional methods only observe the molecular events averaged over a large molecular ensemble, as well as over time and space, and therefore could easily miss dynamic and transient events. The second question is whether previous observations might be a consequence of extensive aggregation of GPI-ARs induced by antibodies. Almost all of the receptor molecules on the cell surface might be cross-linked to form many large clusters. Therefore, we used colloidal gold particles with a 40-nm diameter, conjugated with whole IgG antibody, to induce clusters of a mean of six CD59 molecules. We also limited the number of gold particles attached to the cell to ∼600 gold particles/cell, thus inducing the engagement of ∼4,000 copies of CD59. Under these conditions, signaling responses comparable to those under physiological conditions were observed. Others found that when raft-associating molecules were cross-linked with a colloidal gold particle with a 40-nm diameter, they exhibited alternating periods of apparently simple diffusion and temporary confinement within a zone (every 1–10 s; ; ). These regions were termed transient confinement zones (TCZs). Based on these observations and our working hypothesis, we examined the possibility that TCZs might be induced by Gαi and/or SFKs recruited to gold-induced CD59 clusters. We simultaneously tracked the movements of single GPI-AR clusters and single molecules of Gαi2 and Lyn (an SFK). Our selections of Gαi2, among the other Gα's, and of Lyn, among the other SFKs, are based on the results by and , as well as on our immunofluorescence and biochemical results using the human epithelial T24 used in this study (see Results). We critically examined the timing of the recruitment of Gαi2 and Lyn with respect to the occurrence of TCZs. In this paper, we describe how the TCZ of CD59 clusters may be induced by dynamically and transiently recruiting individual Gαi and Lyn molecules at the cluster, and in the companion paper (see Suzuki et al. on p. of this issue), we report the function of the TCZ. Experimental data, examining the involvement of raft microdomains in the signaling of GPI-ARs, are shown in many display items in this paper, but they are collectively discussed in . In most of the experiments described here, the human epithelial cell line T24 was used. It expresses CD59, Gαi2, and Lyn, and therefore it is expected to undergo the nonlethal signaling responses found in immune cells upon the engagement of CD59 (; ). Because CD59 is ubiquitously expressed in most cells (), the results obtained here would be useful in a wide variety of cells. Unless otherwise stated, all of the single-molecule tracking experiments were performed at 37°C. Various CD59 and control probes used in this work are shown in . Upon the addition of the cross-linking gold probe (IgG-gold; , right) at a concentration of 1.8 × 10 particles/ml, 610 ± 54 (mean ± SEM; = 9 cells) particles/cell were bound to the T24 cell surface at 5 min after the probe addition, whereas almost all of the single-molecule and single-particle tracking experiments were performed within 5 min after the addition of the gold particle (). Almost all of them were found on the apical (dorsal) surface: the 40-nm gold particles probably did not readily enter the space between the bottom (ventral) membrane and the coverslip. Such IgG-gold binding was sufficient to elicit intracellular signaling, at levels comparable to that induced by CD59's natural ligand, C8, at a serum (therefore high) concentration (at a cytolytic membrane attack complex unit of 1,000/coverslip; ; , third from left). The amplitude and the time course of Lyn (and other SFKs) phosphorylation in its activation loop (at Y397, as detected by antibodies to phosphorylated Y418 of c-Src) after the binding of ∼600 IgG-gold probes were similar to those after C8 addition (; ). The intracellular inositol- triphosphate (IP)–Ca responses upon the additions of IgG-gold or C8 were also very similar to each other (see – in ). These results show that CD59 could act as a signaling receptor in epithelial cells, like in immune cells, as well as a complement control factor (; ). The IgG-gold–induced signaling was specific with regard to GPI-anchoring and cross-linking. Multivalent gold probe–induced cross-linking of a transmembrane mutant of CD59 (mycCD59TM; CD59 with a myc tag at its N terminus and the transmembrane domain plus the following 12 amino acids of the low-density lipoprotein (LDL) receptor attached at its C terminus; ; , left; 560 ± 51 particles attached per cell; = 5 cells) and a nonraft phospholipid, 1,2-dioleoyl--glycero-3-phosphoethanolamine (DOPE; , right; 670 ± 48 particles attached per cell; = 8 cells), did not induce intracellular signaling responses. Control gold particles (Fab-gold; , left), prepared by conjugating small numbers of Fab fragments of the anti-CD59 antibody to gold particles, did not elicit either Lyn activation () or IP–Ca signaling (see – in ). (This in itself strongly supports the proposal that Fab-gold is not cross-linking.) An independent observation further indicated that Fab-gold practically does not cross-link CD59, although one should be aware that 20–40% of GPI-anchored proteins may be in clusters smaller than pentamers, with the remaining 60–80% being monomers (). The diffusion coefficient of Fab-gold attached to CD59, observed by single-particle tracking on a 100-ms time scale (D, corresponding to D in ), was the same as that for Fab-Cy3–labeled CD59, estimated by single fluorescent molecule tracking (, left; 0.19 vs. 0.20 μm/s, respectively), whereas the single fluorescent molecule tracking method is sensitive to the difference between IgG-Cy3–labeled CD59 and Fab-Cy3–labeled CD59 (median of 0.09 and 0.20 μm/s, respectively; P < 0.01). The D of IgG-gold is only half that of Fab-gold after C8 addition (in median values of 0.022 vs. 0.042 μm/s, respectively; the Fab-gold D decreases from 0.19 to 0.042 μm/s upon C8 addition), suggesting that the sizes of the IgG-gold–induced CD59 clusters are only slightly greater than those of Fab-gold + C8, based on the concept of oligomerization-induced trapping within the membrane skeleton mesh, as proposed by (Fig. S1, available at ). Therefore, the CD59 cluster size beneath the cross-linking IgG-gold may be comparable to that induced by C8 addition (, right). Single fluorescent molecule tracking revealed that a fluorescent spot representing the complex of five Cy3 anti–CD59-IgG antibody molecules and CD59 (clustered by a secondary antibody and not the IgG-gold), namely, a spot containing 5–10 molecules of CD59, was practically immobile (Fig. S2, available at ). As our IgG-gold probes attached to the cell surface were still mobile, this indicates that the number of CD59 molecules beneath an IgG-gold particle would be <10. Meanwhile, the diffusion coefficient D of IgG-gold (median = 0.022 μm/s) is smaller than those for Fab-gold (0.19 μm/s) and IgG-Cy3–labeled CD59 (median = 0.09 μm/s) by factors of approximately nine and four, respectively, suggesting that IgG-gold may cross-link more than two CD59 molecules. Therefore, it would be safe to conclude that an IgG-gold–induced CD59 cluster contains three to nine molecules of CD59, with six CD59 molecules/IgG-gold–induced cluster as a reasonable number to use, although these IgG-gold–induced clusters may contain other recruited proteins. The terms CD59 cluster and GPI-AR cluster are used with this understanding. It follows that ∼600 hexameric (mean) CD59 clusters, totaling ∼3,600 CD59 molecules per cell, on average, are clustered or engaged in the present work, and that this will be sufficient to trigger robust intracellular SFK activation and calcium mobilization at physiological levels. We next observed the movement of each individual Fab-gold (before and after the C8 addition; Fab-gold-C8) and IgG-gold on the cell surface at video rate (33 ms/frame; and Video 1, available at ). Both Fab-gold-C8 and IgG-gold exhibited apparently simple, but slow, diffusion (at a 300-fold–enhanced frame rate of 0.11 ms/frame, both probes exhibited hop diffusion, with a mean compartment size comparable to that found previously by ; median = 110 nm, at a mean hop rate of once every 200 ms [ = 37]). However, in addition to this apparent simple diffusion observed at video rate, both Fab-gold-C8 and IgG-gold were often temporarily immobilized, as seen in the trajectories in , as statistically detected by software designed to find TCZs (; ; ; ; Fig. S3, available at ; and ). As shown in and in , our results indicate that these temporary immobilization events might be induced by the binding of CD59 clusters to actin filaments (must be bound indirectly) and/or actin-associated membrane microdomains, because they are inhibited by partial depolymerization of actin filament (); therefore, to avoid ruling out the possibility of actin binding (rather than partitioning into zones), we call this temporary immobilization stimulation-induced temporary arrest of lateral diffusion (STALL). Both Fab-gold-C8 and IgG-gold intermittently exhibited STALL, separated by periods of apparently simple Brownian diffusion. The distributions of the durations of STALL and simple diffusion are shown in the histograms shown in , which could be fitted with single exponential decay curves, giving the characteristic lifetimes (exponential decay constants) for the STALL and mobile periods of 0.57 and 1.2 s, respectively, for IgG-gold. This would predict that IgG-gold is in the STALL period ∼32% of the time. Actually, as shown in , IgG-gold exhibited STALL for 162 s out of the total observation period of 440 s, or 36% of the total observation time. The CD59 clusters induced by the binding of C8 (Fab-gold-C8) exhibited STALL for 17% of the observation time (67 s out of 410 s), whereas nonstimulated CD59 molecules (Fab-gold) were temporarily immobilized for only 4.5% of the observation time (24 s out of 535 s). In contrast, DOPE or mycCD59TM, non-DRM/non-putative raft molecules, did not exhibit STALL behavior, even after cross-linking by IgG-gold particles (). For further discussion of membrane compartments and STALL, with their detection with a higher time resolution of 0.11 ms, see Fig. S4. None of the STALL sites were colocalized with caveolae, as marked by caveolin 1–GFP, under the experimental conditions used here (only the initial events, generally within 5 min after IgG-gold addition were observed). For further data, see Fig. S5. The following four observations suggested the involvement of Lyn in inducing STALL of CD59 clusters, formed by IgG-gold particles. First, after the cells were treated with the SFK inhibitor 4-amino-5-(4-chlorophenyl)-7-(-butyl)pyrazolo[3,4-d] pyrimidine (PP2) or pertussis toxin (PTX), an inhibitor for Gαi's, both the Lyn activation (Lyn phosphorylation at Tyr397) and STALL occurrence were inhibited ( and ). These results suggest that the activities of Lyn and Gαi's are required for inducing STALL (according to , Lyn is activated by the direct binding of Gαs and Gαi). Second, other GPI-ARs (PLAP and DAF) in T24 cells or CD59 in other cell lines (NRK and PtK2) also exhibited PP2-dependent STALL, indicating the SFK requirement for inducing STALL (). Third, the temporal fraction of the CD59 cluster in STALL was only 2.8% in SYF cells, which do not express Lyn, Src, Yes, Fyn, and other SFKs, but the STALL temporal fraction increased to 20% after Lyn expression (). In addition, IgG-gold on YF cells, which do express c-Src but not other SFKs, exhibited a 9.6% temporal fraction of STALL. DOPE, a nonraft molecule, did not exhibit a significant level of STALL under any conditions (). These results indicate the necessity of Lyn activity (or c-Src and other SFK activity) for the induction of STALL. Fourth, the treatment of cells with methyl-β-cyclodextrin (MβCD) or saponin, for partial depletion or clustering of cholesterol, blocked both the Lyn activation and STALL of the CD59 clusters. Furthermore, the subsequent replenishment of cholesterol with cholesterol-loaded MβCD reinstated both the Lyn activation and STALL of the CD59 clusters ( and ), showing the high correlation between the STALL occurrence and Lyn activation (see Discussion in ). These observations led us to propose a model in which Lyn is recruited to the molecular complex centered on CD59 clusters and is activated in the molecular complex (by the binding of Gαi2 as described later) and, in turn, induces STALL by phosphorylating as-yet-unknown proteins, which might mediate the binding of CD59 clusters to actin filaments or to domains supported by actin filaments ( and in ). We simultaneously observed each individual CD59 cluster and single molecules of Lyn-GFP (known to be functional; ) expressed in T24 cells (). For such observations, 50-nm latex particles, rather than 40-nm colloidal gold particles, were used to form CD59 clusters, to avoid the signal from the gold particles (see Materials and methods). shows typical trajectories of a CD59 cluster (black), including four STALL periods, and a single Lyn-GFP molecule (orange trajectory), including a period of colocalization (the magenta part of the Lyn-GFP trajectory and the indigo part of the CD59 cluster trajectory; Video 2, available at ). Lyn-GFP is generally located on the cytoplasmic surface of the membrane and is recruited at diffusing CD59 clusters by lateral diffusion on the cytoplasmic surface. However, this Lyn recruitment occurred with no correlation with the STALL of CD59 clusters (). Based on many such experiments, we obtained a histogram of the time difference between the onset of STALL of a CD59 cluster (time 0) and the recruitment of a single Lyn-GFP molecule to that particular CD59 cluster (). The number of incidental overlaps of a Lyn-GFP spot with a CD59 cluster, estimated by up-shifting the CD59 cluster video sequence by 50 video frames (1.67 s), was subtracted (see Materials and methods). This histogram clearly indicates that Lyn is frequently recruited to CD59 clusters but without any correlation with the STALL events. This lack of a temporal correlation between the Lyn-GFP recruitment to a CD59 cluster and the induction of the CD59's STALL was surprising because Lyn is required for STALL induction, as described in the previous subsection (– ). The Lyn recruitment results () suggest that, although Lyn activity is needed for inducing STALL (– ), an upstream molecule that activates Lyn may also have to be recruited to the same CD59 cluster for inducing its STALL. Lyn can be activated by the binding of Gαs and Gαi, but not Gαq, Gα12, or Gβγ, without dephosphorylating the tyrosine residue Y508 near the C terminus (). In fact, we made three observations that suggest the involvement of Gαi2 in Lyn activation and the STALL of CD59 clusters. In the biochemical experiment shown in , we found that IgG-gold–induced Lyn activation is blocked by PTX, a Gαi blocker. PTX also blocked the STALL of CD59 clusters (). Immunofluorescence observations revealed a certain level of colocalization of Gαi2 with CD59 clusters ( and ; but much less in the case of Gαi1, Gαi3, and Gαs, perhaps because of their low expression levels; not depicted). Because PTX totally blocked the IgG-gold–induced Lyn activation, it is difficult to conceive any pathways of Lyn activation other than those mediated by Gαi's. Based on these observations, we formed a working hypothesis in which Gαi2 is also recruited to CD59 clusters, and when it encounters Lyn at the CD59 cluster, it binds and activates Lyn there, inducing STALL of the CD59 cluster by its binding to an actin filament. Therefore, we examined Gαi2 recruitment to CD59 clusters. We simultaneously observed single molecules of Gαi2(YFP) and CD59 clusters (). A typical trajectory of a CD59 cluster that includes a STALL period (, blue), as well as a period of colocalization with a single Gαi2(YFP) molecule (the magenta part of the Gαi2[YFP] trajectory and the indigo part of the CD59 cluster trajectory), is shown in (Video 3, available at ). First, note that, at variance with the general concept of higher membrane localization of trimeric G proteins and Gαi2 than in the cytoplasm, and also in contrast to Lyn-GFP, the residency time of Gαi2(YFP) on the cytoplasmic surface of the plasma membrane is short, on the order of 100 ms (Video 4), consistent with the observations made by and . The colocalized Gαi2(YFP) molecules usually stayed on the membrane only during the colocalization period or longer by only a few video frames. Gαi2(YFP) molecules must be frequently colliding with the inner surface of the plasma membrane, but they were invisible at video rate because they rarely stay on the membrane longer than a video frame time of 33 ms. They stay longer (median of 133 ms) when they bump into CD59 clusters. Interestingly, the recruitment of a Gαi2(YFP) molecule to the CD59 cluster took place right before the beginning of STALL, as exemplified by the trajectories shown in . The distribution of the time difference between the recruitment of Gαi2(YFP) to a CD59 cluster and the onset of STALL () clearly indicates that right after the recruitment of a Gαi2(YFP) molecule to a CD59 cluster, its STALL is initiated, showing a marked contrast with the Lyn recruitment. These results, therefore, are consistent with our working hypothesis, in which the CD59 cluster provides a platform for temporarily recruiting Lyn and Gαi2 and for inducing Lyn activation by the binding of newly recruited Gαi2. Such dynamics would never have been revealed by normal imaging or biochemical methods. The colocalization period of Lyn-GFP at a CD59 cluster was 200 ms (median length of 6 video frames; ). Because this recruitment duration is substantially shorter than the GFP lifetime before photobleaching and blinking in the plasma membrane (∼850 ms), these are not responsible for the short residency time. Therefore, Lyn-GFP molecules are joining and leaving the CD59 clusters continually and rapidly. Very similar observations were made with the recruitment of Gαi2(YFP) to CD59 clusters. The duration of Gαi2(YFP) colocalization with a CD59 cluster is even shorter, ∼133 ms (median of 4 video frames; ). Short-term recruitment of cytoplasmic molecules to the plasma membrane was also reported previously (). These results indicate that the recruitment of Gαi2 and Lyn molecules at a CD59 cluster takes place quite dynamically. However, as described in previous subsections, the physiological CD59 clusters formed by C8 addition are even slightly smaller than those induced by IgG-gold. Therefore, such dynamic recruitment and short dwelling times are not likely to be artifacts as a result of the small sizes of CD59 clusters. Because Lyn activation lasts for >30 min, we initially assumed that each Lyn molecule stays at a CD59 cluster for at least several minutes and were surprised to find the recruitment period of ∼200 ms on average. However, from the viewpoint of regulating the periods and levels of overall Lyn activation, simple addition of many such short recruitment events would be easier than carrying out complex integration of prolonged activation of individual molecules (see in ). The frequency of transient single-molecule recruitment of Lyn-GFP to CD59 monomers, clusters, and mycCD59TM clusters (per minute), normalized by the number density of Lyn-GFP on the plasma membrane (justified because Lyn-GFP recruitment to CD59 always occurs in the membrane), is summarized in . The Lyn-GFP recruitment frequency to the CD59 cluster is about eightfold greater than that of non–cross-linked CD59 or the mycCD59TM cluster. This result is somewhat consistent with the results shown in , in which the temporal fraction in STALL and the frequency of STALL were increased by factors of approximately eight and three, respectively, as CD59 was clustered. To examine the relative importance of lipid–lipid and protein–protein interactions in the recruitment of Lyn-GFP to CD59 clusters (), the recruitment of LynN20-GFP, the N-terminal 20-amino-acid sequence of Lyn, which contains the binding sites for a palmitoyl and a myristoyl chain, fused at its C terminus to GFP (), was examined. LynN20-GFP is recruited to the CD59 cluster 2.4 times less often than Lyn-GFP but 3.3 times more often than the controls (Lyn-GFP to non–cross-linked CD59 or to the mycCD59TM cluster). Furthermore, the residency durations of LynN20-GFP at CD59 clusters and Lyn-GFP at mycCD59TM clusters are considerably shorter than that of Lyn-GFP at CD59 clusters (unpublished data). These results suggest that both lipid–lipid interactions via Lyn's alkyl chains and protein–protein interactions by way of Lyn's protein moiety contributed to the recruitment of Lyn to the CD59 cluster. Furthermore, single molecules of transferrin receptor, a typical nonraft transmembrane protein, labeled with Alexa488-conjugated transferrin, were not significantly recruited to CD59 clusters, suggesting the involvement of “raft-dependent” mechanism for Lyn recruitment and excluding the possibilities of membrane accumulation beneath the CD59 clusters (). First, we extensively cross-linked CD59 by the successive addition of anti-CD59 IgG and secondary antibodies and found that CD59 aggregates were colocalized by Lyn (, left), Gαi2 (middle), or both (right). Note the extensive CD59 aggregation, which is the condition often used for triggering the intracellular signaling. Second, using IgG-gold particles, we observed that ∼20 and 10% of IgG-gold particles were colocalized with Lyn (as well as pY418) and Gαi2, respectively (). Noting that the immunofluorescence result represents a spatiotemporal mean of dynamic events occurring over the periods needed for chemical fixation and over a diffraction-limited space of ∼400 nm, such low levels of colocalization are consistent with the transient recruitment of Lyn and Gαi2 at CD59 clusters. Namely, the dynamic, transient recruitment of Lyn and Gαi2 should not be considered inconsistent with their extensive immunocolocalization to each other at the large CD59 clusters (), Fyn colocalization with PLAP clusters (), or colocalization of Lyn and Gαi2 at IgG-gold particles (). Both PTX, a Gαi blocker, and PP2, an SFK blocker, reduced Lyn recruitment by about twofold (, middle), whereas they further blocked enhanced Lyn activation at CD59 clusters by four- to fivefold (pY418; , top). These results suggest that, upon the activation of Lyn at the CD59 clusters, the activated Lyn stays at the CD59 clusters longer than nonactivated Lyn molecules by a factor of approximately two. Blocking of SFK activation by PP2 did not affect the Gαi2 recruitment. Fluorescein-DOPE, a typical nonraft phospholipid, preincorporated into the plasma membranes before the addition of IgG-gold, was not concentrated in the CD59 cluster domains (), indicating that membrane concentration and membrane undulations, as reported by , are not involved in the colocalization of Lyn and Gαi2 found here. The low levels of immunocolocalization of Lyn and Gαi2 with IgG-gold–induced CD59 clusters, 20 and 10%, respectively, suggest that the colocalization of Lyn and Gαi2 to each other at the same IgG-gold–induced CD59 cluster (three-way colocalization) occurs rarely in the immunocolocalization observations. The three-way quantitative immunocolocalization results (non–single-molecule experiments) showed that ∼4% of the CD59 clusters were colocalized with both Lyn and Gαi2, which is to be compared with the result after MβCD treatment, exhibiting ∼1% three-way colocalization (). The comparison of these results with those using antibody-induced large-scale CD59 clusters (extensive colocalization of Lyn and Gαi2) suggests that an increase in the cluster size would enhance the total number of recruited Lyn molecules during a unit time but possibly without prolonging the residency time of each recruited Lyn molecule. We found that robust SFK phosphorylation (), increases in IP concentration, and Ca mobilization () were induced after the binding of ∼600 IgG-gold particles to the cell surface, which probably engage ∼3,600 CD59 molecules. Gold probe–induced cross-linking of mycCD59TM or DOPE (again, ∼600 gold particles were bound to the cells), which are putative nonraftophilic molecules, did not trigger any observable intracellular signals (). CD59 clusters undergo alternating periods of apparent simple Brownian diffusion (1.2-s lifetime) and STALL (0.57-s lifetime). As described (), STALL is likely to be a key, but temporary, site for relaying the extracellular CD59 signal to the intracellular Ca signaling pathways. Namely, single-molecule tracking found that PLCγ2 molecules are recruited to CD59 clusters only during the STALL periods, and various blocking experiments suggested that PLCγ2 molecules recruited to the STALLed CD59 clusters might produce IP from PIP, which leads to the release of Ca through the IP channel located in the ER membrane. In the present paper, we describe how STALL is induced after the engagement of CD59. Both Gαi2 and Lyn are necessary for inducing STALL of CD59 clusters (). In the cell line lacking SFKs, STALL could not be induced, but STALL was reinstated after transfecting Lyn (). Because PP2, an SFK blocker, inhibited STALL, Lyn phosphorylation in its activation loop may be mediated by autophosphorylation after the initial Lyn activation by Gαi2 (). Single-molecule tracking revealed the recruitment of Gαi2 and Lyn molecules to CD59 clusters. Surprisingly, this occurs transiently (median of 133 and 200 ms, respectively; and ). Although the short recruitment interaction durations could not be logically excluded, these were unexpected. This finding necessitates the reevaluation of our basic concept on how signaling exactly occurs in living cells (see in ). The recruitment of single Lyn molecules to CD59 clusters takes place without any temporal correlation with the occurrence of STALL ( and ), although Lyn activation is required for STALL. In contrast, a CD59's STALL and its Gαi2 recruitment are temporarily highly correlated: STALL is initiated right after the recruitment of each Gαi2 molecule. Furthermore, pharmacological blocking of Gα's inhibited STALL. Collectively, these results suggest that when Gαi2 is recruited to a CD59 cluster (and because Lyn is recruited often), it may bind to and activate the recruited Lyn () at the CD59 cluster, during the short residency periods of both molecules (the colocalization of Lyn and Gαi2 at CD59 clusters was shown at the immunofluorescence level in and ). The direct, single-molecule detection of Gαi2 binding to Lyn and the subsequent Lyn activation at a CD59 cluster is one of the key remaining issues for future studies. The recruitment of Gαi2 and Lyn to CD59 clusters might be mediated by a transmembrane protein (, X), and the recruitment of protein X might be facilitated by raft-based interactions (). The recruitment of Lyn and Gαi2 to CD59 clusters might also be mediated by raft domains located in the two leaflets, perhaps because of interdigitation (; ; ). The latter interaction may be weak but sufficient to cause or facilitate brief recruitments for 0.1–0.2 s. The involvement of raft-based lipid–lipid interactions in these processes, as suggested by the recruitment of LynN20-GFP to CD59 clusters (although at lower frequencies; ), in addition to protein–protein interactions, will be comprehensively discussed by . SYF mouse embryonic fibroblast cells (), which lack c-Src, Fyn, Yes, and Lyn, and YF cells, which express c-Src, but not Fyn, Yes, or Lyn, were cultured in DME (Invitrogen) containing 10% FBS. The PtK2, NRK, SYF, and YF cells were transfected with the cDNA for human CD59. These cells express CD59 of their own species, but to carry out experiments under similar clustering conditions, they were transfected with the human cDNA for CD59. To examine the involvement of Lyn in STALL induction, SYF cells were transfected with the cDNA for Lyn fused with GFP at the C terminus (Lyn-GFP); this GFP construct is known to be functional (). Transfection was performed using Lipofectamine Plus (Life Technologies) according to the manufacturer's recommendations. Partial depletion of cholesterol in the plasma membrane was performed by incubating the cells in 4 mM MβCD (Sigma-Aldrich) at 37°C for 30 min () or in 60 μg/ml saponin (Sigma-Aldrich) on ice for 15 min (). These treatments substantially increased the amount of CD59 recovered in the detergent-soluble fractions in the protocol to prepare DRM. Replenishment of cholesterol was performed by incubating the cholesterol-depleted cells in 10 mM MβCD–cholesterol complex (1:1) for 30 min at 37°C (). The overall amounts of cholesterol per cell after cholesterol depletion with MβCD and after the subsequent repletion were found to be 66 and 118% of the original amount (SD of ± 6%), as determined by a cholesterol E-test kit (Wako). Partial actin depolymerization was performed by incubating the cells in medium containing 50 nM latrunculin B for 10 min (gifts from Dr. G. Marriott, University of Wisconsin–Madison, Madison, WI; ). SFKs were inhibited by treating the cells with 10 μM PP2 (Calbiochem) for 5 min at 37°C (). Heterotrimeric G protein was inhibited by incubating the cells in medium containing 1.7 nM PTX (Calbiochem) at 37°C for 22 h (). To visualize caveolae and coated pits, T24 cells were transfected with caveolin 1–GFP cDNA (a gift from T. Fujimoto, Nagoya University Medical School, Nagoya, Japan; ) and AP2α-GFP (a gift from J. Keen, Thomas Jefferson University, Philadelphia, PA), respectively. Their distributions in the cell totally overlapped with the fluorescent spots after immunostaining with anti–caveolin 1 and anti-clathrin antibodies, respectively, suggesting that their distributions faithfully represent those of caveolae and clathrin-coated pits. The bright-field image of gold particles and cells and the fluorescence image of GFP were simultaneously observed. As a control for GPI anchoring, a transmembrane chimeric protein of CD59 was used (mycCD59TM; its cDNA was provided by M. Maio, Instituto Nazionale di Ricovero e Cura a Carattere Scientifico, Ancona, Italy): the CD59 ectodomain was fused with an N-terminal myc tag and a C-terminal LDL receptor transmembrane domain, which additionally contains the 12 amino acids from the N terminus of the cytoplasmic domain of the LDL receptor (and thus lacks the sequence required for internalization via coated pits; ). IgG-gold particles coated with anti-myc antibody (9E10.2) were used for cross-linking mycCD59TM. To examine the role played by the lipid anchoring chains of Lyn in its recruitment to CD59 clusters, the cDNA for LynN20-GFP, the N-terminal 20-amino-acid sequence of Lyn, which contains the binding sites for palmitoyl and myristoyl chains, fused at its C terminus to GFP, was prepared (), and T24 cells were transfected with the LynN20-GFP cDNA. Fab-gold particles (40 nm in diameter) were prepared as reported previously (). For the preparation of Fab-gold for labeling CD59 (but not triggering the cellular signaling responses) or DOPE (conjugated to fluorescein, which was used as a tag, rather than a fluorescent probe; Invitrogen), one third of the minimal protecting amount of anti-CD59 Fab (MEM43/5 mouse monoclonal antibody; a gift from V. Horejsi), anti-PLAP Fab (rabbit polyclonal antibodies; Zymed Laboratories), anti-DAF Fab (IIH6 mouse monoclonal antibody; a gift from V. Horejsi), or anti-fluorescein Fab (rabbit polyclonal antibodies; Invitrogen) was added to the colloidal gold suspension, and the mixture was incubated on a slowly tumbling shaker for 1 h at room temperature. The gold probe was further stabilized with 0.03% Carbowax 20 M (Sigma-Aldrich). After three washes by sedimentation and resuspension in 0.03% Carbowax 20 M/2 mM phosphate buffer, the gold particles were finally resuspended in HBSS containing 0.03% Carbowax 20 M and were used within 6 h. To further suppress the cross-linking by Fab-gold, a final concentration of 8.3 μg/ml of free Fab (which was not bound to gold particles) was premixed with the coated gold particle suspension, and the mixture was added to the cells cultured on coverslips. The ligand for CD59, C8, was added at a cytolytic membrane attack complex unit of 1,000/coverslip (at a serum concentration; ). For the cross-linking of CD59, PLAP, or DAF, a fivefold minimal protecting amount of anti-CD59 IgG, anti-PLAP IgG, or anti-DAF IgG (whole divalent IgG) was mixed with the gold particle suspension and subjected to the same stabilization and washing procedures as for Fab-gold. For the conjugation of Cy3 with Fab or IgG, a Cy3 monofunctional dye kit (GE Healthcare) was used according to the manufacturer's instructions. The dye/protein ratio was adjusted to ∼1 in all cases. All of the microscopic observations of living cells were performed at 37°C. For single-particle tracking experiments, cells were sparsely seeded on coverslips (4 × 10 cells/coverslip) and were grown for 18–30 h before each experiment. They were used at ∼10% confluency. Care was taken to use cells at about the same level of confluency (∼10%) for every experiment because the macroscopic diffusion rate in the T24 cell membrane was found to depend on the confluency level (the macroscopic diffusion coefficients for CD59 and DOPE were twofold higher in the confluent cell membranes; note that they were practically the same if the confluency levels were the same). The incorporation of Cy3-DOPE or fl-DOPE into cell membranes was done as previously reported (). IgG-gold particles were added to the cells at a final concentration of 1.8 × 10 particles/ml. This concentration was selected so that it is sufficiently high to induce robust intracellular signaling responses, like those after the addition of high concentrations of C8, and yet sufficiently low not to bother single-particle tracking by the close encounters of two particles on the cell surface. Single-particle tracking of gold probes was performed on the apical/dorsal membrane, using a microscope (E800; Nikon) equipped with a 100× 1.4 NA objective lens and a CCD camera (XC-75; Sony), as described previously (; ). Single fluorescent molecule tracking was performed using a home-built, objective lens–type, TIRF microscope, based on the same microscope (; ). The accuracies of the position determinations for stationary probes were estimated from the SDs of the determined coordinates of the probes fixed on poly--lysine–coated coverslips impregnated in 10% polyacrylamide gel. The accuracies for the gold and Cy3 probes at 33-ms resolution were ±2 and ±17 nm (SD), respectively. The term STALL was used instead of TCZ to avoid confusion, because we were unable to rule out the possibility that the temporary immobilization of CD59 clusters is induced by their binding to the actin-based membrane skeleton, rather than by being confined within preexisting zones. However, the detection of STALL was performed by using the software to detect TCZs, developed by . STALL (TCZ) was detected in gold-probe trajectories recorded at a 33-ms resolution for a period of 10 s, as previously described (). The length of the trajectories used for the analysis had no influence on the estimated parameters, as long as it was >5 s. The size of the area covered by a CD59 cluster during STALL was estimated by 2D Gaussian fitting of the determined coordinates of the CD59 cluster during the STALL period. To observe the recruitment of Lyn or Gαi2 to CD59 clusters in live cells, T24 cells were transiently transfected with the cDNA for Lyn fused with GFP (at the C terminus of Lyn) or with that for Gαi2 fused with YFP (which was placed between the 91st and 92nd amino acids of Gαi2's; the cDNA for Gαi2, obtained from T. Haga, Gakushuin University, Tokyo, Japan, was modified according to ). Lyn-GFP has been known to be functional (). and showed that Gαi2(YFP) underwent proper membrane targeting and Gβγ association. Detection of single Lyn-GFP or Gαi2(YFP) molecules was confirmed by single-step photobleaching of each individual fluorescent spot, as well as by the single Gaussian distribution of the signal intensity of the fluorescent spots (). For the simultaneous tracking of single CD59 clusters and single molecules of Lyn-GFP or Gαi2(YFP), CD59 clusters were formed by using 50-nm latex beads coated with anti-CD59 whole IgG, because the 40-nm gold particles gave signals that could not be separated from the fluorescence signals from GFP or YFP, at the level of single molecules and single particles. These two types of particles exhibited practically the same STALL time fractions and durations. Furthermore, as shown in (left) in and the fourth trace from the bottom in (), these 50-nm beads were capable of inducing intracellular signals as effectively as 40-nm gold IgG-gold particles. Colocalization was determined as described by , i.e., when the distance between the center of the bead image and that of the GFP-tagged signaling molecule was <100 nm, they were regarded as being colocalized. This is a condition where ∼98% of bound molecules are classified as colocalized molecules when the binding did occur, under the present signal-to-noise ratio of each image and the present accuracy of superimposing images obtained on two cameras placed on two separate detection arms. The frequency of incidental colocalization events was evaluated by artificially shifting the video sequence for CD59 clusters by 50 frames (1.67 s, which could be an arbitrary period if it is much longer than the median colocalization period) and was subtracted. The distributions of the time difference (lag time) between the colocalization and the onset of STALL () and the distribution of the colocalization period were obtained. The estimation of colocalization period was not affected by the photobleaching of GFP or YFP because the recruitment durations were substantially shorter than the GFP lifetime before photobleaching and blinking (∼850 ms, which was measured for GFP conjugated to stem cell factor receptor, which stays in the plasma membrane). Furthermore, when the laser intensity was increased by a factor of two, the residency times of Lyn-GFP or Gαi2(YFP) at CD59 clusters were not significantly affected. Cells, cultured in a 6-cm dish, were incubated with Fab-gold (for nonstimulated CD59) or IgG-gold (for cross-linking CD59) at 37°C for 1, 2, 5, 10, 30, and 50 min; washed twice with PBS; and lysed with 1% NP-40 + 0.1% SDS lysis buffer containing phosphatase inhibitors and protease inhibitors at room temperature. Lyn was immunoprecipitated from the lysate with anti-Lyn antibodies (Santa Cruz Biotechnology, Inc.) and was blotted with anti–pY418-Src antibodies (Biosource International) and anti-Lyn antibodies. For quantitation of the phosphorylation levels of Lyn and other SFKs, the blot was imaged by a CCD camera and the intensity was quantitated by the NIH ImageJ software. Fig. S1 shows the difference between the membrane compartments (corrals) and STALL and oligomerization-induced trapping of membrane molecules within membrane compartments. Fig. S2 shows the diffusion coefficients for the spots containing five anti-CD59 IgG molecules, showing that they are practically immobile. Fig. S3 gives the definition and the detection protocol for STALL. Fig. S4 demonstrates detection of membrane compartments and STALL in the trajectories observed by high-speed single-particle tracking. Fig. S5 shows that STALL sites are not caveolae. Video 1 provides representative movement of a CD59 cluster (IgG-gold particle) recorded at video rate (replayed in real time), showing alternating periods of apparently simple Brownian diffusion and STALL. The original data for the trajectory shown on the right in a. Video 2 shows single-molecule detection of transient Lyn-GFP recruitment to a CD59 cluster, occurring without temporal correlation with the STALL of the CD59 cluster. Video 3 shows single-molecule detection of transient Gαi2(YFP) recruitment to a CD59 cluster, which occurs right before the onset of the STALL of the CD59 cluster. Video 4 shows single-molecule tracking of Gαi2(YFP), showing that Gαi2(YFP) is recruited from the cytoplasm to the plasma membrane only transiently, i.e., its residency time on the plasma membrane is generally on the order of 0.1 s or less. Online supplemental material is available at .

In the companion paper (see Suzuki et al. on p. of this issue), we report that single individual Gαi2 and Lyn molecules are dynamically and frequently recruited to CD59 clusters (consisting of three to nine molecules) formed beneath a colloidal gold particle 40 nm in diameter, coated with whole IgG antibody (IgG-gold), as determined by single-molecule tracking. These results are consistent with previous reports showing that clustered glycosylphosphatidylinositol-anchored receptors (GPI-ARs) recruit and activate Gαi and Lyn (and other Src-family kinases [SFKs]; ; ; ). Furthermore, we found that right after the recruitment of Gαi2, the CD59 cluster temporarily stops diffusion, which is an SFK (e.g., Lyn) activity-dependent process termed stimulation-induced temporary arrest of lateral diffusion (STALL). Therefore, we proposed that, when a single Gαi2 molecule is recruited at the CD59 cluster, the recruited Gαi2 molecule would bind to and activate Lyn that was also recruited temporarily to the same CD59 cluster, based on the previous observations in which Gαi2 binds to SFKs and activates them without the need for dephosphorylating the tyrosine residue near the C terminus (; ; ). We also proposed that Gαi2-activated Lyn induces STALL of the CD59 cluster, probably by phosphorylating an as-yet-unknown protein. In the present paper, we concentrate on the physiological functions of the STALL, rather than the mechanism for inducing the STALL of CD59 clusters. The involvement of raft domains in recruiting signaling molecules (; ; ; ) is collectively discussed toward the end of the Results in this paper. In another line of earlier studies of GPI-AR signal transduction, the cross-linking of GPI-ARs, e.g., decay accelerating factor (DAF, or CD55) and CD59, was found to trigger the activation of the intracellular inositol- triphosphate (IP)–Ca pathway. This is a nonlethal signaling event found in both immune and nonimmune cells (; for review see ), and it involves the hydrolytic generation of IP from phosphatidylinositol-bisphosphate (PIP) by PLCγ (; ; ), leading to the release of Ca from the stock in the ER through the IP receptor (IP-dependent calcium channel; ; ; ; ). Therefore, the next interesting point may be the relationship between the Gαi2–Lyn and IP–Ca signaling pathways. Previously, showed that the inhibition of SFKs blocked the GPI-AR clustering–induced Ca mobilization. reported that SFK-induced IP–Ca signaling may be mediated by PLCγ (but not by PLCβ). These results suggest that IgG-gold–induced Lyn activation might occur upstream of IP production from PIP, by PLCγ in the signaling cascade. Meanwhile, we showed in that Gαi recruitment (and thus Lyn activation) quickly induces STALL. This led us to form the following working hypothesis. Namely, IP production from PIP may take place exclusively at the CD59 cluster undergoing STALL by recruiting cytoplasmic PLCγ there; therefore, the CD59 cluster undergoing STALL may be a key, albeit temporary (0.57-s lifetime), site for linking the Gαi2-induced Lyn activation to the PLCγ–IP–Ca signaling pathway. We performed the present research based on this working hypothesis. We examined it by carrying out simultaneous observations of single molecules of GFP-conjugated PLCγ2 (GFP-PLCγ2) and single CD59 clusters. We indeed found that single PLCγ2 molecules are recruited to CD59 clusters, almost exclusively during the STALL periods. Furthermore, the occurrence of STALL is a prerequisite to the bulk increases of the cytoplasmic IP and Ca concentrations (). Because GPI-ARs have a strong tendency to partition into the low buoyant density fractions on the sucrose density gradient ultracentrifugation, after extraction with Triton X-100 at low temperatures (called detergent-resistant membrane fractions), the so-called raft microdomains may be involved in the signal transduction of GPI-ARs (; ; ; ; ; ; ). Furthermore, the partitioning of PLCγ into raft microdomains has been suggested (; ). We now raise the possibility that the recruitment of PLCγ at CD59 clusters is facilitated by the affinities of both molecules to the so-called raft microdomains (; ; ). The involvement of cholesterol-based special membrane domains or lipid–lipid interactions in the signal transduction of GPI-ARs is collectively discussed at the end of the Results, together with the data presented in . Other critical issues addressed in this paper are the dynamic nature of the recruitment and the very short residency times of the intracellular signaling molecules (PLCγ2, Gαi2, and Lyn) at the CD59 cluster, which are much shorter than the bulk activation durations of these molecules (). As described in , the residency times of Gαi2 and Lyn (as well as PLCγ2, as shown here) are only 0.1–0.3 s, which was at first surprising. A quantitative understanding of the signaling mechanisms would become possible once the kinetics and the equilibrium of the molecular interactions are known; thus, this issue is fundamentally important. Nevertheless, the lifetimes of the molecular complexes occurring in any signaling pathway are rarely known (). The single fluorescent molecule tracking used here is particularly suited to tackle this problem. By investigating the recruitment of cytoplasmic PLCγ2 molecules to GPI-AR clusters at the level of single molecules, we were able to track the complicated single-molecule dynamics. In fact, without using single-molecule tracking with sufficient spatiotemporal resolution, transient binding in living cells could not have been found. The lifetimes of the molecular complexes, and the residency time of PLCγ2 at the CD59 clusters, were determined. CD59 is widely expressed in nonimmune cells, as well as in immune cells, and is responsible for triggering nonlethal signaling in both cell types, in addition to controlling complement-induced cell death (for review see ). The human epithelial T24 cell line, which was extensively used here, expresses CD59, Gαi2, Lyn, and PLCγ2 and was expected to undergo nonlethal signaling responses upon the binding of CD59's natural ligand, C8, which leads to the formation of CD59 clusters (), or upon the artificial clustering of CD59, as found in other cell types (; ; , ). Therefore, we first established the occurrence of IP–Ca signaling responses in T24 cells (– ), and particularly their amplitude and time courses under the very special conditions of CD59 engagement, which we also used for observing SFK activation (see and in ): the binding of ∼600 IgG-gold particles, each cross-linking approximately six CD59 molecules on average and inducing the total engagement of ∼3,600 CD59 molecules (). We paid special attention to the signal amplitude and on the time courses of the IP–Ca signal (see ). First, we extensively examined GPI-AR cluster–induced IP signaling, using live-cell imaging. Although the increase in the cytoplasmic IP concentration upon GPI-AR cross-linking has been described in the literature cited above (; ; ), these studies were not conducted in living cells and thus remained somewhat cursory; in addition, we had to know if and how the cross-linking by 600 IgG-gold particles elicited IP responses in T24 cells, a cell line that was extensively used here. The increase in the cytoplasmic IP concentration was monitored by observing the partitioning between the cytoplasm (IP) and the plasma membrane (PIP) of the pleckstrin homology (PH) domain of PLCδ1 fused to GFP (; ; ), as parameterized by the relative cytoplasmic and plasma membrane fluorescence signal intensities (I/I) of the PH domain–GFP (; ; see Fig. S1, available at , for the result using confocal microscopy). Typical time courses of IgG-gold–induced Ca mobilization observed with Ca-sensitive fluorescent dyes (see Materials and methods; ) are shown in . Both the levels and time courses of the intracellular IP and Ca responses to ∼600 bound IgG-gold particles were comparable to those observed after the addition of CD59's natural ligand, C8, at a serum (therefore high) concentration (a cytolytic membrane attack complex unit of 1,000/coverslip; ), although the Ca signal started somewhat earlier with IgG-gold (). The involvement of PLCγ was confirmed by the blocking of the IgG-gold–induced IP and Ca responses by the pretreatment of cells with the PLCγ inhibitor U73122, but not by the control drug U73343 (– ). In addition, the IgG-gold–induced Ca mobilization was inhibited by microinjecting the cells with heparin (), indicating that the rise in the cytoplasmic Ca concentration is due to the release of Ca from ER via the IP receptor (). These signaling responses did not occur with the control gold particles (Fab-gold; prepared by conjugating small numbers of Fab fragments of the anti-CD59 antibody to gold particles), which bind to CD59 but do not cross-link it. Our Fab-gold probes were probably bound to single molecules of CD59, as described (see and and related text in ). These results, in addition to the CD59-induced SFK activation shown in , indicate that CD59 acts as a receptor for nonlethal signals in T24 cells, like that in immune cells. Individual CD59 clusters induced by IgG-gold particles exhibited alternating periods of apparently simple Brownian diffusion and temporary immobilization, called STALL, as shown in , as well as in in . The lifetimes (the exponential decay times) of each STALL and simple Brownian period are 0.57 and 1.2 s, respectively (see in ). Partial actin depolymerization greatly reduced the STALL time fraction of IgG-gold–induced CD59 clusters (). Similar observations were made using cross-linking gold particles for other GPI-ARs, placental alkaline phosphatase (PLAP), and DAF and in other cell types (NRK and PtK2 cells; see in ). Closer inspection of the trajectories during the STALL periods () revealed that the CD59 clusters undergo a jittering motion (without macroscopic diffusion), in a limited area of 48 nm in diameter, as estimated by the Gaussian fitting of the determined coordinates during the STALL period (), whereas the median compartment size of the partitioned plasma membrane is 110 nm in the cell type used in this study (T24; ; , bottom). Practically the same STALL diameter was found in NRK cells (52 nm; ), which have a much greater compartment size (230 nm; ; ; , bottom). This result suggests that STALL is not induced by the corralling effect of the actin-based membrane skeleton that is enhanced by the clustering of CD59 (). STALL may be induced by the binding of CD59 clusters to actin filaments (but indirectly, because they are located in/on the opposite leaflets of the membrane) or by the partitioning of the CD59 clusters into actin-supported microdomains. The jittering motion during STALL may be due to the conformational dynamics of the actin filaments. All of these results are consistent with previous observations suggesting a link between raft domains and actin filaments (; ; ). Here, we summarize six results showing very high correlations between IP–Ca signaling and STALL of CD59 clusters (). Namely, any treatment that blocked STALL also inhibited IP generation and Ca mobilization. We will then show that STALL takes place upstream of IP production by PLC, indicating the STALL requirement for IP–Ca signaling. Treating cells with methyl-β-cyclodextrin (MβCD), which may cause partial depletion of cholesterol, greatly reduced the STALL fraction of CD59 clusters () and, at the same time, inhibited the IgG-gold–induced production of IP ( and and ) and the intracellular Ca responses ( and ). Replenishment of cholesterol with cholesterol-loaded MβCD, after partial depletion of cholesterol with MβCD, restored IP–Ca signaling and reinstated the STALL of CD59 clusters (– and ). MβCD may have side effects other than cholesterol depletion, as described later, but our point here is the high correlation between IP–Ca signaling and STALL induction under various conditions. Partial actin depolymerization with latrunculin B, blocking the activities of SFKs with PP2, or blocking Gα's with pertussis toxin (PTX) greatly reduced the STALL of CD59 clusters and, at the same time, blocked the IgG-gold–induced IP–Ca signaling (– and ). Fab-gold–CD59, mycCD59TM clusters, and 1,2-dioleoyl--glycero-3-phosphoethanolamine (DOPE) clusters induced neither STALL nor IP–Ca signaling (). We next examined the effect of inhibiting PLCγ on the STALL of IgG-gold–induced CD59 clusters. The PLCγ blocker U73122, under the conditions where it completely blocked IP production (, right), did not affect STALL at all (whereas the control reagent U73343 had no effect on either the IP concentration or STALL; ). This result indicates that STALL occurs upstream of PLCγ-induced IP production in the signaling cascade after CD59 engagement and that STALL is required for IP–Ca signaling. Given the STALL requirement for IP–Ca signaling, we formed the following working hypothesis: the STALL sites may be the places where IP is produced from PIP. Because this reaction is likely catalyzed by the signaling enzyme PLCγ, we investigated whether PLCγ2 is specifically recruited to the STALL sites, i.e., whether the PLCγ2 recruitment to CD59 clusters takes place exclusively when CD59 clusters are undergoing STALL. Thus, we simultaneously observed CD59 clusters and single molecules of GFP-PLCγ2 () in or on the plasma membrane. (PLCγ1 is also expressed in T24 cells used throughout this study, and we expect that PLCγ1 would behave in a way very similar to PLCγ2 because these two molecules behave similarly in a variety of signal transduction pathways [].) For such observations, 50-nm latex particles, rather than 40-nm colloidal gold particles, were used to form CD59 clusters, to avoid the signal from the gold particles (). Bright-field microscopy was used to observe the beads, and these two types of particles exhibited practically the same STALL time fractions and durations. In addition, these 50-nm beads were signal capable as much as 40-nm colloidal gold particles, as shown in and . shows a typical trajectory of a CD59 cluster and a GFP-PLCγ2 molecule (also shown in ), including a period of colocalization (Video 1, available at ). Only during a STALL period of the CD59 cluster was a GFP-PLCγ2 molecule recruited from the cytoplasm for a period of ∼0.3 s (; colocalization determined as described by ). Note that although the GFP-PLCγ2 molecules must be continuously colliding with the cytoplasmic surface of the plasma membrane, the vast majority of them would not stay longer than a 33-ms video frame time, which would make them invisible in video-rate observations. Those recruited at the CD59 clusters undergoing STALL tend to stay for a period of several video frames, thus making themselves visible. Based on many such experiments, we obtained a distribution (histogram) of the time difference (lag time) between the onset of STALL of a CD59 cluster and the recruitment of a single GFP-PLCγ2 molecule to that particular CD59 cluster (; see in for the definition of the time difference). Time 0 was set at the initiation of the STALL period, and the incidental colocalization was subtracted. This histogram clearly indicates that GFP-PLCγ2 molecules are recruited to CD59 clusters right after the onset of STALL. The yellow bars represent the GFP-PLCγ2 molecules that left the STALLed CD59 clusters before they resumed normal diffusion (i.e., those with recruitment periods completely included within a STALL period). Therefore, this histogram shows that GFP-PLCγ2 recruitment occurs almost exclusively during the STALL periods. Note that, in all of the cases we have observed thus far (see, e.g., Video 1), a CD59 cluster starts showing STALL first, and the recruitment of GFP-PLCγ2 molecules to the CD59 cluster takes place during the STALL period (and not vice versa), consistent with our bulk imaging and biochemical observations indicating that STALL is required for IP–Ca signaling via PLCγ. These results strongly suggest that the CD59 cluster undergoing STALL is the key, but temporary, platform for the transient recruitment of PLCγ2 molecules and IP production and thus for transducing the extracellular CD59 signal to the intracellular signaling network via PLCγ2 recruitment, which eventually leads to the intracellular Ca response. The duration of GFP-PLCγ2 recruitment was 0.25 s (median, corresponding to 7.5 video frames; ), which is much shorter than the STALL lifetime (0.57 s; GFP photobleaching/blinking was not responsible for this short duration, as described in ). As the turnover rates of PLC to produce IP from PIP found in the literature are between 70 and 200 per second (; ), assuming a sufficient supply of PIP in or around the CD59 cluster undergoing STALL, the recruited PLCγ2 can produce 20–50 molecules of IP during the recruitment period of 0.25 s. The biological significance of such transient recruitment is further discussed in the Discussion section (see ). How PLCγ recruited to CD59 clusters during the STALL periods is activated remains unknown. Because Lyn is very often (almost constitutively, but dynamically) recruited to the CD59 clusters, Lyn is a good candidate kinase for phosphorylating PLCγ, triggering its activation (). The recruitment of PLCγ2 at CD59 clusters was further confirmed by immunofluorescence microscopy of IgG-gold–treated cells (5 min and then fixed; ). Approximately 20% of the CD59 clusters (IgG-gold particles) were colocalized with PLCγ2, and this association was greatly diminished by cholesterol depletion with MβCD, blocking of SFK activity with PP2, or partial actin depolymerization with latrunculin B (), which totally correlated with the STALL occurrence of CD59 clusters () and IP–Ca signaling (– and ). Because the temporal STALL fraction of CD59 clusters (IgG-gold particles) is ∼40%, 20% PLCγ2 immunocolocalization is considered to be consistent with the PLCγ2 recruitment only during the STALL period, considering the binding and visualization efficiency in immunofluorescence staining. Fluorescein-DOPE, a typical nonraft phospholipid preincorporated in the plasma membrane, was not concentrated under the IgG-gold particles (see in ), indicating that membrane concentration and membrane undulations, as reported by , are not involved in the colocalization of the PLCγ2 spots found here. When larger CD59 clusters are formed by the sequential addition of the primary and secondary antibodies, they are almost always strongly immobilized (see ) and are capable of activating Lyn and triggering the IP–Ca pathway. We suggest that the immobilization and signaling mechanisms for large CD59 clusters may be the same as those for IgG-gold–induced CD59 clusters. Because many transient bindings simultaneously occur in large CD59 clusters, the clusters are almost always bound to the actin skeleton by at least one remaining bond, thus keeping the cluster immobilized for long periods. Large CD59 clusters may be strongly signal competent, because they recruit many Gαi2, Lyn, and PLCγ2 molecules at the same time, although each recruitment event lasts for periods as short as those found at IgG-gold–induced clusters. In the present study, the elementary recruitment and signaling steps were dissected by using IgG-gold particles to induce small CD59 clusters and by single-molecule tracking. Furthermore, triggering with IgG-gold resembles the physiological stimulus much more than that with large CD59 clusters: after C8 addition, the Fab-gold particles diffused and underwent STALL, as the IgG-gold particles did (). STALL might have further physiological importance in cellular signaling. The external signal may be relayed to the actin membrane skeleton/cytoskeleton at the STALL sites, perhaps inducing local reorganization of the actin filament network (; ). In this subsection, we address the possible involvement of raft microdomains in the CD59 signal transduction by combining the results described here with those in . For this examination, we used four approaches in this research. First, an artificial transmembrane construct of CD59 (mycCD59TM; CD59 with a myc tag at its N terminus and the transmembrane domain plus the following 12 amino acids of the low-density lipoprotein [LDL] receptor attached at its C terminus) was used to examine the effect of replacing the natural GPI anchor with a transmembrane domain (). MycCD59TM clusters, formed by anti–myc-IgG-gold, did not exhibit STALL () and induced neither Lyn activation (see and in ; ) nor IP–Ca signals (, left; ; and ). Second, the effect of cross-linking another typical nonraft molecule, the unsaturated phospholipid DOPE, was examined. DOPE clusters induced by IgG-gold did not exhibit STALL () and induced neither Lyn activation (see and in ) nor Ca mobilization ( and ). The Ca mobilization data may appear to be at variance with the results by . Although we cannot resolve this apparent difference, the cross-linked molecules are different, the levels of clustering used here would be much lower than those used by , and they used Jurkat T cells, which may be very sensitive to various external stimulations. Third, the recruitment frequency of LynN20-GFP (the N-terminal 20-amino-acid sequence of Lyn, containing the binding sites for a palmitoyl and a myristoyl chains, fused at its C terminus to GFP; ) to CD59 clusters was substantially greater than that for Lyn-GFP to non–cross-linked CD59 or to cross-linked mycCD59TM (negative controls, 3.3-fold). Meanwhile, it was considerably smaller than that of Lyn-GFP to CD59 clusters (positive control, 2.4-fold; see Table IV in ). These results suggest that both lipid–lipid interactions via Lyn's alkyl chains and protein–protein interactions by way of Lyn's protein part contributed to the recruitment of Lyn to CD59 clusters, consistent with the findings of . Fourth, the effects of partial cholesterol depletion (with MβCD or saponin) and the subsequent replenishment of cholesterol were examined. warned against the use of MβCD because it enhances the PIP-based F-actin network formation, which could not be recovered by cholesterol replenishment within 12 h. This suggests that if the observed effect of MβCD could be quickly recovered by replenishing cholesterol (generally within 30 min), then the primary effect of MβCD found in that particular experiment probably reflects the direct influence of partial cholesterol depletion. Cholesterol depletion with MβCD (4 mM for 30 min at 37°C) decreased the temporal STALL fraction of CD59 clusters by a factor of approximately five () and Lyn activation by a factor of approximately four (see and in ) while virtually blocking the IP–Ca signaling (– and ). The subsequent cholesterol replenishment restored STALL and reinstated Lyn phosphorylation in the activation loop (see and in ) as well as IP–Ca signaling (– and ; although the initiation of the Ca signal is slower []). Cholesterol clustering with saponin also reduced STALL (see in ). Similar observations were made for other GPI-ARs (PLAP and DAF) in other cell types (NRK and PtK2 cells; see in ). Although the cholesterol replenishment experiments always worked well, drug treatments are always susceptible to side effects and therefore due caution should be paid to the interpretation of the results using MβCD. In addition, single-molecule tracking revealed transient, dynamic recruitment of glycosphingolipid GM1 to CD59 clusters (not depicted, but such transient recruitment is consistent with the lack of colocalization in fixed cells as reported by ), further suggesting the involvement of raft microdomains in the signal transduction of GPI-ARs. These four lines of results suggest that cholesterol-based raft microdomains or lipid–lipid interactions enhance the recruitment of the cytoplasmic signaling molecules to CD59 clusters. Because the specificity of signaling pathways without any involvement of protein–protein interactions is unlikely (; ; ), one of the most critical issues regarding the involvement of raft microdomains in signal transduction in the plasma membrane is whether raft microdomains facilitate or enhance the rate and/or the specificity of molecular recruitment. The aforementioned results, in particular those using LynN20-GFP and mycCD59TM, suggest the occurrence of raft-based facilitation in recruiting Lyn to the CD59 clusters. Based on the results described in this paper and in , we propose the following working model for the signal transduction of GPI-AR, shown in (also see Video 2, available at ). We also point out the places in this model where protein interactions may be facilitated by cholesterol-containing (raft) microdomains. There are many unknowns in this working model, but we believe that the present work clarified some of the important processes in the signal transduction steps of GPI-ARs and made it possible to delineate the next investigation steps, as described in this working model. The signaling steps shown in are described in the following paragraphs. When CD59 is liganded, it forms larger clusters that resemble those formed by IgG-gold particles (; , step 1). Because the subsequent Lyn activation and IP production did not occur after cross-linking mycCD59TM or DOPE, and because they are sensitive to cholesterol depletion as well as the subsequent cholesterol replenishment, we raise the possibility that the CD59 cluster may also be a cholesterol-enriched nanodomain (). In the second step, these CD59 clusters undergo slow, apparently simple Brownian diffusion (, step 2a) while frequently and transiently recruiting Lyn, for a median of ∼0.20 s (, step 2b; see in ). The mechanism for recruiting Lyn is unknown (, step 3). Lipid-based interactions between Lyn and the CD59 cluster raft nanodomains across the bilayer might be involved. Another possibility would be the involvement of an as-yet- unknown transmembrane protein, X, that has certain levels of affinity to the protein moiety of Lyn (and perhaps to Gαi2) and to that of CD59 (avidity to CD59 clusters), as well as an affinity for the raft nanodomains (or to the lipid and lipid- anchoring sequences of Lyn, Gαi2, and CD59; see Table IV in ). The cholesterol requirements for IP–Ca signaling reported previously (; ) might occur at this stage. We also found that Gαi2 molecules are transiently recruited to CD59 clusters for ∼0.13 s (median) and that right after the recruitment, the CD59 cluster is engaged in STALL (, step 4; see , d and e, in ). These results suggest that when the Gαi2 molecule meets Lyn molecules that have also been recruited at the same CD59 cluster, they form a complex at the CD59 cluster, leading to Lyn activation () and inducing further Lyn activation by phosphorylating its activation loop, perhaps by autophosphorylation (). As the present work showed that the CD59 cluster acts as a platform for recruiting Gαi2 and Lyn, the next important step is to show their interactions at a single CD59 cluster at the single-molecule level and, at the same time, to detect the activation of Lyn at the level of each individual single molecule right before the onset of STALL. Both Gαi2 and Lyn soon leave the CD59 cluster, and what these activated proteins do after they leave the CD59 cluster is unknown and would be of interest. This is similar to the activation process of H-Ras, reported previously (; for review see ). H-Ras may be activated in raft microdomains, but the activated H-Ras leaves them to become incorporated in the domain associated with galectin 1. In step 5, Lyn activation concurrently induced both STALL of the CD59 cluster and transient PLCγ2 recruitment at the CD59 cluster (median of ∼0.25 s; ). Understanding the mechanisms of these processes would be an important next step. Because PLCγ2 will produce IP from PIP only at the CD59 cluster that is undergoing STALL (∼0.57 s), the CD59 cluster in STALL is likely to be the key, but transient, platform for relaying the extracellular CD59 signal to the intracellular IP–Ca signals (, step 6). The residency time of PLCγ2 at the STALLed CD59 cluster is short (∼0.25 s; ), but it may be sufficient to produce 20–50 IP molecules, a calculation based on the turnover rate of the enzyme (; ). The direct observation of IP production right at the single CD59 cluster undergoing STALL should be performed in future studies. In step 7, the increase in the IP concentration will lead to the release of Ca from the intracellular pool (). When the present research was undertaken, cell biological and biochemical data accumulated in the literature suggested that (a) the engagement of GPI-ARs may trigger the association of Gαi2 (), SFKs (; ), and PLCγ2 (; ) with the GPI-AR cluster; (b) Gαi2 may bind to Lyn to activate it (); and (c) Lyn might induce the IP–Ca pathway via PLCγ recruitment (). However, the majority of these proposals were based on pharmacological experiments and pull-down assays in the presence of detergents (which jeopardize the legitimacy of the detection of molecular complexes in these assays). Furthermore, these results did not address how these events may occur in a spatiotemporally organized way in living cells (). The single-molecule tracking results obtained in the present study contributed greatly to clarifying this process, by providing a dynamic view of these events as itemized above, although there are still many unknown processes and mechanisms. The cytoplasmic IP signal, as observed by conventional fluorescence imaging, lasts for at least 15 min, or on the order of 1,000 s ( and ). Such a bulk signal must be generated by the superposition of the individual events of PLCγ2 recruitment to the plasma membrane, which occurs at the CD59 clusters undergoing STALL, producing 20–50 IP molecules for each PLCγ2 recruitment period. The cytoplasmic IP signal amplitude (concentration) at a given time may largely be determined by the collective recruitment/activity of thousands of PLCγ2 copies (consumption and long-term accumulation of IP might occur). Therefore, the short recruitment period of ∼0.25 s for each single PLCγ2 molecule at the CD59 cluster in STALL, which is at least 4,000-fold shorter than the overall duration of the bulk IP signal, might initially seem surprising. However, such pulse-like occurrences of individual signaling events might address a long-standing problem in the field of cellular signaling, as raised from the viewpoint of communication system engineering () or systems biology (). Here is the enigma. The recruitment/activation of each individual molecule has been believed to last for a duration comparable to (or perhaps 10 times less than that of) the bulk duration. For the sake of simplicity, assume the overall activation duration to be 1,000 s and the durations for recruitment/activation of each individual molecule to be on the order of 100–1,000 s. Generally, as stated at the beginning of this section, the activation level of the bulk signal is determined by the superposition (integration) of the recruitment/activation of thousands of copies of the signaling molecule. However, this integration presents a difficult problem for the cell, which must regulate and maintain the proper, stable level of the overall activation of the signaling event (see Model A [complex integration model] in ). This is because the active durations of individual molecules are assumed to be long, on the order of 100–1,000 s for the overall activation period of 1,000 s: to reach the correct level of bulk activation, the activation of the next copy of the molecule must be done on the basis of the correct prediction regarding when the previously activated copies of the molecule are turned off. How the cell can achieve such a complicated integration over many thousands of molecules has remained enigmatic. In the case of IP signaling examined here, each individual elementary process for IP production takes place like a quantized burst (∼0.25 s), which is shorter by a factor of at least 4,000 than the bulk activation period. If this were the case, for realizing the proper, stable level of the bulk IP concentration, complicated integration would be unnecessary: the simple addition of the IP pulses would make the bulk signal, as shown in Model B (simple summation model) in (for brevity, the signal amplitude by each individual event is assumed to be the same here; ). Therefore, the short lifetime of each recruitment event of PLCγ2 at a CD59 cluster, and thus the briefness of each burst of IP production, may be critical for the regulation of the enhanced level of IP upon the engagement of CD59. The analogue amplitude of the bulk IP signal is basically proportional to the number of PLCγ2 recruitment events during the unit time (and thus the number of IP bursts per unit time). If the rate of IP consumption is slow, the bulk amplitude is proportional to the cumulative number of PLCγ2 recruitment until the specified time; in this sense, to obtain the bulk amplitude, the loss of IP has to be evaluated. Nevertheless, the digital, pulse-like signal produced by each IP production event greatly simplifies the signaling system. Partial depletion of cholesterol in the plasma membrane was performed by incubating the cells in 4 mM MβCD (Sigma-Aldrich) at 37°C for 30 min () or in 60 μg/ml saponin (Sigma-Aldrich) on ice for 15 min (). These treatments substantially increased the amount of CD59 recovered in the detergent-soluble fractions in the protocol to prepare detergent-resistant membranes. Replenishment of cholesterol was performed by incubating the cholesterol-depleted cells in 10 mM MβCD–cholesterol complex (1:1) for 30 min at 37°C (). The overall amounts of cholesterol per cell after cholesterol depletion with MβCD and after the subsequent repletion were found to be 66 and 118% of the original amount (SD of ±6%), as determined by a cholesterol quantification kit (Wako). Partial actin depolymerization was performed by incubating the cells in the medium containing 50 nM latrunculin B for 10 min (gifts from G. Marriott, University of Wisconsin–Madison, Madison, WI; ). SFKs were inhibited by treating the cells with 10 μM PP2 (Calbiochem) for 5 min at 37°C (). Heterotrimeric G protein was inhibited by incubating the cells in medium containing 1.7 nM PTX (Calbiochem) at 37°C for 22 h (). The involvement of the IP receptor in the GPI-AR–induced Ca mobilization was examined by injecting 10 mg/ml heparin (∼100 femtoliter; Sigma-Aldrich) into the cells (), using a micromanipulator/injector (Eppendorf). As a control for GPI anchoring, a transmembrane chimeric protein of CD59 was used (mycCD59TM; its cDNA was provided by M. Maio, Instituto Nazionale di Ricovero e Cura a Carattere Scientifico, Ancona, Italy): the CD59 ectodomain was fused with an N-terminal myc tag and a C-terminal LDL receptor transmembrane domain, which additionally contains the 12 amino acids from the N terminus of the cytoplasmic domain of the LDL receptor (and thus lacks the sequence required for internalization via coated pits; ). IgG-gold particles coated with anti-myc antibody (9E10.2) were used for cross-linking mycCD59TM. Immunofluorescence staining of IgG-gold particles, and their colocalization with immunofluorescent spots of PLCγ2, was examined in the following way. Cells were incubated with IgG-gold for 5 min at 37°C, fixed with 4% paraformaldehyde for 90 min at room temperature, and permeabilized with 0.01% Triton X-100 in PBS for 1 min. After blocking with 5% skim milk for 90 min, the cells were immunostained with the rabbit anti-PLCγ2 antibodies (BD Biosciences). Fluorescein-conjugated goat anti–mouse IgG antibodies were used for the staining of the IgG-gold particles. These are described in detail in . STALL (transient confinement zone) was detected in gold-probe trajectories recorded at a 33-ms resolution for a period of 10 s, following the method developed by . The only difference is the definition of the size of the area covered by a CD59 cluster during STALL. It was estimated by the 2D Gaussian fitting of the determined coordinates of the CD59 cluster during the STALL period. To observe the recruitment of PLCγ2 to CD59 clusters in live cells, T24 cells were transiently transfected with the cDNA for PLCγ2 fused with GFP (at the N terminus of PLCγ2; obtained from M. Katan [The Institute of Cancer Research, London, England] and slightly modified; ). For the simultaneous tracking of single CD59 clusters and single molecules of PLCγ2, CD59 clusters were formed by using 50-nm latex beads coated with anti-CD59 whole IgG (a gift from V. Horejsi, Academy of Sciences of the Czech Republic, Prague, Czech Republic), because the 40-nm gold particles gave signals that could not be separated from the fluorescence signals from GFP, at the level of single molecules and single particles. These two types of particles exhibited practically the same STALL time fractions and durations. Furthermore, as shown in the left box in and the fourth trace from the bottom in , these 50-nm beads are capable of inducing intracellular signals as effectively as 40-nm IgG-gold particles. The bright-field images of the 50-nm latex beads (forming CD59 clusters beneath them) were obtained simultaneously with the images of GFP-tagged single signaling molecules, using the same conditions and instrument described in . Determination of colocalization is also described in . The increase in the cytoplasmic IP concentration after the addition of IgG-gold particles was observed in cultured T24 cells transfected with the cDNA for the PH domain of PLCδ1 fused with GFP at the C terminus (a gift from K. Hirose, Nagoya University Medical School, Nagoya, Japan; ; ). This PH domain binds to both IP, which is located in the cytoplasm, and PIP, which is located on the inner leaflet of the plasma membrane. Upon the engagement of CD59, PLCγ starts hydrolyzing PIP in the membrane to generate IP, and thus the cytoplasmic IP concentration increases (). This can be detected as the relative increase in the fluorescence signal of PH domain–GFP in the cytoplasm (I) versus that in the plasma membrane (I; ). This change in the fluorescence signal, detected by epifluorescence microscopy, was monitored by measuring the I/I ratio, as described by . The increase in the cytoplasmic IP concentration was also observed by confocal fluorescence microscopy, using a microscope (TE300; Nikon) equipped with a spinning-disc confocal scanner system (CSU22; Yokogawa) and a cooled charge-coupled device camera (Cascade 650; Roper Scientific). Intracellular Ca mobilization in living cells was monitored by epifluorescence microscopy, using fluo-4 or rhod-2 (Invitrogen) as a probe (). T24 cells were incubated in HBSS containing 5 μM fluo-4 AM (Invitrogen) for 30 min at room temperature, washed once with fresh HBSS, and incubated with the IgG-gold or Fab-gold particles. When the effect of DOPE cross-linking on Ca mobilization was examined, fluorescein-conjugated DOPE was incorporated in the cells and then cross-linked by IgG-gold particles coated with anti-fluorescein antibodies. Because of the spectral overlap, fluo-4 could not be used to observe the effect of cross-linking (fluorescein-conjugated) DOPE. Therefore, rhod-2 AM (Invitrogen) was used (). Fig. S1 shows the increase of the cytoplasmic IP concentration induced by CD59 clusters, as observed by the confocal fluorescence microscopy of the PH domain (from PLCδ1)–GFP expressed in T24 cells. Video 1 shows simultaneous observation of a CD59 cluster and the recruitment of single molecules of GFP-PLCγ2, recorded at video rate. Video 2 provides a model for the signal transduction of CD59, leading to the intracellular IP–Ca signal. Online supplemental material is available at .

After germinal vesicle (GV) breakdown, the fully grown oocyte is transcriptionally silent (). After fertilization, chromatin remodeling has been proposed to provide a window of opportunity for transcription factors to bind the regulatory sequences of genes that must be activated for development to proceed (; ). Concomitantly, a transcriptionally repressed state would be necessary to prevent promiscuous gene expression as a result of a “general permissiveness” of the genome (for reviews see ; ). In the mouse, two phases of transcriptional activation lead to the transition from maternal to zygotic control of gene expression (). The major and most studied wave of activation is the second one, which begins at the late 2-cell stage. However, less is known about the first wave, which occurs in the pronuclei of the zygote and represents 40% of the transcriptional levels observed at the 2-cell stage (; ; ). Transcription intermediary factor (TIF) 1 α () was first identified as a transcriptional regulator of nuclear receptors and has been shown to interact with numerous proteins involved in chromatin structure (, ; ; ; ; ). TIF1α is one of four TIFs described in mammals that belong to the tripartite motif superfamily of proteins (, ; ; ). TIF1β () is required for the proper specification of the anteroposterior axis in the mouse (). Little is known about the biological function of TIF1α, and its expression pattern is only known at late stages of postimplantation development (). Here, we have characterized the role of TIF1α in early mouse embryogenesis. We show that TIF1α acts as a modulator of the transcriptional state of a particular set of genes during the first wave of genome activation and that ablation of TIF1α compromises development. We first analyzed the expression pattern of in oocytes and throughout preimplantation development by in situ hybridization and RT-PCR. was expressed from the GV stage oocyte to the blastocyst (). Initially, transcripts were present in all blastomeres, but as development progressed, transcripts became restricted to the inner cells of the embryo (). This became evident at the 16-cell stage, and when the blastocyst formed, expression was restricted to the inner cell mass (ICM). At the GV stage, TIF1α protein was detected in the oocyte cytoplasm (). Shortly after fertilization, TIF1α remained predominantly cytoplasmic, but it moved to both pronuclei at the mid and late zygote stages. TIF1α became localized to discrete regions associated with the nucleolar-like bodies (NLBs), which are a compact mass of DNA surrounded by a perinucleolar chromatin ring that cause the characteristic pattern of DNA staining visible at these stages (; ). This distribution was observed in both male and female pronuclei, although in some cases (11 of 32 zygotes analyzed), TIF1α was only seen in the male pronucleus, most likely reflecting the fact that the male pronucleus undergoes transcriptional activation earlier (). TIF1α remained associated with NLBs through the 2-cell stage and, although less prominent, throughout the 4-cell stage. This pattern of localization in dense spots was specific for TIF1α because TIF1β was uniformly distributed throughout the nucleoplasm of the two pronuclei (). The time when we observed TIF1α translocation into the pronuclei coincides with the time when chromatin remodelers and transcription machinery factors, such as Brahma-related gene 1 (BRG-1; SMARCA4), Brahma (BRM; SMARCA2), and high mobility group box 1 (HMGB-1), translocate into the pronuclei (; ; ). This is also associated with the appearance of the hyperphosphorylated (active) form of the RNA polII (), concomitant with the activation of transcription of the embryonic genome (). Thus, the change of TIF1α localization from the cytoplasm to the pronuclei occurs at the time of embryo genome activation. To examine whether TIF1α is associated with regions of active transcription in the embryo, we assayed whether the sites of 5-bromo UTP (BrUTP) incorporation in vivo colocalize with TIF1α in the zygote. BrUTP staining was detected throughout the pronuclear nucleoplasm of the zygote, and sites of higher BrUTP accumulation were observed in the periphery and the proximity of the NLBs (, +BrUTP; and Fig. S2, available at ). Immunostaining for TIF1α in BrUTP-injected zygotes revealed that TIF1α colocalized with some of these sites of greater BrUTP incorporation. Note, however, that not all of the BrUTP sites colocalized with TIF1α (). This data indicates that TIF1α is recruited to specific sites of RNA synthesis at the late zygote stage. Because the fully grown oocyte is transcriptionally silent (), chromatin remodeling is expected to be required after fertilization to enable embryo genome activation. The TIFs are characterized by the presence of a bromodomain in the C terminus, and it is known that bromodomain-containing proteins can have a role in chromatin remodeling, gene repression, and gene activation (; ; ). This led us to examine whether TIF1α colocalizes with chromatin remodelers. We assayed whether the sites of TIF1α accumulation relate to the localization of the ATPase subunits of the mammalian types switching defective–sucrose nonfermenting (SWI–SNF) and Imitation of Switch (ISWI) remodeling complexes. SNF2H (SMARCA5) is the ATPase subunit of the mammalian ISWI complex (). At the late zygote stage, SNF2H localized to small foci throughout the nucleoplasm of both pronuclei and to larger foci around the NLBs (). BRG-1 is the ATPase subunit of the mammalian SWI–SNF complex (). BRG-1 localized to bigger foci than those of SNF2H and displayed increased accumulation around the NLBs (). A similar distribution has been reported for BRG-1 in earlier zygotes (). As expected, TIF1α did not colocalize with SNF2H in early zygotes (, top). In contrast at the late zygote stage, we found that sites around the NLBs that were enriched with TIF1α were also enriched with both SNF2H and BRG-1 (, middle and bottom). Moreover, similar to the pattern of BrUTP incorporation, not all SNF2H and BRG-1 foci contained TIF1α. Thus, sites of accumulation of TIF1α are also enriched with chromatin remodelers in the late zygote stage. We next wished to assess the function of TIF1α at the beginning of development of the mouse embryo. To this end, we used two methods: RNAi and injection of antibodies. For RNAi, zygotes were microinjected with double-stranded RNA (dsRNA) for TIF1α at the fertilization cone stage. Injections of dsRNA for GFP as well as noninjected embryos were used as negative controls. We found that embryos injected with TIF1α dsRNA proceeded through the first cleavage and reached the 2-cell stage at the same time as the control embryos. However, although the control embryos developed normally to the blastocyst stage (noninjected, 96%, = 85; dsGFP, 92%, = 70; five independent experiments), the majority of the embryos injected with dsRNA for TIF1α arrested at the 2–4-cell stage (66%; = 80; five independent experiments; ). 19% of these embryos arrested at the 2-cell stage, 30% arrested at the 3-cell stage, and 15% developed only to the 4-cell stage (). To examine whether the down-regulation of TIF1α upon RNAi in zygotes was efficient, we analyzed embryos that had been injected with dsRNA for TIF1α or for GFP by Western blot, which showed that TIF1α protein was efficiently knocked down (). We also verified that injection of dsRNA for TIF1α was specific: it did not result in the reduction of TIF1β, E-cadherin, or β-actin mRNA levels in these embryos, and the protein levels of tubulin were unchanged ( and Fig. S1 a, available at , and see ). As an additional approach to test TIF1α function, we performed a similar series of experiments, this time blocking TIF1α protein through injection of antibodies. Although the majority of the Flag antibody–injected (87%; = 35; four independent experiments) and noninjected control embryos (100%; = 56; four independent experiments) developed to the blastocyst stage, 86% of the embryos injected with antibodies against TIF1α arrested between the 2- and 4-cell stages ( = 50; four independent experiments; ). Most embryos (46%) arrested at the 3-cell stage, and 20% of the embryos developed to the 4-cell stage. Thus, injection of antibodies, similarly to RNAi, caused the majority of embryos to arrest at the 2–4-cell stage. Although the arrest was slightly stronger upon antibody injection, this is unsurprising, as the injection of antibodies may result in a more immediate neutralization of TIF1α than RNAi. These results show that reducing the levels of TIF1α by two complementary approaches (either through interference with the message or with the protein) results in a decreased number of embryos that develop to the blastocyst stage. To further understand the phenotype resulting from ablation of TIF1α, we injected TIF1α antibodies before the onset of genome activation and examined the localization of SNF2H and BRG-1 at the late zygote stage, that is, at the time of genome activation. We also analyzed the localization of the active (Ser5-phosphorylated) form of the RNA polII in the injected zygotes. We found that ablation of TIF1α resulted in a change in the distribution of the active Ser5-phosphorylated form of RNA polII (). The active RNA polII localized in a patchy and foci-network distribution instead of the homogeneous pattern throughout the nucleoplasm observed in the control embryos ( = 10; ), suggesting an effect on transcription. Blocking of TIF1α resulted in the mislocalization of BRG-1, which was only barely detected in the pronuclei of the injected zygotes and showed a diffuse staining in the cytoplasm ( = 9; ). The distribution of SNF2H was also affected: the small foci observed throughout the nucleoplasm of the control embryos were no longer visible after ablation of TIF1α. Instead, SNF2H accumulated in few larger foci ( = 10; ). Given that heterochromatin protein 1 (HP1) recruitment and histone acetylation increase gradually in the zygote and that both are involved in regulation of gene expression (; ; ), we also examined the effect of TIF1α ablation on HP1 localization and histone acetylation. Moreover, HP1 proteins can associate with both active and silent chromatin (). Ablation of TIF1α did not affect HP1β ( = 15) or HP1γ ( = 9) localization. Similarly, the acetylation status of lysines 14 and 18 of histone H3 remained unchanged and that of histone H4 was not drastically affected ( = 8; ). Similar results were observed when the same experiments were performed upon RNAi (unpublished data). Given that TIF1α ablation provoked a change in the localization of the RNA polII, we next wished to assess whether ablation of TIF1α elicited a general defect in transcription. To this end, we analyzed the pattern of staining of BrUTP incorporation in the late zygote after ablation of TIF1α. The embryos remained transcriptionally active after interference with TIF1α (). Ablation of TIF1α did not appear to result in striking differences in the pattern of BrUTP incorporation in comparison with the control groups. However, an in-depth observation revealed that the general signal of fluorescence was more disperse, and the accumulation of BrUTP around the NLBs seemed slightly enhanced. This observation was confirmed by quantification of the area containing transcription foci, which showed a small and reproducible increase in the area being transcribed in the embryos after ablation of TIF1α compared with the two control groups (). Thus, blocking of TIF1α did not abolish transcription, consistent with our observation that TIF1α localizes only to specific sites of active transcription (), but induced a significant change in the area of BrUTP incorporation. Mislocalization of BRG-1 and SNF2H observed by immunofluorescence after TIF1α ablation suggested that TIF1α might be involved in the nuclear localization of these two chromatin remodeling proteins in the late zygote. Given that not all of the BRG-1 colocalized with TIF1α (), the reduced signal of BRG-1 staining in the pronuclei resulting from TIF1α ablation suggests that TIF1α may play a role in the nucleation of BRG-1. Alternatively, the absence of TIF1α could affect the expression of SNF2H and/or BRG-1. We attempted to examine by Western blot whether the protein levels of SNF2H and/or BRG-1 were affected after ablation of TIF1α, but because of technical limitations attributable to the amount of material, we could not draw any conclusion. However, we found that the mRNA levels of both BRG-1 and SNF2H were maintained in the embryos after ablation of TIF1α (Fig. S2 b). We observed a slight decrease of the mRNA levels of SNF2H upon TIF1α ablation, which could correlate with the decreased staining that we observed in our immunofluorescence experiments. Thus, our data indicate that ablation of TIF1α function results in the mislocalization of BRG-1 and SNF2H in the zygote. The mislocalization of the active form of the RNA polII, together with the change in the transcriptionally active area resulting from TIF1α loss, could indicate that specific sites of initiation of transcription may be disrupted and/or mislocalized in the zygote after TIF1α ablation. Therefore, we next examined whether TIF1α binds to specific genes in the zygote and whether the expression of these genes would be misregulated as a consequence of TIF1α ablation. To this end, we first used a chromatin immunoprecipitation (ChIP) cloning approach in late zygotes, which we modified to circumvent the constraint of the requirement of large amounts of material (see the supplemental text, available at ). Our approach allowed us to identify 18 candidate target genes of TIF1α in the late zygote. These encode proteins that perform diverse cellular functions (). Second, to validate these target genes and to explore whether the genes identified by ChIP cloning are indeed regulated by TIF1α, we randomly chose 10 of them and examined their expression in embryos that had been subjected to TIF1α RNAi. We injected dsTIF1α in zygotes before the onset of genome activation (at the fertilization cone stage) and performed RT-PCR at a time when the embryos would have been gone through genome activation. We found that 9 out of the 10 genes that we analyzed were indeed misregulated after TIF1α interference (). The changes elicited in the levels of gene expression varied from complete loss of the corresponding mRNA ( and ), to partial () or very slight ( and ) down-regulation, to robust up-regulation (, , , and ). Although the expression pattern between the zygote and the 4-cell stage of most of these genes is not known, expression has been shown to increase its mRNA levels around the zygote stage (), consistent with it being one of the genes that requires TIF1α to be activated at the zygote stage (). To verify whether the changes in gene expression upon TIF1α RNAi were specific, we analyzed the mRNA levels of three genes as internal control: β-actin, TIF1β, and E-cadherin. None of these three genes showed changes in their mRNA levels ( and Fig. S2). This suggests, in agreement with what we observed for the BrUTP incorporation profile, that down-regulation of TIF1α does not elicit a general defect in transcription but only affects the expression of a specific set of genes. Moreover, TIF1α acts not only as an activator of its target genes but can also prevent the activation of others. Importantly, genes such as , (HNF1β), , and have documented functions in early embryonic development and/or cell growth (; ; ; , ). Thus, our data indicate that TIF1α is required to determine the transcriptional state (active or repressed) of a set of genes in the late zygote. After observing the mislocalization of SNF2H and BRG-1 upon TIF1α ablation, we hypothesized that if these chromatin remodelers are relevant for its function in the zygote, the expression of at least some of the TIF1α target genes should be affected when either BRG-1 or SNF2H are knocked down. To test this hypothesis and given that SNF2H can coimmunoprecipitate with TIF1α (Fig. S4, available at ), we performed RNAi for SNF2H using the same conditions as for TIF1α RNAi. Early zygotes at the fertilization cone stage were microinjected with dsRNA for SNF2H; injections of dsRNA for GFP as well as noninjected embryos were used as negative controls. The embryos subjected to SNF2H RNAi divided to the 2-cell stage normally ( = 52). The control embryos developed normally to the blastocyst stage (noninjected, 95%, = 23; dsGFP, 91%, = 23). Although approximately half of the embryos injected with dsRNA for SNF2H developed to the morula and blastocyst stages (54%; = 52), the other half showed developmental arrest between the 2- and 8-cell stages (46%; = 52; ). We then analyzed the same genes that we analyzed after RNAi for TIF1α. RT-PCR of these genes revealed a subset of genes (, , and ) that showed a drastic down-regulation in their mRNA levels after injection of dsRNA for SNF2H (). These genes corresponded to the genes that were down-regulated upon TIF1α RNAi (). Similar to what we observed for TIF1α RNAi, we also observed a slight decrease in the expression of and after RNAi for SNF2H. We did not observe any effect on the expression of , , , or , which remained not expressed in the embryos after RNAi for SNF2H (). This was in contrast to what we observed after RNAi for TIF1α, which resulted in a robust up-regulation of the corresponding mRNA for this second group of genes (, compare a and b). shows that SNF2H knockdown was induced efficiently in the embryos. Thus, although the effect on gene regulation elicited upon TIF1α ablation was both up- and down-regulation of target genes, RNAi for SNF2H resulted only in down-regulation of the same target genes that were down-regulated after TIF1α RNAi. This data suggests that TIF1α regulates activation of gene expression of a subset of its target genes in the zygote through SNF2H function. We have investigated the role of TIF1α in the early development of the mouse embryo. We show that at the onset of genome activation, TIF1α translocates into the pronuclei and accumulates in specific regions of RNA synthesis that are enriched with chromatin remodelers. Ablation of TIF1α results in the mislocalization of RNA polII, SNF2H, and BRG-1, and in the misregulation of a particular set of genes. Thus, TIF1α is a maternal factor that functions in the first wave of embryonic genome activation as a modulator of the transcriptional state of a subset of genes. TIF1α was originally cloned because of its ability to interact with nuclear receptors (). Although we cannot rule out a functional interaction with the nuclear receptors at this stage, the expression of the nuclear receptors known to interact with TIF1α is undetectable in the stages of development that are within the time window of our study (). Our data show that TIF1α plays a role as a modulator of embryonic transcription and suggest that its function in the zygote is required for the proper localization of chromatin remodelers and the RNA polII. Several reports have documented TIF1α acting as a repressor or as an activator in cultured cells and, therefore, its role as a coactivator appears controversial (, ; ; ). Although the discrepancies could be explained by the differences in the systems used in those studies, it is also likely that TIF1α plays a dual role in regulating repression versus activation of specific genes. This is supported by the effects on gene expression observed here after TIF1α RNAi in early mouse embryos. TIF1α function could also be regulated, as it may associate with different chromatin remodeling complexes, ultimately causing changes in the transcription of selected genes. Thus, the proteins TIF1α associates with would determine the specificity and the outcome on transcription (activation versus repression). Indeed, remodeling complexes containing BRG-1 and SNF2H can lead to both activation and repression of gene expression (, ; ). Our results suggest that TIF1α regulates activation of a subset of its target genes through SNF2H function. Further, lack of recruitment of BRG-1 may also account for some of the changes in gene expression that we observed after TIF1α ablation. Mislocalization of SNF2H, BRG-1, and RNA polII itself suggests that TIF1α may be involved at least partially in the localization of these remodeling complexes in the zygote. In support, we found that SNF2H can coimmunoprecipitate with TIF1α (Fig. S4). Moreover, TIF1α's ability to nucleate the formation of a ternary complex with coactivators has recently been documented (). Thus, we propose that recruitment of TIF1α to specific sites in the genome would ensure the localization of initiating RNA polII on one hand and of chromatin remodeling complexes on the other, and the “choice” of particular chromatin remodeling complexes will determine the outcome on transcription. The tripartite motif proteins have been implicated in processes such as cell differentiation, growth, and development. In , Bonus, a TIF homologue, is essential for cell viability and embryogenesis (). Of the four TIF members reported in mammals (, ; ; ), TIF1β has been shown to be required for the specification of the anteroposterior axis in the mouse (). However, in view of the observation that TIF1β is also expressed in early embryos, it remains to be established whether TIF1β also plays a role earlier in development. Although the protein motifs in the TIFs are conserved—a tripartite domain composed of a coiled-coiled, a RING (really interesting new gene) domain, and a B-box, and a bromodomain in the C terminus ()—some molecular differences have been documented that translate into functional differences among the TIFs: only TIF1β can target histone deacetylase activity, thereby acting as a corepressor, and it localizes to heterochromatin, the latter via interactions with HP1 (; ). Additionally, TIF1β has so far not been reported to interact with nuclear receptors, in contrast to TIF1α (). Moreover, TIF1α possesses a kinase activity () that has not been documented for the other TIFs. Likewise, the RING domain of TIF1γ acts as a ubiquitinase (), but this activity has so far not been detected in TIF1α or -β. Interestingly, this RING-like ubiquitinase activity is required for ectoderm induction in (). The functional heterogeneity of the TIFs may account for the different roles that so far have been assigned to some of them during embryogenesis. Our work now documents a role for TIF1α in early development and in regulation of transcription in the mouse zygote. In this context, it is important to note that altogether the group of TIF1α target genes that we have identified cover several cellular processes that are essential for early development, such as translation () and adhesion (, , and ; ; ). In fact, the expression of some translation initiation factors correlates with the maternal-to-zygotic transition in mouse embryos (). The target genes under the “unknown” category include a conserved mRNA for a protein containing a highly basic lysine domain of unknown function (), a protein with domains predicted to be involved in RNA processing and transcriptional regulation (), and an mRNA deadenylase (). Although the relevance of each of these genes in early development remains to be investigated, their coordinated expression may be of functional significance in the control of the maternal-to-zygotic transition. The expression pattern of TIF1α in the preimplantation embryo is reminiscent of that of , which is expressed initially in all blastomeres but then becomes restricted to the ICM, and whose expression is essential for maintaining the pluripotency of the ICM cells (). Moreover, TIF1α expression decreases upon differentiation of embryonic stem cells (). It is also noteworthy that TIF1α has been reported to be a direct target gene of Nanog in mouse embryonic stem cells (). Thus, in the future, it will be interesting to determine whether expression of TIF1α contributes to the establishment or the maintenance of the pluripotent capacities of the early mouse embryo. Such a role for TIF1α is supported by the failure of most embryos lacking TIF1α to develop to the blastocyst stage, and by the changes in the localization of SNF2H and BRG-1 resulting from TIF1α ablation, both of which are required for ICM and/or trophectoderm survival in the mouse (; ). During the early stages of development, decisions about cell fate determination, pluripotency, and patterning are made. Thus, the chromatin has to be dynamically remodeled for opening and closing specific regions in response to those events. TIF1α could take part in this process by activating or repressing particular sets of genes. Our data suggest that TIF1α is a factor involved in epigenetic mechanisms in early mammalian development. Embryos were collected from F1 (C57BL/6 × CBA/H) 6-wk-old superovulated females as described previously (). For the RNAi experiments, F1 females were mated with EF-1α MmGFP transgenic males (). All other experiments were performed with F1 × F1 crosses. Zygotes and cleavage stage embryos were collected at the indicated hours after human chorionic gonadotrophin (hCG) injection and cultured in KSOM medium (Specialty Media, Inc.) as described previously (). All animals were handled in accordance to Home Office legislation. Freshly collected embryos at various stages were fixed in 4% paraformaldehyde in PBS. In situ hybridizations were performed as described previously (), except that the embryos were not dehydrated and the proteinase K treatment was omitted. The TIF1α probe was prepared and labeled with digoxygenin-UTP using the pSK.TIF1α plasmid (provided by R. Losson, Institut de Génétique et de Biologie Moléculaire et Cellulaire [IGBMC], Strasbourg, France) as a template (). After removal of the zona pellucida with acid Tyrode's solution (Sigma-Aldrich), embryos were washed three times in PBS and fixed in 5% paraformaldehyde, 0.04% Triton X-100, 0.3% Tween, and 0.2% sucrose in PBS for 20 min at 37°C. After permeabilization with 0.5% Triton X-100 in PBS for 20 min, the embryos were washed three times in 0.1% Tween in PBS (PBS-T), blocked in 3% BSA in PBS-T, and incubated with the primary antibodies for ∼12 h at 4°C. Embryos were then washed twice in PBS-T, blocked for 30 min, and incubated for 2 h at 25°C with the corresponding secondary antibodies. After two washes in PBS-T, the DNA was stained with TOTO-3 (Invitrogen) and the embryos were mounted in Vectashield (Vector Laboratories) and analyzed using a 60×/1.40 oil objective (Nikon) in an upright confocal laser microscope (Radiance; Bio-Rad Laboratories) using the LaserSharp 2000 software (Bio-Rad Laboratories) at room temperature. The antibodies used in this work are as follows: TIF1α (Santa Cruz Biotechnology, Inc.), KAP1 (TIF1β; Abcam), RNA polII (recognizing the CTD phosphorylated in Ser5; CTD4H8; Upstate Biotechnology), BRG-1 (Santa Cruz Biotechnology, Inc.), hSNF2H (provided by P. Varga-Weisz, The Babraham Institute, Cambridge, UK; ), tubulin (Sigma-Aldrich), HP1β (IGBMC), HP1γ (IGBMC), acetylated histone H4 (Upstate Biotechnology), acetylated K14 histone H3 (provided by B. Turner, University of Birmingham, Birmingham, UK), and acetylated K18 histone H3 (Abcam). Secondary antibodies were coupled with either Alexa Fluor 568 or 488. Images were then prepared or analyzed using Photoshop 7 (Adobe) and Volocity (Improvision), respectively. BrUTP labeling was performed as described previously (). Embryos were collected 24 h after hCG injection and microinjected using a transjector (model 5246; Eppendorf) with 1–2 pl of 100 mM BrUTP (Sigma-Aldrich) in 2 mM Pipes and 140 mM KCl, pH 7.4. Embryos were fixed after 3 h of culture and processed for immunostaining using an anti-BrdU antibody (Sigma-Aldrich). For quantification of BrUTP incorporation in pronuclei after microinjection of antibodies, the embryos were collected at 20 h after hCG injection and microinjected with BrUTP followed by microinjection of antibodies as described (see Microinjection of antibodies). For the analysis after immunostaining, the area of the pronucleus of injected zygotes was defined and cropped using Volocity. The pixels were then selected under a 30% tolerance level, and the area displaying BrUTP staining was quantified using the measurement tools of the software as recommended by the manufacturer. Zygotes were collected and microinjected at 20 h after hCG injection with 1–2 pl of 1 μg/μl long dsRNA for TIF1α, long dsRNA for SNF2H, or long dsRNA for GFP (). The sequence for the dsRNA for TIF1α spans nucleotides 1284–1771, which shares no homology with the other members of the family. For the SNF2H RNAi experiments, dsRNA spanning nucleotides 2041–2520 of the cDNA was used. For RT-PCR analysis, embryos were collected ∼42 h after dsRNA injection and processed for RT in pools of five embryos, each using the Dynabeads mRNA direct micro kit (Dynal). Embryos were collected at the same stages for all the samples to avoid variation resulting from embryos derived from different stages. Half of the mRNA extracted (10 μl) was used for the reverse-transcriptase controls and the other half for cDNA synthesis. PCR was performed with 1/20 of the cDNA (0.5 μl), such that all genes were analyzed in the same sample using 60 cycles for amplification, except for and , in which 35 cycles were used, and , , and , in which 50 cycles were used. It was verified that the cycling conditions were within the exponential phase of amplification for each gene. The products were transferred onto a Hybond N+ membrane (GE Healthcare), hybridized against the corresponding probes, and exposed for autoradiography. Antibodies against TIF1α and Flag (Sigma-Aldrich) were microdialyzed overnight at 4°C against Tris-EDTA, pH 8.0, and concentrated using a filter (Centricon; Amicon) to a final concentration of 215 ng/μl (). Zygotes collected at 20 h after hCG injection were microinjected with ∼1–2 pl of antibody solution and cultured. For the immunostaining analysis, the embryos were fixed after 7 h of culture. Embryos from five different experiments were collected 42 h after dsRNA injection, washed three times in PBS, pooled (155 embryos per group), and subjected to Tris-Glycine PAGE-SDS. Competition assays with the corresponding TIF1α-blocking peptide were performed to ensure the specificity of the antibody (Fig. S3, available at ). We first assessed the ability of the antibody to immunoprecipitate TIF1α in pronuclei extracts (Fig. S3). For the ChIP, 413 zygotes were collected in M2 at 27 h after hCG injection, formaldehyde cross-linked, washed, and lysed in 5 mM Pipes, pH 8.0, 85 mM KCl, and 0.05% NP-40. The pronuclei were then lysed and sonicated. For the immunoprecipitation, 1 μg of TIF1α antibody was used after preclearing of the chromatin. The samples were then extensively washed, eluted with 50 mM NaHCO and 1% SDS, and treated with proteinase K at 65°C for 4 h. After purification, DNA was incubated with T4 DNA polymerase and ligated to two unidirectional linkers (). The samples were amplified by PCR, cloned into pGEM-T Easy (Promega), and sequenced. Out of 25 clones sequenced, 19 contained inserts that corresponded to regions of different genes, many of them unknown. The remaining six clones contained background sequences corresponding to the cloning vector or . Two clones contained sequences of the same gene, which led us to select 18 candidate genes. We provide a detailed protocol in the supplemental text. Fig. S1 shows typical BrUTP accumulation in late zygote and 2-cell stage embryos. Fig. S2 provides evidence that RNAi for TIF1α does not induce down-regulation of TIF1β and an analysis of the mRNA levels of SNF2H and BRG-1 upon TIF1α RNAi. Fig. S3 shows the characterization of the TIF1α antibody used in this work. Fig. S4 depicts coimmunoprecipitation of SNF2H with TIF1α. The supplemental text provides a detailed protocol for ChIP cloning in zygotes. Online supplemental material is available at .

Tbx5 and -4 belong to the family of T-box transcription factors that share a homologous DNA binding domain (T-domain) first described in the mouse (or ) gene product (; ). Studies in chicken (; ; ), zebrafish (; ), and mouse (; ; ) revealed that both and - play critical roles in the outgrowth and specification of vertebrate forelimbs and hindlimbs, respectively. In addition to the limbs, has been shown to be required for proper heart development in zebrafish () and mouse (). In the chicken, expression has also been described in the heart, complementing the asymmetrical expression in this organ and suggesting parallel pathways for these transcription factors in the limbs and heart (). Although there is strong evidence that and - are critical for embryonic development, little is known about how the transcription factors are regulated and function at the cellular level. In a protein–protein interaction screen, we recently identified from chicken a new protein called LMP4, by its ability to interact with the C-terminal transactivation domain of the Tbx5 and -4 transcription factors (). In chicken embryos, is expressed in the developing eye, heart, forelimbs, and hindlimbs, all organs that express either or - (; ; ). LMP4 is a member of an emerging class of scaffolding proteins, denoted PDZ-LIM proteins, which appear to function in fundamental biological processes, including cytoskeletal organization, cell lineage specification, and organ development (; ). PDZ-LIM proteins contain cassettes of two different types of protein–protein interaction domains: a single N-terminal PDZ domain and one or three C-terminal LIM domains. The PDZ domain is an 85-amino-acid β-barrel protein interaction motif that binds to both C-terminal peptides and internal sequences of target proteins (). The PDZ domains of the PDZ-LIM proteins Enigma homologue (ENH) 1 and CLP-36 both bind to α-actinin, and this interaction localizes the proteins to actin filaments (; ). The LIM domain is a 55-amino-acid sequence that contains two zinc finger–like motifs with conserved cysteine residues (). The LIM domains of PDZ-LIM proteins have been found to interact with protein kinases, such as Clik1 (), PKC (), and receptor tyrosine kinases (; ). All of the described binding partners for PDZ-LIM proteins suggest a role for this protein family as mediators, regulating protein function and/or signaling. PDZ-LIM family proteins can be subdivided into two subclasses depending on the number of LIM domains present. For example, CLP-36 contains a single C-terminal LIM domain, whereas ENH1 and LMP1 contain three. The ENH and LMP proteins share significant sequence homology between their PDZ and LIM domains. However, there still appears to be specificity within the binding motifs. The PDZ domains of rat ENH1 and human Enigma (an LMP protein) bind to α-actinin () and β-tropomyosin (), respectively, whereas the LIM domains of each protein bind different isoforms of PKC (). We have proposed that LMP4 interacts with Tbx5 and -4 and regulates their activities by localizing the transcription factors out of the nucleus (). Building on our previous developmental studies, we focus on Tbx5 and use it as a model to understand the mechanism of LMP4–Tbx interactions. We have conducted a detailed cellular investigation using cell biology and biochemical techniques to test our hypothesis and uncover a novel mechanism that regulates Tbx protein subcellular localization and transcriptional activity. Previously, we showed coexpression of and mRNA in chicken embryos during heart and forelimb development. Additionally, Tbx5 and LMP4 protein binding was shown in vitro, and initial experiments with transfected cells suggested that the Tbx5 and LMP4 proteins colocalized at cytoplasmic sites (). To extend our earlier findings, we wished to determine whether such protein interactions occur within a developmental context in vivo. To gain insight into this question, we turned to the developing chicken wing. Wings from Hamburger-Hamilton (HH) stage 36 () chicken embryos were sectioned, and serial sections were stained with Tbx5- () and LMP4-specific antibodies (Fig. S1 and supplemental text, available at ). As expected, we detected Tbx5 protein in mesenchymal cells but not in the outer epidermal layer (; and not depicted). Interestingly, Tbx5 was localized both within the nucleus and at cytoplasmic sites; however, the ratio of nuclear to cytoplasmic distribution varied in different regions of the limb. This finding demonstrates that Tbx5 protein in the developing chicken wing is not strictly localized to the nucleus. Cytoplasmic localization of TBX5 has also been reported in the human lung (). In adjacent tissue sections, the expression of LMP4 was predominantly cytoplasmic, revealing a punctate and filamentous pattern, reminiscent of Tbx5 cytoplasmic localization (). No obvious nuclear localization was detected for LMP4. The expression of Tbx5 and LMP4 at the RNA level in HH stage 36 wings was confirmed by RT-PCR, whereas the Tbx5-specific antiserum did not detect protein in respective hindlimb cryosections (unpublished data). Of note, in experiments with developing chicken hearts, we have observed similar subcellular localization patterns for Tbx5 and LMP4 (unpublished data). This would indicate a more general role for nuclear and cytoplasmic Tbx5 distribution and suggest functional significance of the LMP4–Tbx5 interaction in vivo. Finding that Tbx5 protein displayed cytoplasmic distribution during development supported our model that LMP4 may regulate Tbx5 on a cellular level. To elucidate in more detail the mechanism of Tbx5 and LMP4 interactions, we turned to cultured cells. During chicken heart development, epicardial cells arise from the proepicardial organ, migrate over the heart, and through a process of epithelial-to-mesenchymal transition (EMT) give rise to cardiac fibroblasts and coronary smooth muscle cells that contribute to the myocardial wall, atrioventricular cushion and valves, and coronary vasculature (; ). Epicardial cells have been shown to natively express Tbx5 (), and in vitro cultures of these cells have been used as a cell differentiation model during EMT (; ). We have used such a chicken primary epicardial cell culture system to determine if Tbx5 and LMP4 subcellular localization would change after stimulating the cells to differentiate. Cultured epicardial cells maintained in serum-free media remain in an undifferentiated state, characterized by predominantly cortical actin and absence of differentiation markers such as calponin (). In contrast, cells cultured in chicken embryonic heart–conditioned media differentiate into epicardially derived cells (EPDCs), which lose their cortical actin organization and begin to express the smooth muscle marker calponin (; ). Monolayers of chicken epicardial cells were grown from HH stage 25 hearts in serum-free media. These epicardial cells were then processed for indirect confocal microscopy to detect endogenous Tbx5 and LMP4 using specific antibodies. Alternatively, cultures were induced to differentiate into EPDCs using embryonic heart–conditioned media followed by immunocytochemical analysis. Using the Tbx5-specific antiserum, the transcription factor was detected in cultured epicardial cells and EPDCs by confocal microscopy () and confirmed by Western blot using cell lysates (not depicted). Interestingly, Tbx5 protein localization was found to change in the chicken primary heart cultures depending on cellular context. In epicardial cells cultured in serum-free media, the actin cytoskeleton was predominantly cortical, consistent with undifferentiated epicardial cells (). A further indication for the undifferentiated status of these cells is the absence of calponin expression (Fig. S2, available at ; ). In these epicardial cultures, Tbx5 was predominantly nuclear (), similar to previous observations in transfected cells (; ; ). However, when epicardial cells were shifted from serum-free to heart-conditioned culture media, they differentiated into EPDCs (). The altered differentiation status was indicated by the drastic change in cell morphology, as outlined by the reorganization of actin, from predominantly cortical to mostly filamentous stress fibers (). In addition, the cells expressed the differentiation marker calponin (Fig. S2). In the EPDCs, Tbx5 changed its localization from predominantly nuclear to a combination of nuclear and cytoplasmic distribution (). Furthermore, a significant amount of the cytoplasmic Tbx5 protein colocalized with phalloidin-stained actin filaments (). The distribution of Tbx5 within the EPDCs demonstrates that the transcription factor is not strictly localized to the nucleus and that its localization is regulated depending on cell differentiation status, a finding that has not previously been described. Our model suggests that Tbx5 cytoplasmic localization is due to direct interactions with LMP4, and our previous study showed that LMP4 localizes Tbx5 to actin in transfected COS-7 cells (). To correlate the dynamic localization of Tbx5 to the LMP4 binding partner, we assayed for expression of LMP4 in the epicardial cells. As detected by indirect fluorescence () and by Western blot of cell lysates (not depicted), using the LMP4-specific antiserum, chicken epicardial and EPDC cultures natively expressed LMP4. In the undifferentiated epicardial cells, LMP4 predominantly colocalized with cortical actin, outlining cell margins (). LMP4, as well as other PDZ-LIM protein family members, has been shown to associate with the actin cytoskeleton, supporting the localization in epicardial cells (; ; ; ). Additional cytoplasmic localization was also observed in epicardial cultures for LMP4, along with some staining in or around the nucleus. When epicardial cells were shifted to differentiation medium, the cells began to express calponin (Fig. S2) and LMP4 remained predominantly localized to actin, now organized to stress fibers (). No distinctive nuclear localization could be detected in differentiating EPDCs. Thus, although Tbx5 displayed a dramatic localization shift, LMP4 remained cytoplasmic and associated with actin as the epicardial cells differentiated. Interestingly, the filamentous localization patterns for Tbx5 and LMP4 in the differentiating EPDCs were strikingly similar. Costaining EPDCs for both Tbx5 and LMP4 demonstrated that the two proteins colocalized along the actin cytoskeleton, supporting the notion that the proteins bind each other (). The epicardial cell data suggest a potential new role for Tbx5 and/or LMP4 during EMT in the heart, and this possibility is currently under investigation (unpublished data). In addition, the shift of subcellular localization, concomitant with the change of culture conditions of the epicardial cells, suggests that Tbx5 localization may be regulated by external signals. Our group, as well as others, has shown nuclear localization of Tbx5 in transfected cells (; ; ). However, many of these experiments used large fusion proteins such as EGFP for detection, which we have found to cause Tbx5 to function at suboptimal levels (see ). To reduce the risk for functional interference, we have constructed nontagged and small C-terminal epitope–tagged Tbx5 expression plasmids. Nontagged Tbx5 or Tbx5-HA expression constructs were transfected into COS-7 cells, and protein localization was detected by indirect fluorescence using anti-HA or Tbx5-specific antibodies. Identical results were obtained with tagged or nontagged Tbx5; however, for consistency with other experiments, data for Tbx5-HA are shown (). Using confocal immunofluorescence detection, Tbx5-HA displayed a clear nuclear localization in COS-7 cells. This microscopic examination was further verified by Western blot analysis of separated cytoplasmic and nuclear fractions (). Empty vector controls displayed no specific localization by Western blot or immunofluorescence ( and not depicted). For detection, we used the specific Tbx5 and anti-HA antibodies interchangeably with identical results (unpublished data). Previously, we showed that LMP4 localizes to cytoplasmic sites using an HcRed-LMP4 fusion expression construct (). To verify our earlier findings, similar to Tbx5, we used expression plasmids containing LMP4 without a tag or with a small C-terminal myc-epitope tag for transfections. LMP4 or LMP4-myc transfected into COS-7 cells was detected by indirect fluorescence using the LMP4-specific antiserum. (F–I) represents the data with LMP4-myc; however, identical results were obtained with nontagged LMP4 (not depicted). LMP4 protein displayed a filamentous distribution that overlapped with phalloidin-stained actin. This pattern is comparable to our previous data with HcRed-LMP4 and those obtained using an anti-myc antibody for detection (). Cellular fractionation confirmed LMP4 cytoplasmic localization (). We note that we could not detect with either method conclusive nuclear localization for LMP4 in these single-transfection experiments. Empty vector controls displayed no specific localization by Western blot or immunofluorescence ( and not depicted). Thus, using cell biology and biochemical methods, individually expressed Tbx5 and LMP4 localize to separate subcellular compartments in COS-7 cells—the nucleus and actin cytoskeleton, respectively. We next cotransfected COS-7 cells with LMP4-myc and Tbx5-HA, and protein localization was determined by indirect fluorescence using anti-HA and anti-myc antibodies (). The anti-HA antibody detected Tbx5-HA within the nucleus but also at cytoplasmic structures (). LMP4-myc, as detected by anti-myc, produced a localization pattern comparable to single transfections, indicating association with actin filaments (). However, we also observed a higher level of nonfilamentous cytoplasmic staining for this protein. Comparing Tbx5 and LMP4 localization in the merged image revealed colocalization of Tbx5 and LMP4 within the cytoplasm (), along polymerized actin (phalloidin stain not depicted; ), and at additional unidentified cytoplasmic sites. We note that in transfected COS-7 cells, we were only able to detect cytoplasmic localized Tbx5 in the presence of LMP4. The colocalization of Tbx5 and LMP4 within the cytoplasm supports our previous in vitro binding studies and suggests that the two proteins interact in cells (). To confirm the interactions by independent means, we performed protein coimmunoprecipitations. Lysates of COS-7 cells cotransfected with LMP4-myc and Tbx5-HA were subjected to immunoprecipitation with anti-myc antibodies and Western blot analysis (). Probing the Western blot with the Tbx5-specific antibody demonstrated that the transcription factor coprecipitated with LMP4 (, third lane). The vector controls and individual Tbx5-HA transfections did not result in immunoprecipitates with the anti-myc antibody (, first and second lanes). Nuclear/cytoplasmic fractionation experiments of cells coexpressing Tbx5 and LMP4 did not reveal any evidence for nuclear localization of LMP4, despite the appearance of some weak staining in the nuclear area of cotransfected cells (). The fractionation experiments also did not detect Tbx5, LMP4, or the Tbx5–LMP4 complex within the soluble fraction, indicating that they are part of additional protein complexes (unpublished data). The coimmunoprecipitation of LMP4 and Tbx5 supports the cellular colocalization and confirms the binding of the two proteins in the cell. In reciprocal coimmunoprecipitation experiments, Tbx5-HA was also able to coprecipitate LMP4-myc (unpublished data). Thus, Tbx5 localization outside the nucleus is a result of its interaction with LMP4. Because the Tbx5–LMP4 complex forms at actin filaments, it was important to investigate what role an intact actin cytoskeleton would have in mediating the interaction. To determine this, Tbx5 and LMP4 localization was observed in COS-7 cells with destabilized actin. 24 h after transfection of Tbx5-HA and LMP4-myc, cells were treated with 2 μM latrunculin A for 1 h to disrupt filamentous actin and then processed for confocal microscopy using indirect fluorescence (). Using the anti-HA antibody, Tbx5 was detected both in the nucleus and cytoplasm of actin-disrupted cells (). LMP4 was detected with the anti-myc antibody and found only in the cytoplasm of actin-disrupted cells (). As in nontreated cells, both Tbx5 and LMP4 appear to colocalize within the cytoplasm of latrunculin A–treated cells (); however, the proteins displayed no clear subcellular localization (). Despite the lack of specific localization, Tbx5 and LMP4 were still able to interact as demonstrated by imaging and coimmunoprecipitation (). Similar results were also obtained when actin was disrupted using 5 μM cytochalasin B for 1 h on cotransfected COS-7 cells (unpublished data). It remains to be determined whether an intact actin cytoskeleton is needed for initial Tbx5–LMP4 binding or whether it has a predominant role in the proper subcellular localization of the protein complex. However, it appears that the interaction of both proteins is maintained despite the lack of a complete actin cytoskeleton. Tbx5 has been shown to function as a transcription factor, activating target genes in the developing limb and heart (; ). In the mouse, the limb-specific and the heart-specific () genes have been shown to be immediate downstream targets of Tbx5 (; ). DNA fragments containing the respective promoters were ligated to the gene, and the resulting reporter constructs were used to determine Tbx5 transcriptional activity as a function of subcellular relocalization. COS-7 cells were transfected with constant amounts of the reporter and plasmids and increasing amounts of expression plasmids. To achieve optimal protein activities in this assay, nontagged and small-tagged constructs were used. Tbx5 alone revealed a robust activation of both the and reporters (), verifying that the chicken Tbx5 transcription factor can activate the respective mouse gene promoters at levels comparable to those of mouse Tbx5 (; ). Cotransfecting along with , however, resulted in decreased Tbx5 activity on both of the reporters. The ability of LMP4 to repress Tbx5 transcriptional activity was dose dependent, as increasing amounts of caused a linear reduction in Tbx5 activity (). The maximum amount of tested (300 ng) led to a repression of (25 ng) of 90 and 83% using the and promoters, respectively. Of note, our work with EGFP-Tbx5 and LMP4-EYFP revealed that such large fusion proteins have a significant reduction in activity. For example, transfection of COS-7 cells with 25 ng resulted in an ∼30% reduction of transcriptional activity on the promoter as compared with an equivalent amount of or nontagged (). All luciferase reporter data were normalized to luciferase to account for variability in transfection efficiency and expression. Luciferase assays were performed in triplicate, and data were collected from two independent experiments. Thus, LMP4 modulates Tbx5 transcriptional activity by relocalizing the transcription factor out of the nucleus. To determine if Tbx5 relocalization and transcriptional modulation by LMP4 was due to shuttling of the transcription factor out of the nucleus, FRAP experiments in COS-7 cells were performed. For this series of experiments, it was essential to use an fusion construct to visualize protein in living cells. EGFP-Tbx5 revealed a reduced level of transcriptional activity compared with nontagged or small HA-tagged Tbx5 (); however, the fusion protein retained the ability to colocalize with LMP4 within the cytoplasm of cotransfected cells (not depicted; ). COS-7 cells were cotransfected with and or and grown on glass-bottomed culture dishes for live cell confocal microscopy. As expected, in the background of LMP4, EGFP-Tbx5 was detected in both nuclear and cytoplasmic compartments (). To observe shuttling of the transcription factor, the EGFP fluorescence signal in either the cytoplasm or nucleus was bleached using the 488-nm laser at maximum intensity, and its recovery within the bleached compartment was observed over time. After photobleaching the cytoplasm, the EGFP fluorescent signal showed significant recovery over a 30-min time window ( and Video 1, available at ). It is important to note that the fluorescence recovery in the cytoplasm occurred with concomitant decrease of fluorescence in the nucleus, indicative of active shuttling of EGFP-Tbx5 out of the nucleus to cytoplasmic sites. Quantitative data analysis for a representative cell is shown in . Likewise, reciprocal experiments involving photobleaching the nucleus ( and Video 2) displayed EGFP fluorescence recovery of the nuclear compartment at the expense of the cytoplasmic signal over a similar 30-min time frame. This result indicates movement of EGFP-Tbx5 from cytoplasmic sites into the nucleus. As a control, whole cell FRAP of cotransfected cells was performed. No EGFP recovery was observed in control cells within the same time period in this setup, indicating that the experimental fluorescence recovery is due to protein shuttling and not to translation of additional EGFP-Tbx5 or maturation of EGFP (Fig. S3). Therefore, in the presence of LMP4, Tbx5 subcellular localization is dynamic and the transcription factor can shuttle between the cytoplasm and nucleus. We previously hypothesized that Tbx5 transcriptional activity may be modulated by dynamic interactions between Tbx5 and LMP4 proteins and localization of the complex to the actin cytoskeleton (). Here, we provide evidence in support of the Tbx5–LMP4 regulatory model. Individually, Tbx5 and LMP4 localize to separate cellular compartments— the nucleus and cytoplasm, respectively. However, when expressed within the same cell, Tbx5 is no longer strictly nuclear. The Tbx5 and LMP4 proteins bind and colocalize in the cytoplasm, predominantly in association with actin. The change in Tbx5 localization caused by LMP4 represses its ability to activate target promoters, as shown by in vitro luciferase reporter assays. The regulatory model would also imply a dynamic Tbx5–LMP4 complex assembly/disassembly, responding to external stimuli or signal transduction pathways and ultimately modulating Tbx5 protein activity. This notion is supported by relocalization of Tbx5 in differentiating epicardial cells and by dynamic shuttling of Tbx5 between the cytoplasmic and nuclear compartments in transfected COS-7 cells, as observed by FRAP. In addition, the coexpression of both Tbx5 and LMP4 proteins in developing chicken wings and, importantly, the localization of native Tbx5 outside of the nucleus in these limb cells indicate in vivo relevance. In chicken epicardial cells, we observed coexpression of Tbx5 and LMP4; however, in contrast to COS-7 cells, Tbx5 localization was predominantly nuclear in these cells. Cytoplasmic and actin-associated Tbx5 was not observed until cells were induced to differentiate. Cultured epicardial cells are well known to undergo such EMTs when stimulated with embryonic heart–conditioned medium or TGFs such as TGFβ (; ; ; ). For the first time, we have demonstrated a concomitant relocalization of Tbx5 proteins from the nucleus to the cytoplasm during this process, strongly suggesting that specific signaling pathways, potentially involving TGFβ-like factors, are involved in regulating Tbx5–LMP4 interactions and Tbx5 activity. The data presented are in agreement with the Tbx5–LMP4 regulatory model and point toward a complex pathway regulating Tbx5 activity by altering its localization depending on the developmental context of the cell. The nuclear concentration of many transcription factors is a dynamic balance that is determined by competing processes of nuclear import and export and by the presence of anchor proteins in both the nucleus and the cytoplasm. For example, NF-κB/Rel proteins, which are involved in diverse biological processes, were initially identified as constitutive nuclear transcription factors. Subsequent analysis, however, revealed that in most cells, NF-κB is sequestered in the cytoplasm via its interaction with IκB family proteins and only released into the nucleus in response to specific stimuli (). Additionally, the GLI-1 transcription factor has been shown to be not strictly nuclear but also cytoplasmic (; ). GLI-1 cytoplasmic localization has been shown to be due to interactions with Suppressor-of-Fused, and the export of GLI-1 from the nucleus regulates its transcriptional activity (). A similar novel mechanism may be emerging with LMP4 and Tbx5. When complexed with LMP4, a pool of Tbx5 is localized outside the nucleus in association with the actin cytoskeleton, thereby limiting the transcription factor's availability and activity in the nucleus. However, the specific signaling cascades that regulate the expression, localization, and function of Tbx5 have yet to be identified. Based on our initial studies with primary chicken epicardial cultures, it appears that specific stimuli are involved in the relocalization of Tbx5 during differentiation. Recently, TGFβ has been shown to induce differentiation of chicken epicardial cells into EPDCs (). It will be of interest to determine whether TGFβ in concert with LMP4 is required for Tbx5 relocalization in differentiating epicardial cells or to identify the nature of other specific upstream signals. In this context, it is noteworthy that PDZ-LIM proteins are thought to have diverse roles as regulators of cytoarchitecture, cell motility, signal transduction, and gene expression (; ). Several family member proteins such as Enigma, CLP-36, and the Cypher/ZASP proteins interact via their PDZ domains with the cytoskeleton (; ; ). Consistent with these findings, it is not surprising that in our studies, LMP4 is also colocalizing with the actin cytoskeleton. Of note, CLP-36's C-terminal LIM domain binds to and relocalizes the nuclear Clik1 kinase to actin stress fibers (). LIM domains in general are known to mediate protein interactions, and the close LMP4 family member Enigma binds to the insulin receptor (), receptor tyrosine kinases (), and PKC (). Although the exact roles of PDZ-LIM proteins such as Enigma in signaling cascades are currently speculative, the association with signal receptors and/or transducers provides an attractive link and points to an involvement in regulated signaling events, eliciting a change in binding partners. The presence of LMP4 in epicardial cultures, which display a differentiation response to external signals along with a significant relocalization of Tbx5, also suggests involvement of this PDZ-LIM protein in a signaling cascade. A cytoplasmic distribution for TBX5 in human lung during development has been indicated (). Our observations with developing wings, primary epicardial cells, and transfected cells reveal a previously unidentified localization of Tbx5 in both the nucleus and the cytoplasm. The actin-associated distribution of the Tbx5 transcription factor is particularly striking in the primary chicken EPDCs and would suggest that an equilibrium of Tbx5 in the nucleus and cytoplasm is important for the proper maintenance of its functions within the cell and, in turn, the organism. Although the system may compensate for some changes in protein levels, acting as a capacitor, significant over- or underexpression would be expected to result in deleterious consequences. Few reports are available on gain-of-function/overexpression phenotypes in higher vertebrates, but those available support our findings. For instance, retroviral overexpression of in the respective chicken wing/leg bud resulted in limb truncations, similar to misexpression of the respective dominant-negative constructs (). In addition, skeletal and cardiac malformations known as Holt-Oram syndrome (HOS) in humans are caused by mutations in . The majority of mutations critical for disease manifestation are thought to result in early protein terminations and haploinsufficiency; however, increased dosage, such as chromosome 12q2 duplication, has been reported to also result in HOS (; ). These data would suggest a more generalized effect of gene dosage, both under- and overexpression, in causing fairly similar, if not identical, phenotypes. The data presented here would also imply that a change in Tbx5 level would have direct consequences on Tbx protein distribution in the cell. This in turn may interfere with the differentiation program such as EMT of epicardial cells. In this context, it may be of significance that epicardial cells contribute to the myocardial wall, atrioventricular cushions, and valves, all cardiac structures that are predominantly affected in HOS. Therefore, balanced cellular Tbx5 levels and appropriate localization appear to be critical, and LMP4 may play a central role in this regulation. Experimental data coming from many animal models have provided clear evidence for the importance of Tbx5 in eye, limb, and heart development, and gene and RNA studies have provided some clues for the roles Tbx5 has in the cell. However, in addition to its role as a transcription factor, our new data may also point to unknown functions of Tbx5 outside the nucleus when associated with actin. An attractive possibility would be a direct role in regulating actin dynamics, a notion supported by the finding that Tbx5 function is involved in cell migration in zebrafish fin development (). A role in migratory behavior would be quite plausible also in light of the complex phenotypes of misexpression that have been observed in humans and animal models. This hypothesis can be tested, and future experiments mislocalizing Tbx5 to distinct cellular compartments and examining the resulting functional consequences will provide new insights into this question. COS-7 cells were grown in DME supplemented with 10% FBS, 1% -glutamine, and penicillin/streptomycin. Cells were transfected using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. For reporter assays, COS-7 cells were grown in 12-well culture dishes and transfected with 300 ng reporter plasmid, 10 ng luciferase plasmid, and expression plasmids as described. The total amount of DNA transfected was held constant at 1 μg. Transfected cells were cultured for 36 h before lysis. Luciferase activity was measured using the Dual-Luciferase assay system (Promega), and samples were read on a Lumat LB 9501 (Berthold). All reporter assays were performed in triplicate, and the collected data from two independent experiments were normalized to the luciferase activity. Tbx5-HA expression plasmids were provided by M. Logan (Medical Research Council, London, UK), and luciferase reporter constructs were provided by B. Bruneau (Hospital for Sick Kids, Toronto, Canada). To obtain chicken epicardial cell cultures, HH stage 25 () hearts were dissected and placed on fibronectin-coated coverslips in MEM without -glutamine (). After 24 h, the hearts were removed and the cells were either fixed for immunocytochemistry or induced to differentiate into EPDCs. Differentiation was induced by culturing cells with whole heart–conditioned media (), and after 3 d, cells were fixed and processed for immunocytochemistry. COS-7 cells were fixed in 4% PFA followed by 1% Triton X-100 extraction and sequential incubation with primary and secondary antibodies in 1% BSA. Affinity-purified rabbit polyclonal anti-LMP4 (Fig. S1) and anti-Tbx5 () were used at a 1:500 dilution. Anti-HA (HA-7; Sigma-Aldrich), anti-myc (9E10; Sigma-Aldrich), and anti-calponin (CP-93; Sigma-Aldrich) were diluted 1:500. Primary antibodies were detected using Alexa 488– and Alexa 546–conjugated secondary antibodies at 1:500 dilutions (Invitrogen). Filamentous actin was detected using Alexa Fluor 488 or 633 phalloidin (Invitrogen). Nuclei were stained using DAPI (Roche). For double-staining experiments, LMP4 antibodies were directly coupled to rhodamine using the EZ-Label protein labeling kit (Pierce Biotechnology). Confocal microscopy was performed using a 510 META system (Carl Zeiss MicroImaging, Inc.) equipped with a Plan Apochromat 63×/1.4 oil differential interference contrast lens. Images were processed in Photoshop CS2 (Adobe). Chicken wings from HH stage 36 were dissected in cold PBS, embedded in Tissue Tek OCT (Sakura Finetek), and frozen over dry ice. 10-μm sections were cut on a cryostat (CM3050S; Leica), fixed in 4% PFA, and processed for immunohistochemistry as described. Cells were transfected with EGFP-Tbx5 and either LMP4-myc or HcRed-LMP4. Cells were grown on uncoated glass-bottomed 35-mm culture dishes (No. 1.0; MatTek) containing DME/10% FBS and equilibrated on a 37°C heated stage fitted on a laser-scanning microscope (LSM 510; Carl Zeiss MicroImaging, Inc.). EGFP photobleaching was performed using the 488-nm laser line at 100% intensity. Changes in pixel intensity were analyzed using OpenLab 4.0 (Improvision). Whole cell bleaching of a cotransfected cell was performed to determine the onset of EGFP protein synthesis and maturation. COS-7 cells were grown to 80–90% confluency in 10-cm culture dishes and transfected with the described plasmids. A modification of the subcellular fractionation protocol from was used. 24 h after transfection, the cells were rinsed with cold PBS, trypsinized, and pelleted at 1,500 rpm for 15 min at 4°C. The cells were lysed in homogenization buffer (10 mM Tris, pH 7.4, 15 mM NaCl, 60 mM KCl, 1 mM EDTA, 0.1 mM EGTA, 0.5% Nonidet-P40, and 5% sucrose) containing protease inhibitors (P8340; Sigma-Aldrich) for 10 min on ice, followed by further disruption with 15 strokes in a tightly fitting Dounce homogenizer. The homogenate was centrifuged at 6,000 rpm for 1 min at 4°C to pellet the nuclei. The supernatant was further centrifuged at 10,000 rpm for 10 min at 4°C, and this supernatant was saved as the cytosolic fraction. The nuclear pellet was passed through 5 ml sucrose buffer (10 mM Tris, pH 7.4, 15 mM NaCl, 60 mM KCl, and 10% sucrose) at 3,000 rpm for 5 min at 4°C, washed three times with wash buffer (10 mM Tris, pH 7.4, 15 mM NaCl, and 60 mM KCl), and resuspended in homogenization buffer that had the NaCl adjusted to 0.5 M. After incubation for 30 min at 4°C with rocking to extract the nuclear proteins, the extract was centrifuged at 10,000 rpm for 10 min at 4°C and the supernatant was saved as the nuclear fraction. Protein concentrations were determined for each of the fractions by BCA assay (Pierce Biotechnology) for subsequent SDS-PAGE () and immunoblot analysis () with the indicated antibodies. For coimmunoprecipitation, COS-7 cells were grown to 80–90% confluency in 10-cm culture dishes and transfected with 10 μg Tbx5-HA and 14 μg LMP4-myc. After 24 h, cells were lysed in lysis buffer (25 mM Tris-HCl, 100 mM NaF, 10 mM EGTA, 5 mM EDTA, 250 mM NaCl, 1% NP-40, 50 mM NaPO·HO, 0.5% DOC, and 10 mM ATP) containing protease inhibitors. Lysates were incubated on ice for 20 min followed by centrifugation at 52,000 rpm for 10 min. The supernatant was incubated with anti-myc–conjugated protein A–Sepharose beads (GE Healthcare) overnight at 4°C. The Sepharose beads were washed in lysis buffer, and the bound protein was eluted with SDS buffer, boiled, and analyzed by immunoblotting with the indicated antibodies. COS-7 cells were treated with 2 μM latrunculin A (Sigma-Aldrich) or 5 μM cytochalasin D (Sigma-Aldrich) for 60 min at 37°C. Parallel cultures were treated with the vehicle DMSO as a control. After treatment, the cells were immediately prepared for cell imaging or biochemical analysis. Full-length chicken Tbx5 was cloned into a pcDNA3.1 expression vector containing a HA tetramer tag. Tbx5 was additionally placed as an N-terminal fusion into a modified pEGFP-C1 vector suitable for the Gateway recombination system (Invitrogen). Full-length chicken LMP4 was cloned into pcDNA3.1 containing a myc C-terminal tag. Chicken LMP4 was also recombined as an N-terminal fusion into a modified HcRed-C1 expression vector suitable for the Gateway recombination system. Mouse ENH1 (available from GenBank/EMBL/DDBJ under accession no. ) fragments were cloned from mouse brain cDNA into the pGEX-6P-2 prokaryotic expression vector to create an N-terminal GST fusion protein (GE Healthcare). The PDZ/proline-rich fragment covers amino acids 1–414 and was amplified using forward primer 5′-ACGCGTCGACCATGAGCAACTACAGTGTGTCATTG-3′ and reverse primer 5′-ATAGTTTAGCGGCCGCTCACATGGGGGTCCGCTTGCCCG-3′. The ENH1 LIM 1/2/3 fragment covers amino acids 412–593 and was amplified using forward primer 5′-ACGCGTCGACCATGTGTGCCCACTGCAACCA-3′ and reverse primer 5′-ATAGTTTAGCGGCCGCTGATTTTCAAAAATTCACAGAATGAG-3′. Recombinant mouse ENH1 peptides were expressed in BL21 as described previously (). To identify a region in chicken LMP4 suitable for specific antibody production, a multiprotein sequence alignment (MacVector 7.0; Accelrys) was conducted to compare two closely related subclasses of PDZ-LIM proteins: LMP and ENH proteins. LMP protein sequences from human LMP1 (), rat LMP1 (), and chicken LMP4 () were compared with ENH sequences from human ENH1 (), mouse ENH1 (), and rat ENH1 (). From this alignment, a 17-amino-acid peptide (DPAFAERYAPDKTSTVL) was identified that was conserved in LMP proteins but not in ENH proteins. In addition, based on its predicted antigenicity and hydrophobicity, the peptide was suitable to elicit a good immune response. Peptide synthesis and rabbit immunization was performed by Invitrogen custom antibody services. The final rabbit antiserum was affinity purified on antigen peptide–conjugated columns and tested for specificity (Fig. S1 and supplemental text). Fig. S1 and the accompanying supplemental text describe the production and testing of the LMP4-specific antisera. Fig. S2 shows the calponin control staining in epicardial cells for . Fig. S3 describes the negative control whole cell FRAP. Representative time-lapse videos of cytoplasmic FRAP () and nuclear FRAP () are provided online in Videos 1 and 2, respectively. Online supplemental material is available at .

The GW182 protein is a critical component of cytoplasmic RNP bodies that have been shown to function in mRNA degradation, storage, and, recently, microRNA (miRNA)- and siRNA-based gene silencing (; ; ; ; ; ). GW182 was named for the presence of multiple glycine (G)–tryptophan (W) amino acid pairs in the N-terminal region of a 182-kD protein with a predicted C-terminal RNA recognition motif (RRM). It localizes into cytoplasmic GW bodies (GWBs; ; ) that also contain factors involved in 5′–3′ mRNA decay, including the exonuclease XRN1, decapping enzymes DCP1 and DCP2, and the LSm1–7 decapping activator, pointing to a role for GWBs in regulating mRNA stability (; ; ). These bodies may participate in additional roles in mRNA regulation, as they also contain the m7G cap–binding protein eIF4E and the eIF4E transporter but no other components of translation machinery (; ). Importantly, intact GWBs are required for the functioning of the RNAi pathway in human cells potentially via direct interaction between GW182 (and the related TNRC6B protein) and Argonaute1 (Ago1) and 2 (Ago2; ; ,; ). GWBs are thought to be analogous to cytoplasmic processing bodies (PBs). They are involved in mRNA decapping and 5′–3′ exonucleolytic decay (), and their integrity depends on the presence of nontranslating mRNAs (; ; ). Both PBs and GWBs dissociate when polysomes are stabilized with drugs such as cycloheximide (; ; ). However, despite similar compositions, there are functional differences between GWBs and PBs. GWBs increase in size and number in proliferating cells (), whereas PBs increase in size and number during growth limitation and increased cell density (). GWBs and PBs also differ in their responses to stress, as PBs increase in size and number in response to environmental stress. This is likely caused by decreased translation initiation because this response can be reproduced using a temperature-sensitive allele of Prt1p, a subunit of the eIF3 complex (). In stressed mammalian cells, stalled preinitiation complex mRNAs are first targeted to stress granules (SGs), which may function as triage sites where mRNAs are sorted for future degradation, storage, or reinitiation of translation. Observation of interactions between SGs and GWBs in live cells suggest that transcripts may be exported from SGs to GWBs for degradation (). We have characterized the role of (), the orthologue of the human GW182 gene family. GW localizes to punctate structures in the cytoplasm of embryos and cultured S2 cells. GWBs are electron-dense nonmembrane-bound cytoplasmic foci. These structures are targeted by human GW182 and its paralogues TNRC6B and TNRC6C in cells. Unlike what is seen in some mammalian cells, only some foci colocalize with the previously identified GWB components LSm4, the Xrn1 orthologue Pacman (PCM), and AGO2 (; ; ; ; ). There is a requirement for the zygotic expression of full-length GW during early embryonic nuclear divisions. This suggests a critical role for GWB-based cytoplasmic RNA regulation in beginning with early embryo development. The mutation was isolated in a screen for recessive lethal zygotic mutations on the fourth chromosome and mapped to a region predicted to contain a single gene, CG31992 (). This gene encodes a 143-kD protein containing a C-terminal RRM domain and an N-terminal glycine- and tryptophan-rich region (20% G or W), which are features also found in the human GW182 protein (). There are three human GW-like proteins (). The gene is also proposed to be part of this family, although it lacks an RRM domain (). Although many vertebrate species have up to three GW-related proteins, invertebrates seem to have only a single form (). The mutant allele encodes a 100-kD truncated protein containing the GW-rich region but not the C-terminal RRM domain as a result of a nonsense mutation (). The location of this gene on chromosome four required an alternate approach to confirm the genotype of mutant embryos as a result of the lack of early developmental markers on this chromosome. We confirmed the presence of the mutation in individual embryos by PCR amplification of the region flanking the mutation (). We raised a polyclonal GW antibody that recognized a 160-kD protein (), which is within ∼10% of the predicted molecular mass of 143 kD. This antibody also recognized the 100-kD truncated GW protein in homozygotes. This truncated protein is also present in heterozygous adults, strongly suggesting that it is functionally inactive and has no dominant-negative effects (). Heterozygous / parents produced embryos with disorganized internal structures 90– 130 min after egg deposition (AED; ). allele ( = 200) by PCR. In early embryos, GW localizes to foci surrounding cortical nuclei (). Homozygous mutant embryos failed to cellularize, and DNA, GW, and membrane can be seen forming disorganized aggregates (). The highest relative levels of GW were found during early embryonic development and pupariation (). The presumptive maternal GW contribution to the embryo appears to be depleted by 60–70 min AED followed by an increase in GW levels starting at 80 min AED (). The activation of zygotic transcription was confirmed by Northern blotting. There is a significant maternal contribution of mRNA (). Corresponding to the increase observed in GW protein levels, the relative levels of mRNA increase at 80–90 min AED (). GW localizes to punctate cytoplasmic bodies in embryos () and S2 cells ( and Fig. S1, available at ). In transmission EM sections, GWBs appeared as electron-dense nonmembrane-bound cytoplasmic particles (). Because of these similarities to human GWBs (), we tested the functional conservation between human GW182 and GW. To assay GW in living cells, we created a transgenic cell line expressing a GW-GFP fusion that localized to cytoplasmic foci. GFP alone showed diffuse fluorescence throughout the cell (Fig. S1). Several key proteins found in GWBs/PBs, including PCM, the Xrn1 homologue, AGO2, and a representative of the LSm proteins, LSm4, colocalize with GW (). This colocalization is not caused by aggregation of the GFP tag, as FLAG-AGO2 costains with endogenous GW. The association between AGO2 and GW was further confirmed by coimmunoprecipitation of AGO2 with GW (Fig. S2). Functional conservation with human GWBs is also suggested by the targeting of GFP-GW and RFP fusions of the human GW182 and its paralogues TNRC6B and TNRC6C in S2 cells (). Both human GW182 and GW contain an RRM domain within the C-terminal of the protein (). Concomitant with a requirement for intact RNA for the formation of GWBs and PBs (; ), we have shown a requirement for intact RNA for the formation of GWBs. After RNase treatment, only 15% of cells had localized GWBs compared with 97% of untreated cells (). Syncytial embryos undergo 14 synchronous nuclear cycles (NCs) during early development before they cellularize (). In homozygous embryos, defects in nuclear spacing and morphology were observed beginning at approximately NC10, as they migrate to the embryo cortex. Mutant embryos had fewer cortical nuclei, and these had irregular spacing (). These nuclei had abnormally positioned centrosomes (), and examination of the ultrastructure of 2-h AED mutant embryos showed larger than normal nuclei and an abnormal clearing of the embryo cortex. By 3 h AED, no recognizable nuclei were found, and large multivesicular bodies and homogeneous patches devoid of organelles were seen (). Higher magnification of the homogenous regions showed that they were composed of filamentous elements (), which may represent large tubulin aggregates. Homozygous mutant embryos that do not express full-length GW are extremely fragile as a result of what appears to be abnormal cellularization (). Thus, we examined the localization of chromatin in live embryos expressing histone-GFP, which can be used to track chromatin dynamics after NC10 (Video 1, available at ; ). In homozygous embryos, fewer nuclei reached the cortex at NC10, and the majority of those that did could not successfully complete subsequent mitosis (Video 2). The remaining GFP-labeled chromatin could be seen fusing into large aggregates within the cytoplasm, which is similar to the pattern observed with DNA staining of fixed embryos ( and ). The rapid degradation of internal structures that occurs in homozygous embryos made linking specific effects to the loss of function difficult. Therefore, we interfered with GW function in a localized manner by injecting anti-GW antibody into live embryos. Loss of GW function occurs in a graded manner starting closest to the injection site. When GW antibody was injected into histone-GFP–expressing embryos, the chromosomes failed to successfully separate during mitosis similar to what is seen in / embryos (Video 3, available at ). As the effect of the anti-GW antibody diffused anteriorly, additional nuclei were observed failing to separate with each NC. In both anti-GW–injected and mutant embryos, the chromatin was observed forming ring-shaped patterns that broke apart with time. Additionally, one to two NCs after injection, the nuclei were no longer anchored at the cortex as they moved freely within the embryonic cytoplasm (Video 3). Live embryos expressing GFP fusions that selectively mark the spindles (tubulin), pseudocleavage furrows (actin), or nuclei (nuclear localization sequence) were treated in a similar fashion. The pseudocleavage furrows act as barriers between adjacent spindles and regress during late anaphase and telophase (). These can be monitored by following the actin network that forms apical caps over the cortical nuclei that correspondingly divides with each NC (; ). As each nucleus enters prophase, the centrosomes normally migrate to opposite poles, and the apical actin caps reorganize into the pseudocleavage furrows. Subsequently, the nuclear envelope is broken down, and the spindle poles begin to separate during chromosome separation (; ; ). A tubulin-GFP fusion faithfully marks the localization of the spindles during embryonic nuclear divisions (Video 4, available at ). The defects in tubulin localization induced by anti-GW injection (Video 5) are similar to those detected in fixed mutant embryos by indirect immunofluorescence using antibodies to tubulin or centrosomin (). In both cases, nuclei were often observed with an abnormal number of spindles, which subsequently broke down to form large tubulin aggregates (Video 5). The dynamics of actin reorganization during the cell cycle in wild-type embryos can be seen using an actin-GFP fusion (Video 6). Blocking GW function by antibody injection at NC10 causes a stabilization of actin in the hexagonal pattern that is associated with pseudocleavage furrows beginning at the site of injection (; and Video 7). The stabilized actin configuration was seen even after 30 min following injection ( and Video 7) but eventually breaks down into a large aggregate (Video 7). The number and size of nuclei can be monitored in developing embryos expressing an NLS-GFP fusion (). The effect of the blocking of GW function on nuclear proliferation was assayed by injecting antibody at interphase of NC13 and observing the resulting effects at the time when NC14 would have occurred in wild-type embryos (130 min AED; ). When anti-GW was injected at any point before NC9, significantly fewer nuclei are observed at the embryo periphery (). These nuclei were on average 8–10 times greater in diameter than stage 14 nuclei of control injected embryos (). When anti-GW was injected later, a graded response was observed. In embryos injected at 1 h 40 min AED, three distinct regions of enlarged nuclei were seen with a distinct boundary between nuclei that was eight and four times greater in size as well as between nuclei that was four times and twice the size farther from the site of injection ( and Video 8, available at ). Embryos injected at later time points (1 h 50 min) showed nuclei twice the normal size in the area proximal to the injection point, whereas the diameter and number of nuclei in the anterior and posterior were similar to wild type. Additionally, in these embryos, the posterior pole cells developed normally (). This graded response to a presumptive gradient of anti-GW activity could be correlated to the number of nuclear divisions that elapsed between the time of injection and 130 min AED. A video of a live embryo expressing NLS-GFP injected with anti-GW antibody at NC10 shows that with subsequent three mitotic cycles, a corresponding increase in nuclear size could be observed beginning at the site of injection and progressing anteriorly (Video 8). Finally, because AGO2 and GW colocalize in some GWBs ( and Fig. S2), we also tested the effect of injection of anti-AGO2 antibody using a similar assay (). In all cases ( = 12), the injection of anti-AGO2 at 1 h AED produced an effect similar to the injection of anti-GW at the same time (). Our results confirm that GW is homologous to human GW182 and that GWBs are analogous to human GWBs and yeast PBs. GW localizes to rapidly moving (Video 9, available at ) and electron-dense, nonmembrane-bound cytoplasmic structures (). Colocalization of GW to homologues of known GWB or PB components LSm4, AGO2, and PCM (Xrn1) shows that GWBs are of similar composition to PBs and GWBs. Another similarity between GWBs and PBs is that GWBs also require intact RNA to maintain their integrity (). Functionally, human and GW homologues are all targeted to the same foci when coexpressed in S2 cells (). However, not all GWBs contain the mRNA decay enzymes LSm4 and PCM or AGO2 associated with GWBs or PBs. There is an apparent lack of interdependence in functions of the nonsense-mediated decay, RNAi, and miRNA pathways in S2 cells, as the depletion of proteins involved in one pathway did not affect the function of another (). Thus, the variable composition of GWBs provides evidence that there may be distinct functions for these cytoplasmic structures. It may be possible to discern functionally distinct classes of GWBs by analyzing relative localizations of other mRNA-processing proteins as they become known. There have been several exhaustive screens to identify zygotically transcribed genes that affect precellular embryonic development (; ). Currently, a total of seven genes are thought to be expressed before the cellular blastoderm stage (; ). However, these screens focused on the X chromosome and autosomes two and three, but not four (). We propose that represents an additional zygotically expressed gene required for successful completion of the early embryo development in . The reduction in GW protein observed at 60–70 min AED () suggests that maternally supplied GW is depleted. This would be subsequently replenished by zygotic transcription, as shown by rising mRNA levels beginning at 70–80 min AED (), a time of rapid nuclear division that culminates in the cellularization and subsequent gastrulation steps of embryo development (). Notably, increased levels of GW expression are also observed during pupal development (), which is another time of rapid cell proliferation (). The increase in GW expression during periods of rapid cell division is consistent with elevated GW182 levels observed in proliferating human cells (). The function of GWBs described in mammalian cells suggests a potential role for these structures in development. In many organisms, siRNA and miRNA, which are produced by Dicer-mediated cleavage of longer double-stranded or hairpin RNA precursors, regulate several developmental functions (for review see ). For both siRNA and miRNA activity, the RNA-induced silencing complex (RISC) binds and selectively suppresses or degrades complementary target mRNA (; ; ; ). Several recent studies have identified a link between GWBs and the RNAi pathway. RISC components Ago1–4 localize to GWBs (; ), as do reporter mRNAs targeted for miRNA-mediated translational repression (). In addition, intact GWBs are required for siRNA silencing (; ). The effects of miRNA expression on development were characterized in a screen of 46 embryonically expressed miRNAs. Injection of antisense RNA to block these miRNAs into 30-min AED embryos revealed 25 miRNAs with visible phenotypes affecting a variety of developmental processes. Blocking miR-9 resulted in several severe defects, including nuclear division and migration, actin cytoskeleton formation, and cellularization (). A role for components of the RNAi machinery in the timing of heterochromatin formation and accurate chromosome separation has been reported in (; ) and the trypanosome (). mutants show several defects in early embryogenesis, including defects in centromeres, nuclear division, nuclear migration, and germ cell migration. However, homozygous mutants are, for the most part, fertile and viable (). Therefore, cytoplasmic-based RISC-mediated miRNA may have an effect on the control of timing of protein reorganization associated with cytoskeletal and mitotic events during early development. The putative GW protein orthologue - localizes to cytoplasmic foci with a composition similar to PBs and GWBs and forms complexes with ALG-1 () Dicer-1 and miRNAs. However, and RNAi components , , and function in the heterochronic pathway that regulates developmental timing in many postembryonic cell lineages (; ), while is required in embryogenesis for ventral epithelial closure (). The phenotypes associated with blocking GW function suggest that functional GWBs are required for the completion of nuclear divisions during early embryonic development. These effects, although similar to mutants, are far more severe. Injection of anti-AGO2 antibody into early embryos caused a reduction in number and enlargement in the size of the embryonic nuclei detected by NLS-GFP (). The more severe defects resulting from GW depletion may be caused by the nature of the mutation, which does not completely block protein function (), or may be the consequence of additional functions of GWBs (which are not related to AGO2) and, by extension, RISC function. GW is expressed throughout development and is required for the viability of cultured cells (). Our data suggest that one function of GWBs is to coordinate the regulation of embryonic development in a posttranscriptional fashion. Subsets of eukaryotic mRNAs involved in the same cellular processes are often associated with specific RNA-binding proteins, depending on growth conditions (; ). In one proposed model, RNP particles like GWBs coordinately regulate mRNAs encoding functionally related proteins, which is analogous to the operon-based coordination of prokaryotic gene expression (). Thus, mRNAs with similar cis-elements would be recognized and trafficked by a common RNP to collectively regulate their translation or degradation (; , ; ; ; ). Our data provide evidence that GWBs mirror human GWB composition and function, providing an excellent model for genetic dissection of the potential role of GWBs in regulating mRNAs during development. The open reading frame and 3′ untranslated region were amplified from cDNA LD47780 with primers 5′, CGCAGACGTCTTATGCGTGAAGCCC and 3′, TGCGGACGTCGACATATACATACATATGTATG and were cloned into pZero Blunt (Invitrogen) to make pZB. A GFP-GW fusion was expressed in S2 cells by recloning from pZB into the AatII site of pP(GS[hsEGFP3′]) () to make pPGFP. Approximately 10 cells were transfected with 1.6 μg pPGFP and 0.1 μg pCoHygro using 7 μl Cellfectin (Invitrogen), and stably transformed cells were selected using 300 μg/ml hygromycin. The open reading frame was amplified from the LD22664 cDNA with 5′PCM, CACCATGGGCGTTCCCAAGTTCTTTC and 3′PCM, AGTTGGATGCGGGGAGTCGGG primers and cloned into pENTR/D (Invitrogen) to make pENTR. It was then recombined into pAWR (provided by T. Murphy, Carnegie Institute, Troy, MI) to create a C-terminal RFP fusion under control of the promoter. The LSm4 homologue (CG33677) was amplified from the RE35747 cDNA with 5′LSM, CACCATGCTGCCACTTTC and 3′LSM, CGATCCGAAGAACTATTTCCTATT primers, cloned into pENTR/D, and recombined into pAWR as described above. cDNAs of human GW182 and GW182-related proteins, which were provided by E. Chan (University of Floirda, Gainesville, FL), were also recombined into pAWR as described above. The open reading frame was amplified from the REO4347 cDNA () using the primers 5′-GGGGACAAGTTTGTACAAAAAAGCAGGCTCCATGGGAAAAAAAGATAAGAACA-3′ and 5′-GGGGACCACTTTGTACAAGAAAGCTGGGTCGACAAAGTACATGGGGTT-3′, recombined into pDONR221 (Invitrogen), and recombined into pAWR. Double-stranded RNA was made using primers specific to the 3′ untranslated region of : 5′, RNAi TAATACGACTCACTATAGGGAAGATCAATTACCAGTTCCA and 3′, RNAi TAATACGACTCACTATAGGGACATATACATACATATGTATG, allowing direct synthesis of double-stranded RNA from the PCR product using the Megascript in vitro transcription system (Ambion). The mutant was identified during an ethylmethylsulfonate mutagenesis screen for recessive lethal loci located on chromosome four. This mutation was mapped to the 102C region, and only two nucleotide changes were identified: causing W967stop in and N144I in the N-terminal region of CG1838 (). However, CG1838 contains a conserved proteolytic cleavage site, which would remove N144 from the mature protein (). The HS-GFP-GW strain was generated by transferring pPGFP into the pP(GS[ , hsEGFP3′]) vector () and germline transformation of P( = Δ2−3) 99B/TM6 embryos (). The - strain carries the histone2AvD-GFP fusion (). All other fly strains were obtained from the Bloomington Stock Centre. The 5′ XhoI fragment of pZB encoding the first 1,061 amino acids of GW was subcloned into pRSETA (Invitrogen), and recombinant protein was purified on Ni nitrilotriacetic acid agarose (QIAGEN), repurified by SDS-PAGE, electroeluted from polyacrylamide (), and injected into Hartley guinea pigs (Charles River Laboratories). Western blot analysis confirmed reactivity with the initial 100-kD recombinant protein, the endogenous 160-kD GW protein in embryos and S2 cells, as well as a 200-kD GFP-GW fusion. GW antibody was affinity purified using 100 μg of fusion protein bound to a 1-ml HiTrap -hydroxysuccinimide–activated high performance column (GE Healthcare) and eluted using Immunopure gentle elution buffer (Pierce Chemical Co.). The eluted antibody was concentrated to 15 μg/μl using an ultrafiltration unit (centricon-10; Millipore) in a Tris, pH 8.0, and 50% glycerol solution. Anti-GW serum recognized cytoplasmic foci colocalizing with GFP-GW in stably transformed S2 cells fixed with 2% PFA (), whereas no specific signal was seen with the preimmune serum. embryos were fixed as described previously (), rehydrated in 1× PBS, and treated for 30 min with 10 μg/ml DNase-free RNase (Sigma Aldrich). The following primary antibodies were used: mouse anti–α-tubulin (1:100; Sigma-Aldrich), anti-actin (1:100; Sigma- Aldrich), rabbit anticentrosomin (1:100; a gift from T. Kaufman, Indiana University, Bloomington, IN), and antiphosphotyrosine (1:1,000; Cell Signaling). All secondary antibodies were AlexaFluor-conjugated 488, 546, or 647 (Invitrogen) used at 1:2,000. DNA was stained using PicoGreen (1:1,000; Invitrogen). All imaging was performed at 25°C. Confocal images were obtained using a spinning disk confocal system (Ultraview ERS; PerkinElmer) mated with a camera (Orca AG; Hamamatsu) and a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with a 63× NA 1.4 plan-Apochromat lens. Extracts were prepared in 2.5× SDS gel sample buffer (157 mM Tris-HCL, 0.025% bromophenol blue, 5% SDS, 25% glycerol, and 50 mM DTT), immediately heated to 98°C, and centrifuged for 5 min at 12,000 . Approximately 200 μg of protein per 1 μl of sample buffer (embryos, larvae, and pupae) or one adult per 8 μl SDS sample buffer was loaded in each lane (). Protein loading was standardized using E7 anti–β-tubulin monoclonal antibody (Developmental Studies Hybridoma Bank). Early developmental extracts contained five visually staged embryos in 25 μl of gel sample buffer for each time point, and the equivalent protein from one embryo was loaded per lane. Proteins were fractionated on 6% polyacrylamide gels, transferred to nitrocellulose, and incubated with anti-GW serum (1:1,000) and 1 μg/ml E7 anti–β-tubulin monoclonal antibody. This was followed by HRP-conjugated anti–guinea pig or anti–mouse secondary antibodies (1:50,000; Jackson ImmunoResearch Laboratories) and detected using Super Signal West Pico Chemiluminescent Substrate (Pierce Chemical Co.). Equal amounts of total RNA extracted from staged embryo TRIzol (Invitrogen) were separated on a 1.2% agarose gel (0.67% formaldehyde) and transferred to BrightStar-Plus Membrane (Ambion) using 10× SSC and 120 mJ UV cross-linked for 45 s. Blots were hybridized to digoxygenin- labeled antisense (1:5,000) RNAs that were in vitro transcribed using T7 RNA polymerase (New England Biolabs, Inc.) from the LD47780 cDNA-cut EagI and RpL32 (loading control) RNA probes T3 transcribed from RH03940 cut with EcoRI overnight at 68°C in 3 M urea, 5× SSC, 0.1% (wt/vol) -lauroylsarcosine, 0.02% (wt/vol) SDS, 0.5% milk powder, and 0.2 mg/ml sonicated salmon sperm DNA. The membrane was then washed for 15 min with 0.1× SSC and 0.1% SDS, washed for 15 min with 2× SSC and 0.1% SDS, blocked for 30 min (0.1 M maleic acid, 0.15 M NaCl, 1% acelyated BSA, and 0.1% Tween 20, pH 7.5), and incubated with sheep antidigoxygenin-HRP (1:10,000) for 1 min (Roche). This was followed by two 15-min washes with blocking buffer and detection using North2South chemiluminescent substrate (Pierce Chemical Co.). S2 cells were imaged in cell media (Perbio) in coverglass chambers (Lab-Tek). Visually staged embryos were prepared under Halocarbon 700 oil (Sigma-Aldrich) on coverslips as described previously (), injected with 0.25 ng affinity-purified anti-GW antibody, guinea pig preimmune serum, or affinity-purified rabbit anti-Ago2 (ab5072; Abcam), and diluted in 1× PBS. Approximately 100–150 pl of antibody solution was injected, determined by estimation of the size of the liquid drops (). RNase treatment of the cells expressing GFP-GW was performed as described previously (). Mitochondria were stained using 100 nm Mitotracker red CMXRos (Invitrogen). All imaging was performed at 25°C. Time-lapse confocal images were obtained using a spinning disk confocal system (Ultraview ERS; PerkinElmer) mated to a camera (Orca AG; Hamamatsu) and a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with a 20× NA 0.75 plan-Apochromat lens. 30–40 optical sections at a resolution of 672 × 512 with 2 × 2 binning were collected every 10 s. A maximum projection of each time point was generated, and uncompressed AVI videos were exported using Ultraview software (PerkinElmer). Each video was converted to QuickTime format using QuickTime Pro software (Apple). Still images of GFP-expressing cells and embryos were obtained using a confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.) and software using a 63× NA 1.4 plan-Apochromat and 20× NA 0.75 plan-Apochromat lenses, respectively. embryos were fixed 8–12 h AED using high pressure freezing () and embedded in LR white resin (London Resin Company). 70-nm thin sections were contrast stained with uranyl acetate and incubated with 1:25 anti-GW antibody or 1:25 preimmune serum followed by donkey anti–guinea pig IgG conjugated to 6 nM gold (1:25; Jackson ImmunoResearch Laboratories). Embryos were collected, aged for 1–3 h, dechorionated in 50% bleach, and fixed in heptane (Sigma-Aldrich) saturated with 25% glutaraldehyde (Sigma-Aldrich) in 50 mM sodium cocadylate, pH 7.0, for 20 min at 25°C. Mutant embryos were selected via direct phenotypic observation of nuclear morphology after staining with PicoGreen (Invitrogen), hand devitinellized under heptane, postfixed in 1% osmium tetroxide (EM Sciences), and embedded in Epon resin (). Thin sections were stained with lead citrate and uranyl acetate before sectioning and were imaged using a transmission electron microscope (TEM2000; Philips), digital camera (MegaView III; Soft Imaging System), and analySIS software (Soft Imaging System). The / genotype was confirmed by genomic PCR with the following primers: 5′outside (intron 6), TGTAACAGGCAGAAGGAAGCGTTTCCGACCAT and 3′outside (exon 9), GGCAGTCAATCCTGGCGGGGGACCTCGAGACG followed by a second nested PCR reaction with 5′inside (intron 6), CCATCTGTCCGTATGAACTTCGAG and 3′inside (exon 9), TCCGAAGTCGCGGTACATTGTTGA using 50 μl PCR Supermix (Invitrogen). The stop mutation (TGG to TGA) in disrupts an NcoI recognition sequence and was initially identified by digesting purified PCR products (Qiaquick; QIAGEN) with NcoI. Mutants were verified by DNA sequencing. Approximately 10 S2 cells were transfected with 10 μg of the plasmid HSFLAG-Ago2. 48 h after transfection, cells were heat shocked for 40 min at 37°C and allowed to recover for 40 min at 25°C. Cells were lysed in 2 ml radioimmunoprecipitation buffer (1% sodium deoxycholate, 1% NP-40, 0.2% SDS, 150 mM NaCl, 50 mM Tris, pH 7.4, complete EDTA-free protease inhibitors [Roche], and 1 mM PMSF). The extract was incubated with 6 μl anti-GW antibody for 30 min and incubated for 2 h in the presence of 40 μl protein A–Sepharose beads (GE Healthcare) at 4°C. After washing, bound proteins were eluted with 2× SDS gel sample buffer, fractionated on a 6% low bisacrylamide (118:1) polyacrylamide gel, and transferred to nitrocellulose. Flag-AGO2 was detected with mouse anti-Flag M2 antibody (1:100; Sigma-Aldrich). Video 1 shows chromatin organization during early development of a wild-type embryo injected with guinea pig preimmune serum. Video 2 shows an abnormal pattern of chromosome division in a homozygous mutant expressing histone-GFP. Video 3 shows histone-GFP–expressing embryos after the localized depletion of GW function by antibody injection at the anterior pole during interphase of NC10. Video 4 shows localization of the spindles during early development in living embryos. Video 5 shows anti-GW antibody injection into the posterior pole of tubulin-GFP–expressing embryos at NC10. Video 6 shows the dynamic pattern of actin localization monitored using the actin-binding domain of moesin-GFP expressed in live embryos. Video 7 shows that blocking GW function by anti-GW antibody injection at NC10 into moesin-GFP–expressing embryos leads to stabilization and then breakdown of the cortical actin network. Video 8 shows that injection of anti-GW antibody into embryos expressing NLS-GFP at NC10 causes progressive nuclear enlargement of the posterior pole and regional nuclear enlargement with each subsequent NC. Video 9 presents the visualization of GWBs in living S2 cells. Fig. S1 shows GFP-GW and GFP expression in S2 cells. Fig. S2 shows Flag-AGO2 colocalized and associated with endogenous GW in S2 cells, and Fig. S3 shows RNAi knockdown of mRNA phenocopies of the mutation. Online supplemental material is available at .

Vesicle traffic from the ER to the Golgi complex is mediated by the sequential action of the COPII and COPI coat protein complexes (; ; ). COPII (coat protein II) vesicles form at the transitional ER (tER), which is a specialized subdomain of the ER that is distinct from the rough ER (). These vesicles may tether and fuse to each other (homotypic fusion) to form the vesicular tubular clusters (VTCs), which is a pre-Golgi structure (also called the intermediate compartment) that lies adjacent to the tER (; ; ; ; ). COPI (coat protein I) is then recruited onto VTCs before coated structures move cargo to the Golgi apparatus (; ; ; ). Although components of the ER–Golgi trafficking machinery are highly conserved from yeast to man (), no equivalent to a pre-Golgi compartment has been identified in the yeast . COPII vesicle tethering/fusion has been reconstituted in vitro with yeast membrane fractions (; ). However, in these assays, vesicles released from the ER in vitro are thought to fuse with Golgi membranes (heterotypic fusion). Tethering factors have been proposed to mediate the initial interaction of a transport vesicle with its acceptor compartment and are implicated in maintaining the identity of organelles (). Tethers are either large oligomeric complexes or long coiled-coil proteins that activate or act downstream of the GTPases (Rabs) that regulate membrane traffic (; ; ). Most peripherally associate with membranes and are recruited directly onto the compartment on which they function (). Although many tethers have been identified, the mechanism of vesicle tethering has remained elusive. To address questions concerning the specificity of vesicle traffic and organelle identity, we have chosen to focus our studies on the tether called TRAPP. The 10 subunits of TRAPP were initially identified in the yeast (, ). TRAPP subunits are found in two complexes (TRAPPI and TRAPPII) that act at different stages of membrane traffic. TRAPPI acts in ER–Golgi traffic, whereas TRAPPII is required for traffic through the early endosome (; ). Bet3p, which is the most highly conserved TRAPP component, is 54% identical to its mammalian counterpart and stably localizes to Golgi membranes (; ). In yeast, the temperature-sensitive mutant blocks ER–Golgi traffic at 37°C (). Although mammalian Bet3p (mBet3p) has been shown to act in ER–Golgi traffic after COPII vesicle budding and before Rab1, it is unknown if it participates in VTC biogenesis (). We show that mBet3p is enriched at the tER and adjacent VTCs. Microinjected α-mBet3p results in the accumulation of cargo in structures that colocalize with the COPII coat. Furthermore, the inactivation of mBet3p blocks homotypic COPII vesicle tethering in vitro. These findings imply that mBet3p mediates the biogenesis of VTCs by linking COPII vesicles to each other. Previous studies identified a cytosolic pool of mBet3p (; ). However, the enrichment of this protein to a particular intracellular structure was not demonstrated in these studies. To begin to address the role of mBet3p in VTC biogenesis, we prepared polyclonal antibodies to this TRAPP subunit to determine the intracellular compartment where mBet3p resides. Immunofluorescence microscopy revealed that mBet3p localizes to punctate structures in the perinuclear region of COS-7 cells (). This localization was not observed when antibody was pretreated to a nitrocellulose strip containing immobilized recombinant mBet3p (). In HeLa cells, the localization of mBet3p was predominantly cytosolic. However, when the cytosolic pool of mBet3p was removed by digitonin before fixation, its perinuclear localization was revealed (unpublished data). The perinuclear localization of mBet3p was also observed in several other mammalian cell lines, including BSC-1 () and NRK cells (). mBet3p was largely found adjacent to ERGIC-53, which is a marker for the pre-Golgi compartment (), and COPI (). COPI is a heptameric coat complex that is recruited to pre-Golgi and early Golgi structures (). Interestingly, by immunofluorescence, mBet3p has a punctate appearance rather than a continuous ribbonlike pattern, which is typical of Golgi proteins. Although the Golgi disassembles in the presence of the drug BFA, we found that, surprisingly, the localization of mBet3p was largely resistant to BFA (, top and bottom). The observation that the localization of a component of a COPII tethering complex is resistant to BFA prompted us to examine if mBet3p is associated with the tER. The tER is a BFA-resistant subdomain of the ER that specializes in the formation of COPII vesicles (; ; ). As a marker for the tER we used Sec31p, which is a subunit of the COPII coat ( Sec31p-labeled structures were most abundant in the vicinity of the Golgi apparatus and largely colocalized with mBet3p ( and Fig. S1, available at ). To address the relationship of mBet3p to the tER, we asked if the localization of mBet3p was perturbed by conditions that are known to disrupt or alter tER sites. When cells are treated with the microtubule-disrupting agent nocodazole, the tER disperses and tER sites cluster near Golgi fragments (). Although the cytosolic pool of mBet3p appeared to increase in the presence of nocodazole, mBet3p largely remained associated with the tER (, top) and clustered near the Golgi marker GM130 (, bottom). The colocalization of mBet3p with the tER marker Sec31p is illustrated in Video 1 (available at ), which is a 3D reconstruction of confocal images obtained from serial sections. To determine if the localization of mBet3p is changed by other conditions that are known to alter the localization of the tER, we microinjected Sar1p H79G into cells. Sar1p H79G is the constitutively active GTP-bound form of Sar1p that blocks export from the ER (). In cells expressing Sar1p H79G, ER exit sites have been shown to cluster (). Golgi glycosylation enzymes like mannosidase II (Man II) are recycled back to the ER (), whereas the localization of the golgin GM130 remains unchanged (; ; ). Unlike Man II and GM130, mBet3p always clustered with the ER exit sites (). Together, these findings imply that mBet3p is recruited to membranes at or near the tER. Like Sec31p, its localization is resistant to BFA and sensitive to reagents that block export from the ER or disrupt tER sites. To more precisely determine the localization of mBet3p, immuno-EM was performed. For this analysis, the tER was defined as a cluster of Sec31p-labeled (large gold particles) membranes. Immuno-EM demonstrated that COPII vesicles and COPII budding profiles (, arrow) contained mBet3p (small gold particles). The observed localization of mBet3p to the tER was significant, as quantitation revealed that the labeling density of mBet3p at the tER was 7.6–22 times higher than the labeling density of mBet3p over the ER and other organelles (). Although no significant labeling was observed on Golgi cisternae (), mBet3p was found on tubules with a dense lumen in proximity to multivesicular bodies (MVBs). Some of these tubules were connected to MVBs ( and Fig. S2 A, available at ). This observation is consistent with the finding that there is more than one pool of mBet3p (; ). It is also consistent with our recent finding that the yeast TRAPPII complex, which also contains Bet3p, localizes to the early endosome (). In vitro transport studies have shown that mBet3p acts after COPII vesicle budding, but before the GTPase Rab1 (). However, the role of mBet3p in VTC biogenesis was not addressed in these earlier studies. Homotypic COPII vesicle tethering and fusion have recently been reconstituted in vitro (). Using this in vitro assay, we directly addressed whether mBet3p is required for the interaction of COPII vesicles with each other. For this experiment, COPII vesicle populations that were marked in two different ways were generated from permeabilized cells. One population of vesicles contained Myc-tagged temperature-sensitive vesicular stomatitis virus glycoprotein (ts045 VSV-Myc), and a second population of vesicles contained untagged radiolabeled VSV-G* (). After the vesicles were formed in a first-stage incubation, the permeabilized donor cells were removed and the differently marked vesicle populations were combined and incubated together in a second-stage incubation in the presence of cytosol and an ATP-regenerating system. VSV-G*–containing radiolabeled vesicles that interact with VSV-G–Myc vesicles were coisolated from the supernatant with anti-Myc antibody and detected by autoradiography. When the vesicles fused, heterotrimeric VSV-G that contained both VSV-G–Myc and VSV-G* was formed and detected by immunoprecipitation in the presence of detergent (). Interestingly, α-mBet3p, but not control IgG, inhibited the coisolation of vesicles and their fusion when added to the second-stage incubation (). Furthermore, this inhibition was completely reversed when the antibody was added in the presence of excess recombinant His-mBet3p (Fig. S2 B). In contrast, antibody directed against the ER–Golgi SNARE syntaxin-5 inhibited the fusion of COPII vesicles without dramatically affecting their coisolation (). In support of the proposal that mBet3p mediates homotypic COPII vesicle tethering, Western blot analysis revealed the presence of mBet3p on the reaction product (, bottom), nascent VTCs that also contained VSV-G– Myc (, top). The localization of mBet3p to nascent VTCs was specific, as mBet3p was not present on the untagged control or when vesicle budding was blocked with the GDP-locked form of Sar1 (, bottom). A single-stage transport assay was also performed in the presence of α-mBet3p and vesicle release was monitored from the permeabilized cells. In this experiment, α-mBet3p was added to the cells in vitro before vesicles were formed. Studies in both mammalian cells and yeast have previously demonstrated that blocking the consumption of COPII vesicles increases their release from permeabilized cells in vitro (; ). Consistent with the finding that COPII vesicles fail to fuse when mBet3p is inactivated, we observed that α-mBet3p stimulated the release of vesicles from the cells (). This finding also demonstrates that mBet3p is not required for vesicle budding and is consistent with a previous study that showed α-mBet3p acts after COPII vesicles bud from the ER in vitro (; ). In vitro, α-mBet3p blocks the conversion of VSV-G from an endo H–sensitive ER form to an endo H–resistant Golgi form (). However, the role of mBet3p in vivo has not been addressed. To determine where cargo accumulates when mBet3p is inactivated in vivo, we monitored the trafficking of two different marker proteins, tsO45 VSV-G and the plasma membrane marker CD8. One marker was used to quantitate the trafficking defect when α-mBet3p was injected into cells, and the other marker was used to address where cargo accumulates. Quantitation of tsO45 VSV-G transport in HeLa cells revealed that the inactivation of mBet3p resulted in a nearly 70% decrease in the trafficking of YFP-tagged tsO45 VSV-G to the cell surface (). To address where cargo accumulates when mBet3p is inactivated, cDNA encoding CD8 was microinjected into the nuclei of BSC-1 cells. The CD8 mRNA was then allowed to accumulate in the presence of cycloheximide before α-mBet3p, or control IgG, was microinjected (). After 1 h, the cycloheximide was removed and CD8 was synthesized during a second incubation. Newly synthesized CD8 was then chased out of the ER during a final incubation in the presence of cycloheximide ( . In control cells, CD8 was found at the cell surface and the perinuclear region of the cell (; IgG control). Perinuclear CD8 colocalized with Golgin-84 in IgG injected cells (unpublished data). In the presence of α-mBet3p, Golgin-84 localization was fragmented. This finding is consistent with the observation that α-mBet3p disrupts the Golgi (see the next section). When mBet3p was inactivated, CD8 accumulated in spots that surround the nucleus. These spots largely colocalized with Sec31p, but were adjacent to Golgin-84–containing membranes (), implying that cargo accumulates in membranes that contain the COPII coat. These results are also consistent with previous studies showing that α-mBet3p blocks the trafficking of VSV-G before it is converted to an endo H–resistant form in the Golgi (). Thus, both in vitro and in vivo trafficking studies support the hypothesis that mBet3p is required for the interaction of COPII vesicles with each other. We also determined the consequences of depleting mammalian cells of mBet3p by specifically reducing its expression with siRNA. Two different siRNA duplexes were used for these studies, BetC () and BetA (not depicted). Similar results were obtained with both siRNA duplexes. BetC, but not the control duplex (luciferase), reduced the expression of mBet3p (). The loss of mBet3p did not lead to a reduction of Sec31p or the ER protein ER60 (). However, the localization of Sec31p was dispersed, and its intense perinuclear localization, which marks the tER, was lost (). In addition, pre-Golgi and Golgi compartments were disrupted ( and Fig. S2 C). The localization of β-tubulin, however, remained unaltered (Fig. S2 D). Similar effects on the Golgi were obtained when cells were micro-injected with α-mBet3p (Fig. S3, available at ). EM analysis confirmed that the depletion of mBet3p disrupts Golgi architecture (, compare the Golgi [G] in the top and bottom images), and leads to a dramatic accumulation of vesicles (, arrows). Although tER was difficult to identify by EM in the siRNA-treated mBet3p-depleted cells, highly dilated rough ER was observed (Fig. S4, available at ). Quantification revealed a threefold decrease in the amount of membrane that represents Golgi cisternae. These findings provide in vivo evidence that mBet3p links COPII vesicles to each other at or near the tER, an event that appears to be required for the maintenance of Golgi structure. Although the late stages of ER–Golgi traffic have been visualized by live cell imaging (; ), little is known about the homotypic fusion of COPII vesicles and how the pre-Golgi structure is formed from these vesicles. We provide evidence that mBet3p, the most highly conserved component of the TRAPP complex, plays a pivotal role in these events. Although it was previously shown that the SNARE syntaxin-5 is required for homotypic vesicle fusion (), the findings we report here are the first to link the tether TRAPP to VTC biogenesis. Our results imply that mBet3p is recruited to the tER region to mediate homotypic COPII vesicle tethering. Several lines of evidence support this proposal. First, we show that the localization of mBet3p is perturbed by conditions that are known to alter or disrupt the tER, suggesting that mBet3p may be recruited to membranes by one or more components of the tER. Second, mBet3p is required for homotypic COPII vesicle tethering in vitro and the inactivation of mBet3p in vivo leads to the accumulation of cargo in membranes that colocalize with the COPII subunit Sec31p. Finally, a role for mBet3p in VTC biogenesis is supported by data that shows the loss of mBet3p disrupts Golgi structure and the VTC marker ERGIC-53. Immuno-EM revealed that mBet3p is enriched on more than one membrane, the tER region, and endosomal structures. Surprisingly, although yeast Bet3p was reported to reside on the Golgi (; ), we did not observe significant labeling of mBet3p on Golgi cisternae. In the yeast , tER, Golgi, and endosomal marker proteins are all found on punctate structures (). Yeast Bet3p was originally reported to reside on the Golgi because a significant fraction of the Bet3p containing puncta enlarged in a mutant (), a putative Golgi proliferating mutant (). However, our more recent studies suggest that Sec7p and yeast TRAPPII (which contains Bet3p) may reside on endosomal membranes (). The localization of mBet3p to both the tER region and endosomal structures is consistent with studies in yeast showing that the two TRAPP complexes (TRAPPI and TRAPPII) act at different stages of membrane traffic. Yeast TRAPPI, which contains Bet3p and several other subunits, is required for COPII vesicle tethering (; ). Because it is difficult to morphologically distinguish membranes in the ER–Golgi branch of the yeast secretory pathway, no equivalent to a pre-Golgi compartment has been identified, and COPII vesicles are thought to tether directly to the Golgi. In mammalian cells, COPII vesicles do not fuse directly with the Golgi, instead it is thought they fuse with each other to form VTCs (). The recent finding that the ER–Golgi intermediate compartment is stable (Ben-Takaya et al., 2005; ), however, has raised the possibility that homotypic fusion is not exclusive and COPII vesicles may also heterotypically fuse with stable pre-Golgi membranes. Nonetheless, our in vitro and in vivo findings clearly show that mBet3p is required for the biogenesis of VTCs and Golgi membranes. The high degree of homology between yeast and mammalian Bet3p also raises the possibility that yeast COPII vesicles may undergo homotypic tethering and fusion before they fuse with the Golgi. Elucidation of the mechanism of COPII vesicle tethering in yeast and mammalian cells will be needed to resolve this issue. In the future it will also be important to address how mammalian TRAPP mediates subsequent steps in membrane traffic. Experiments are currently in progress to address these questions. Monoclonal antibody directed against GM130 was purchased from BD Biosciences. CD8 antibody, monoclonal antibodies directed against the COPI coat (CM1A10), the lumenal domain of VSV-G protein, rodent Man II (53FC3), and Golgin-84 were obtained from G. Warren (Yale University, New Haven, CT). Monoclonal antibody directed against ERGIC-53 was obtained from H.-P. Hauri (University of Basel, Basel, Switzerland). Anti–β-tubulin antibody (sc-5274) was purchased from Santa Cruz Biotechnology, Inc. Anti-ER60 antibody was obtained from P. Kim (University of Cincinnati, Cincinnati, OH). Monoclonal antibody directed against GalT (GTL-2) was obtained from D. Shima (Eyetech Research Center, Woburn, MA). Anti-Sec31p antibody () was obtained from F. Gorelick (Yale University, New Haven, CT), and polyclonal anti–Man II antibody was obtained from M. Farquhar (University of California, San Diego, San Diego, CA). Polyclonal rabbit anti-mBet3p antibody was previously described (). Rabbit IgG specific to mBet3p was purified on an Affigel column (Bio-Rad Laboratories) preloaded with His-tagged Bet3p. Bound IgG was eluted with 0.1 M glycine, pH 2.5. For immunofluorescence, fluorescent-labeled secondary antibodies were purchased from Invitrogen. To deplete α-mBet3p of determinants directed against mBet3p, affinity-purified antibody (1 μg/ml in buffer C) was incubated (with agitation) for 2 h with a nitrocellulose strip containing 60 μg of purified His-tagged protein. Antibody was mock treated with a blank nitrocellulose strip as a control. Affinity-purified α-mBet3p, control IgG, or Sar1p was injected into COS-7 or NRK cells as previously described (). The antibody- injected samples were fixed and stained after 2 h, whereas the Sar1p-injected samples were fixed and stained after 6 h. Experiments were performed as previously described (). The VSV-G–Myc–expressing cells and the pulse-radiolabeled VSV-G*–containing cells were permeabilized by scraping the cells. For the first-stage incubation, each reaction was 1,595 μl and contained 352 μl of water, 55 μl of 0.1 M magnesium acetate, 110 μl of an ATP-regenerating system, 33 μl of 1 M Hepes, pH 7.2, 220 μl of 20/18/50 buffer (20 mM Hepes, pH 7.2, 18 mM CaCl, and 50 mM EGTA), 550 μl of rat liver cytosol dialyzed into 25/125 buffer (25 mM Hepes, pH 7.2, and 125 mM potassium acetate), and 275 μl of either semi-intact NRK cell population. Reactions were incubated at 32°C for 30 min (stage I incubation), centrifuged at 4,000 for 1 min, and then centrifuged again at 15,000 for 1 min. The supernatant, which contained the vesicles, was saved. In the stage II incubation, each reaction was 200 μl and contained 72.5 μl of VSV-G–Myc vesicles, 72.5 μl of VSV-G* vesicles, and 55 μl of 25/125 buffer or purified proteins (antibodies, GST, or GST-Sec23p) dissolved in 25/125 buffer. After a 20-min preincubation on ice, the reactions were incubated at 32°C for 60 min (stage II incubation). For coisolation assays, the final intact vesicle suspensions were processed for immunoisolation using biotinylated anti-Myc antibodies and streptavidin–Sepharose. Bound vesicles were analyzed by SDS-PAGE (8%) and autoradiography, and the coisolated VSV-G* was quantified. For heterotrimer assays, the final vesicle suspensions were incubated with 2% Triton X-100 for 20 min and then centrifuged at 100,000 for 30 min. The 100,000 supernatants containing solubilized VSV-G trimers were then processed for immunoprecipitation using biotinylated anti-Myc antibody and streptavidin–Sepharose, and the coprecipitating VSV-G* was quantified. VSV-G transport was assayed as previously described () with several minor exceptions. HeLa cells were first injected with 1.5 mg/ml α-mBet3p or 1.5 mg/ml control IgG; 2 h later, plasmid encoding 0.2 mg/ml YFP-tagged tsO45 VSV-G was injected. Protein transported to the cell surface was detected with a monoclonal antibody to the lumenal domain of VSV-G that was conjugated to Alexa Fluor 647. Total VSV-G was detected by YFP. The extent of VSV-G transport to the cell surface was determined by the ratio of surface to total measured fluorescence, as previously described (). A total of nine cells injected with either α-mBet3p antibody or IgG were quantitated. The trafficking of the plasma membrane marker CD8 was monitored, as previously described (). Cells were injected with 1.5 mg/ml α-mBet3p or 1.5 mg/ml IgG after the CD8 plasmid was injected in the presence of 100 μg/ml cycloheximide. The incubation was then continued for 1 h at 37°C. siRNA oligonucleotides specific for human Bet3p were designed according to the method developed by Dharmacon Research, Inc. Two oligonucleotides, BetA (5′-UCACUCCAAGCAUUACUAAUU-3′) and BetC (5′-GGAGACGGUGUGACAGAAAUU-3′) were chosen because they are specific and target to regions that are conserved between human and monkey Bet3 sequences. Duplex RNA was introduced into COS-7 or HeLa cells, using OligofectAMINE (Invitrogen) according to the manufacturer's instructions. 3 d after transfection, the cells were analyzed. In three independent experiments, the reduction of mBet3p (determined by Western blot analysis) ranged from 74 to 99% using BetA, and from 87 to 99% for BetC. Samples were quantitated using Image 1.61/68 K software (National Institutes of Health, Bethesda, MD). Immuno-EM was performed in NRK cells as previously described (). Estimation of labeling densities was determined on digital images captured using a charge-coupled device camera (Morada; Soft Imaging System). At a magnification of 43,000×, sections were scanned and 118 fields were captured at random. Transitional elements were defined as tubular vesicular clusters found in the proximity of ER membranes labeled for Sec31p (at least two gold particles). The labeling density for mBet3p was calculated by counting the number of 5-nm gold particles falling on or within 15 nm of membranes, and dividing these numbers by the length of the membranes expressed in micrometers. Length was estimated by counting the number of intersections of membranes with the vertical and horizontal lines of a grid at an estimated distance of 110 nm between lines (). HeLa cells were processed for EM analysis as previously described (), with the following two exceptions: cells were scraped after fixation, and the section thickness was 60 nm. Quantitation of Golgi membrane was done directly at the EM. Sections were scanned horizontally in a random fashion at a magnification of 9,900×, and the number of intersections of membrane with a fixed point were counted. For each sample, >900 intersections with the plasma membrane were counted. The number of intersections with stacked Golgi cisternae was divided by the number of intersections with the plasma membrane (or nuclear envelope). A threefold decrease in the ratio of Golgi to plasma membrane (or Golgi to nuclear envelope) was observed in the mBet3p-depleted sample. HeLa cells were used for this analysis because it was easier to obtain better cryosections with these cells than with COS-7 cells. Figure S1 shows a still from Video 1. Fig. S2 shows that the loss of mBet3p blocks COPII vesicle tethering and disrupts Golgi architecture. Fig. S3 shows that microinjected α-mBet3p, but not control IgG, disrupts the pre-Golgi compartment marked by ERGIC-53 in COS-7 cells. Fig. S4 shows the rough ER is highly dilated in siRNA-treated mBet3p-depleted HeLa cells. Video 1 shows the colocalization of mBet3p and Sec31p in nocodazole-treated cells. Online supplemental material is available at .

The production of nitric oxide (NO) by the vascular endothelium is important for cardiovascular homeostasis, as endogenous NO regulates many fundamental cellular processes, including growth, mitochondrial respiration, differentiation, and migration. Endothelial NO synthase (eNOS or NOS3) synthesizes NO in the endothelium lining all blood vessels, and genetic deletion of eNOS causes many cardiovascular phenotypes, including increased blood pressure, impaired angiogenesis, abnormal vascular remodeling, and accelerated atherosclerosis (). eNOS is a peripheral membrane protein that is modified by dual acylation (N-myristoylation and S-palmitoylation), which targets it to specific biological membranes (). All dually acylated proteins, including eNOS, src family members, and certain G-protein α subunits are cotranslationally N-myristoylated on cytoplasmic ribosomes followed by posttranslational cysteine palmitoylation. eNOS is N-myristoylated at glycine-2 and posttranslational S-palmitoylated on cysteines 15 and 26 (; ; ; ). N-myristoylation and S-palmitoylation mediate localization of eNOS to the Golgi complex and cholesterol-rich microdomains of the plasma membranes, including caveolae and lipid rafts (; ; ). Acylation-defective mutants of eNOS that cannot target either domain impair basal and agonist-stimulated NO release (, ). Little is known about the enzymatic mechanisms for dual palmitoylation in mammalian cells. This fatty acid modification is reversible, unlike N-myristoylation, which is permanent (). In the context of eNOS, palmitate turnover is 45 min, whereas myristate turnover occurs with the protein backbone (both ∼20 h; ). Palmitoylation and depalmitoylation of proteins may be regulated by extracellular signals, providing a mechanism for dynamic regulation of protein localization (; ; ). Recently, a new family of acyl transferase enzymes that catalyzed the protein palmitoylation was discovered (). Genetic screens in yeast identified Erf2/4 () and Akr1p () as palmitoyl transferases for yeast Ras2 and casein kinase2 (Yck2). Deletion of Erf2/4 or Ark1 reduces palmitoylation of Ras2 or Yck2, respectively. Erf2/4 or Ark1 share a common region, the Asp-His-His-Cys motif (DHHC), within a cysteine-rich domain (CRD). The DHHC and CRD domains are essential for palmitoyl acyl transferase (PAT) activity (; ). The human homologues of the yeast Erf2–Erf4 complex are DHHC-9 and a Golgi-localized protein designated GCP16. This complex has been shown to palmitoylate H- and N-Ras in vitro (). 23 genes encoding proteins with DHHC-CRD domains have been identified in mouse and human databases (). Some of these proteins are known as Golgi-specific DHHC zinc finger protein (GODZ/DHHC-3; ), the c-Abl–associated protein Abl-philin2 (Aph2/DHHC-16; ), Sertoli cell DHHC protein (SERZ-β/DHHC-7; ), Huntingtin interacting protein 14 (HIP14/DHHC-17; ), and DHHC-15, which palmitoylates the neuronal scaffold protein PSD-95 (). In the present work, we screened the 23 known DHHCs to examine which isoforms can palmitoylate eNOS. We found that five mammalian DHHC proteins (DHHC-2, -3, -7, -8, and -21) palmitoylate eNOS, are present in human umbilical vein endothelial cells (HUVECs), and colocalize with eNOS on the Golgi apparatus. Finally, inhibition of DHHC-21 reduces eNOS palmitoylation, mislocalizes eNOS, and antagonizes NO release from endothelial cells. Human embryonic kidney (HEK) 293 cells were cotransfected with each of the palmitoyl transferase cDNAs together with eNOS and the biosynthetic incorporation of [H]-palmitate into eNOS examined by fluorography. As shown in , only five clones (DHHC-2, -3, -7, -8, and -21) markedly enhanced incorporation of [H]-palmitate into eNOS (, top, see fold increase in label relative to total eNOS), defining them as putative eNOS PATs. The incorporation of palmitate into thioester linkages is sensitive to the strong base hydroxylamine. As shown in , the incorporation of [H]-palmitate into eNOS is reduced by hydroxylamine, demonstrating that this occurs via a thioester linkage similar to the palmitoylation of PSD-95 by DHHC-3, which was recently described (). In addition, mutation in the core DHHC domain of these enzymes (mutation of DHHC to DHHS) reduces the incorporation of [H]-palmitate into eNOS (), consistent with previous reports (; ). Next, we examined the expression of the five candidate eNOS PATs by RT-PCR in different human cell lines, including HUVECs, human aortic smooth muscle cells (HASMCs), HEK 293, lung carcinoma (A549), and prostate carcinoma cells (LnCAP). As shown in , all of tested DHHCs are expressed in the cell lines studied. However, in HUVEC, the major cell type that expresses eNOS, DHHC-21 was the most highly expressed, followed by DHHC-2 and -3. Cysteines 15 and 26 have been identified as the major sites of palmitoylation of eNOS (; ). As displayed in , transfection of DHHC-21, but not -11 (as a control), increases the incorporation of [H]-palmitate into eNOS, an effect eliminated by mutation of cysteines 15 and 26 to serine (C15/26S eNOS). We expressed the HA-tagged DHHC proteins in COS-7 cells and localized them by immunofluorescence microscopy. As shown in , DHHC-21 (middle) colocalized with the Golgi matrix protein GM-130 (left), as did DHHC-2, -3, -7, and -8 (Fig. S1, available at ). Increasing the magnification of these images by 4× (, bottom) depicts clear colocalization of DHHC-21 with GM130 and a lesser amount of protein in the plasma membrane (, arrows). Next, we determined the proportional distribution of these enzymes in sodium carbonate using a discontinuous sucrose gradient. In this method, tightly embedded membrane proteins are buoyant (at the 5–30% sucrose interface in fractions 2–4), whereas soluble proteins remain at the bottom of the gradient (in fractions 7–10). In all the gradients, the distribution of caveolin-1, the coat protein of caveolae, and β-COP, a marker of Golgi and post-Golgi vesicles were used to confirm adequate fractionation. As shown in , the DHHCs and eNOS cofractionated into two pools, light membranes enriched in caveolin-1 and heavy membranes enriched in β-COP and β-actin (not depicted). Densitometric quantification (, right) of protein localization showed that a fraction of all DHHCs sediment as integral membrane proteins, with the greatest proportional amount for DHHC-2 and -21. Next, we examined colocalization of eNOS with DHHC enzymes by immunofluorescence microscopy in transfected cells. COS-7 cells were cotransfected with the cDNAs for eNOS and HA-tagged DHHC-21 and localized with antibodies against eNOS and the HA epitope. As shown in , eNOS and DHHC-21 (other DHHCs are shown in Fig. S2, available at ) colocalized in the Golgi region and to a lesser extent in plasma membrane. Because palmitoylation is a cytoplasmic event and eNOS is a peripheral membrane protein, we determined whether DHHC enzymes and eNOS can interact. As seen in (left), the HA-tagged DHHCs and eNOS were well expressed in total cell lysates prepared from the cells. Immunoprecipitation of HA-tagged DHHC enzymes using an anti-HA monoclonal antibody detected the coassociation with eNOS (, right, third to seventh lanes) with the DHHCs, whereas no immunoreactive protein was observed after immunoprecipitation of lysates from cells expressing eNOS alone (, right, first lane). Previously, we targeted eNOS to the Golgi versus the plasma membrane using different domains of syntaxin-3 (; ). To determine the importance of localization on the interaction between eNOS and DHHC enzymes, we cotransfected Golgi-targeted eNOS (eNOS S17) or plasma membrane–targeted eNOS (eNOS S25) and DHHC-3, -7, and -21. As shown in , the interaction between the Golgi-targeted eNOS and the DHHC enzymes was stronger than the plasma membrane–targeted isoform, consistent with the majority of DHHCs and eNOS present on the Golgi of transfected cells (). Next, we assessed whether the phosphorylation state could affect the interaction of eNOS with DHHCs. To address this question, DHHCs were cotransfected with wild-type (WT) eNOS and two eNOS phosphomutants, eNOS S1179A and S1179D. Serine 1179 is a key phosphorylation site for several kinases, including Akt, AMP kinase, and cAMP protein kinase (), and phosphorylation of this site is associated with eNOS activation. Mutation of S1179 to D renders eNOS constitutively active, whereas S1179 to A is less activated (; ; ). As seen in , in all cases, WT and phosphomutants were associated with DHHC enzymes, indicating that phosphorylation of this particular residue was not critical for the interaction of eNOS with DHHC-3 and -21. To examine whether the localization of eNOS is important for the interaction of eNOS with DHHCs, COS cells were transfected with WT and different mutants of eNOS that influence eNOS acylation and targeting (). Cells were transfected with WT eNOS (N-myristoylated and S-palmitoylated, Golgi, and plasmalemma targeted), G2A eNOS (neither N-myristoylated nor S-palmitoylated and cytosolic), C15/26S eNOS (N-myristoylated but not S-palmitoylated; diffuse perinuclear localization), or L2S eNOS (N-myristoylated but not palmitoylated because of the mutation of the intervening leucines between the two palmitoylation sites, C15 and C26; diffuse perinuclear pattern; , ; ) with DHHC-3–HA. As shown in (left), WT eNOS is primarily Golgi targeted and G2A eNOS is diffusely distributed throughout the cells, whereas the palmitoylation-deficient mutants, C15/26S and L2S eNOS, were retained in the perinuclear region but are more diffusely distributed compared with WT eNOS. WT eNOS clearly colocalized with HA-tagged DHHC-3, whereas G2A eNOS did not, and the palmitoylation mutants (C15/26S and L2S eNOS) had a partially overlapping perinuclear pattern. Next, we performed coimmunoprecipitation experiments to determine the role of acylation and subcellular localization on the coassociation of eNOS with DHHC. The eNOS constructs and DHHC-3 or -21 were equally expressed based on Western blotting of the proteins in cell lysates (, left), and immunoprecipitation of DHHC-3–HA resulted in ample recovery (, middle) of DHHC-3 and -21. Interestingly, immunoprecipitation of DHHC-3 and -21 resulted in the coassociation with WT eNOS but not with the other acylation-defective eNOS mutants (, right). These results indicate that the fatty acylation and targeting are important for the interaction between eNOS and the DHHC-PATs. Finally, we examined whether the DHHCs can influence eNOS function by measuring NO release (as nitrite after 24 h of accumulation). Cotransfection of COS-7 cells with WT eNOS and DHHC-3, -7, and -21 increased the basal accumulation of NO into the media (). However, this effect was eliminated in cells cotransfected with acylation-defective G2A or C15/26S eNOS (). Although several DHHC enzymes can palmitoylate eNOS, DHHC-21 mRNA was the most highly expressed in endothelial cells, and the protein localized on the Golgi and interacted with eNOS. To determine the role of DHHC-21 in palmitoylating eNOS in endothelial cells, an RNAi approach was used. As shown in , transfection of endothelial cells with DHHC-21 siRNA duplex RNAs (at 12.5 nM) reduced the expression of DHHC-21 (via quantitative RT-PCR) expression in a time-dependent manner, with maximal reduction at 48 h, where nonsilencing RNA as a control was ineffective. The reduction in DHHC-21 mRNA levels resulted in a marked diminution in the incorporation of [H]-palmitate into eNOS by ∼60% (). Because eNOS palmitoylation is crucial for the perinuclear targeting of eNOS onto the Golgi, we performed an immunofluorescence analysis of eNOS in DHHC-21–depleted endothelial cells. As shown in (middle), knock down of DHHC-21 resulted in a more diffuse, perinuclear pattern of eNOS immunoreactivity compared with cells treated with nonsilencing RNA (left). In addition, treatment of endothelial cells with 2-bromopalmitate, a substrate-based inhibitor of palmitoylation, resulted in a similar but more extensive mislocalization of eNOS (, right). We next examined the effect of DHHC-21 knockdown on basal and agonist-stimulated NO release from endothelial cells. In endothelial cells, increases in cytoplasmic calcium activate calmodulin, which binds to the canonical CaM binding domain in eNOS and serves as an allosteric regulator of electron flux through eNOS, leading to a burst of NO release (; ; ). Reduction of DHHC-21, but not treatment with nonsilencing RNA, caused a slight decrease the basal accumulation of NO into the media (, left). More important, reduction in DHHC-21 reduced the release of NO stimulated by both ionomycin (middle) and ATP (right), indicating the importance of eNOS palmitoylation via DHHC-21 for eNOS activation. To examine the specificity of DHHC-21 knockdown on eNOS function, we also reduced the levels of DHHC-3 with RNAi. As seen in , transfection of endothelial cells with DHHC-3 siRNA (at 12.5 nM) reduced the expression of DHHC-3 (via quantitative RT-PCR) expression where nonsilencing RNA as a control was ineffective. However, under these conditions, basal and ATP-stimulated NO was not affected (). The most salient feature of this paper is the identification of five DHHC-PAT cDNAs that can palmitoylate the dually acylated, peripheral membrane protein eNOS in vivo. DHHC-2, -3, -7, -8, and -21 colocalize with GM-130, a peripheral membrane protein of the Golgi in transfected cells, suggesting that palmitoylation can occur on the cytoplasmic aspect of the Golgi complex. The aforementioned DHHC-PATs also colocalize and associate with eNOS, providing evidence for the compartmentalization of protein palmitoylation. Most critical, reduction of DHHC-21 in endothelial cells diminishes eNOS palmitoylation, mislocalizes eNOS, and impairs agonist-stimulated NO release. Although cysteine palmitoylation has long been recognized as important for protein trafficking and function, enzymes responsible for this have been elusive. Genetic and biochemical studies in yeast initially identified that two genes containing CRDs are essential for palmitoylation of Ras2 and casein kinase (; ). This led to the discovery of mammalian DHHC genes and identification of their specific substrates. Out of the 23 independent mammalian genes encoding DHHC-containing proteins, only a few enzyme substrates have been identified. DHHC-2, -3, -7, and -15 can palmitoylate PSD-95 (), DHHC-3 (GODZ; ) can palmitoylate the GABA-A receptor (), DHHC-17 can palmitoylate several neuronal substrates, including SNAP-25 and PSD-95 (), and DHHC-9 can palmitoylate H- and N-Ras (). In the present study, only DHHC-2, -3, -7, -8, and -21 significantly increased the incorporation of palmitate into eNOS, suggesting some degree of substrate specificity of the DHHCs. As DHHC-2, -3, and -7 can palmitoylate the nonmyristoylated proteins PSD95 and GαS, these candidates are less likely to be specific N-myristoyl–requiring PATs (). RNAi-mediated knockdown of endogenous DHHC-21 in endothelial cells reduces the incorporation of [H]-palmitate into eNOS, mislocalizes eNOS, and reduces agonist-stimulated NO release, whereas knock down of DHHC-3 did not reduce NO release, suggesting that although at least five DHHCs can significantly palmitoylate eNOS in transfected cells, DHHC-21 may be a more specific eNOS PAT. The reasons for the apparent specificity of DHHC-21 toward eNOS in endothelial cells are not known, but it may be due to the amount of DHHC-21 protein relative to other DHHCs expressed in endothelial cells or to the posttranslational modification of DHHCs that may determine its substrate specificity or regulate its activity in a cell-specific context. In addition, we cannot rule the possibility that other DHHCs palmitoylate eNOS in endothelial cells, as a substantial knock down of DHHC-21 reduced palmitoylation by only 50% and NO release by 70%. Additional experiments using isoform-selective antibodies and recombinant DHHC proteins with various substrates will help deconvolve DHHC substrate specificity and kinetics in more detail. Other interesting findings are that DHHC-2, -3, -7, -8, and -21 all enhance the transfer of palmitate into eNOS and are strongly colocalized with GM130, a Golgi matrix protein localized to the cytoplasmic face of the Golgi, consistent with our previous suggestion that the Golgi is the site for eNOS palmitoylation (). Palmitoylation of substrates on the Golgi would stabilize protein association via kinetic trapping () and allow for protein localization either on the Golgi or movement to the plasma membrane, as recently confirmed for Ras (). Supporting this are data showing that DHHC-9, an H- and N-Ras PAT, colocalizes and interacts with GCP16, which is a Golgi-localized cosubunit () that appears necessary for DHHC-9 activity toward Ras and movement of Ras to the plasma membrane. For most dually acylated proteins, N-myristoylation is a cotranslational modification that favors posttranslational cysteine palmitoylation. For eNOS, mutation of the palmitoylation sites does not affect N-myristoylation but blocks tight perinuclear targeting and NO release from cells (, , 1997). The mistargeting and dysfunction of these mutants may reflect a structural change or lack of palmitoylation; however, our RNAi experiments, which show that reduction in DHHC-21 decreases eNOS palmitoylation and impairs targeting, indicate that palmitoylation of eNOS is crucial for proper targeting and function. The discovery of DHHC-PATs in vascular cells represents a new, exciting area. Many proteins involved in signal transduction are palmitoylated, and this fatty acid modification is often necessary for function. Understanding the tissue and cellular distribution of DHHCs and their substrates is critical for identification of PAT inhibitors that may be therapeutically useful. Our data showing that DHHC-21 is a major eNOS PAT suggests that antagonizing this enzyme may be useful for reducing NO-dependent changes in vascular permeability and tumor angiogenesis. Mouse monoclonal antibodies against caveolin-1, GM 130, and eNOS were provided by BD Biosciences. Rabbit polyclonal antibody anti–β-COP was purchased from Affinity BioReagents, Inc. Mouse monoclonal against β-actin antibody was provided by Sigma-Aldrich. Rat monoclonal antibody anti-HA high affinity (3F10) was purchased from Roche. Rabbit polyclonal antibody against phospho-eNOS (Ser 1179) was obtained from Zymed Laboratories. Goat anti–mouse IRDye 800–conjugated antibody was purchased from Rockland. Goat anti–rabbit Alexa fluor 680 antibody was provided by Invitrogen, and goat anti–rat Texas red dye–conjugated antibody was purchased from Jackson ImmunoResearch Laboratories. The chemical products were provided by Sigma-Aldrich. HEK 293, COS-7, HASMC, A 459, LnCap, and EA.hy.926 cells were grown in high-glucose DME (Invitrogen) supplemented with FBS (Hyclone), penicillin-streptomycin (Sigma-Aldrich), and HAT (EA.hy.926 only; Sigma-Aldrich) at 37°C in a humidified atmosphere of 5% CO. HUVECs were grown in EGM-2 media (Clonetics). cDNAs encoding DHHC proteins were cloned in PEF-Bos-HA (BD Biosciences) as previously described (). WT, C15/26S, G2A, and L2S eNOS cDNAs were constructed and subcloned into the mammalian expression vector pcDNA3 (Invitrogen) as previously described (). S17, S25, S1179A, and S1179D eNOS cDNAs were constructed and subcloned in pcDNA3 vector (Invitrogen) as previously described (, ; ). The sequences of the PCR fragments cloned were verified by DNA sequencing. For cell transfection, semiconfluent (60%) COS-7 and HEK 293 cells were grown in 6-well plates and transfected with the different plasmids using Lipofectamine 2000. A β-gal plasmid was used to normalize DNA quantities. COS-7 and EA.hy.926 cells grown on coverslips were fixed with 4% PFA for 5 min at room temperature and rinsed twice with PBS. The cells were then permeabilized with 0.1% Triton X-100 for 10 min, washed twice with PBS, and incubated with blocking solution (5% normal goat serum in PBS) for 45 min at room temperature. Next, the cells were incubated with the primary antibodies (diluted 1:500) overnight at 4°C and washed twice with blocking solution, followed by a 45-min incubation with fluorophore-conjugated secondary antibody (FITC or TRITC; diluted 1:250) at room temperature. The coverslips were then mounted on glass slides with Gelvatol/DAPI (Sigma-Aldrich) and analyzed with an epifluorescence microscope (Axiovert; Carl Zeiss MicroImaging, Inc.) with a 63× objective. Images were acquired using a charge-coupled device camera (Axio; Carl Zeiss MicroImaging, Inc.). Analysis of different images was performed using Openlab software (Improvision) after subtracting background. 48 h after transfection of eNOS and DHHC plasmids into COS-7 cells, the media was removed and the cells were supplemented with serum-free DME for 24 h. The media was then processed for the measurement of nitrite (NO ) by a NO-specific chemiluminescence analyzer (Sievers) as described previously (). The same cells were then incubated with fresh, serum-free DME for 30 min to calculate the preagonist nitrite accumulation. Subsequently, the cells were incubated with 1 μM ionomicin for another 30 min to allow postagonist nitrite accumulation. The medium was then harvested, and nitrite accumulation was measured. In other experiments, DHHC-21 and -3 were silenced after treatment with specific siRNAs for 48 h, and EA.hy.926 cells were analyzed for basal and stimulated NO release as described. Transfected COS-7 cells (150-mm dish) were washed twice with PBS and scrapped into 2 ml of ice-cold 500 mM sodium carbonate, pH 11, supplemented with 1 mg/ml protease inhibitor cocktail (Roche), Dounce homogenized, and sonicated (three 20-s bursts at 30% of maximal power). The homogenate was then adjusted to 42.5% sucrose by the addition of 2 ml 85% sucrose prepared in MBS (25 mM MES, pH 6.5, and 0.5 M NaCl) and placed at the bottom of an ultracentrifuge tube. A 5–30% discontinuous sucrose gradient was formed (3 ml of 5% sucrose and 5 ml of 30% sucrose, both in MES containing 250 mM sodium carbonate) and centrifuged at 35,000 rpm for 18 h in a rotor (SW40; Beckman Coulter). Gradient fractions (1 ml) were collected from the top of the tube, and 50 μl of each fraction (1–12) was used for Western blotting analysis as described previously (). The percentage of total proteins in different fractions was determined by densitometry and plotted as percentage of total protein (NIH program). EA.hy.926 and COS-7 cells were lysed in ice-cold buffer containing 50 mM Tris-HCl, pH 7.5, 10% glycerol, 125 mM NaCl, 1% NP-40, 5.3 mM NaF, 1.5 mM NaP, 1mM orthovanadate, 1 mg/ml protease inhibitor cocktail (Roche), and 0.25 mg/ml AEBSF (Roche). Cell lysates were rotated at 4°C for 30 min before the insoluble material was removed by centrifugation at 12,000 for 10 min. After normalizing for equal protein concentration, lysates were precleared by incubation with protein G–agarose for 45 min at 4°C with rocking. Precleared samples were then immunoprecipitated with anti-HA and anti-eNOS antibodies. Proteins in both the cell lysates and immunoprecipitates were heated in SDS sample buffer before separation by SDS-PAGE. After overnight transfer of the proteins onto nitrocellulose membranes, Western blots were performed using the antibodies described above (see Chemicals and antibodies). Transfected HEK 293 cells were preincubated for 30 min in serum-free DME with 10 mg/ml of fatty acid–free BSA (Sigma-Aldrich). Cells were then labeled with 0.5 mCi/ml [H]-palmitic acid (PerkinElmer) for 4 h in the preincubation medium. Cells were washed with PBS, scraped with SDS-PAGE sample buffer (62.5 mM Tris-HCl, pH 6.8, 10% glycerol, 2% SDS, 0,001% bromophenol blue, and 10 mM DTT), and boiled for 2 min. In other experiments, EA.hy.926 cells were solubilized by incubation in 1 ml of lysis buffer containing 50 mM Tris-HCl, pH 7.5, 10% glycerol, 125 mM NaCl, 1% NP-40, 5.3 mM NaF, 1.5 mM NaP, 1 mM orthovanadate, 1 mg/ml protease inhibitor cocktail, and 0.25 mg/ml AEBSF and subjected to eNOS immunoprecipitation as described above (see Immunoprecipitation and immunoblotting). For fluorography, protein samples were separated by SDS-PAGE. Gels were treated with Amplify (GE Healthcare) for 30 min, dried under vacuum, and exposed to a film (Biomax MS; Kodak). Autoradiographs were scanned with a scanner (Canon) and quantified using the IMAGE J (NIH) program. DHHC-21 (available from GenBank/EMBL/DDBJ under accession no. ) DNA and DHHC-3 (accession no. ) target sequences were designed by QIAGEN (HP guaranteed siRNA). In all experiments, we used the same concentration of each of the four siRNAs (12.5 nM). The target sequences for DHHC-21 were 5′-CAGGCAGTTATAAGATTGCAA-3′ (siRNA-1), 5′-CTAGTATAACTAGATAGTATA-3′ (siRNA-2), 5′-TAGCTAGTGTTAGGAAGTGAA-3′ (siRNA-3), and 5′-CACCTTCTTATAGTATAGGTA-3′ (siRNA-4). The target sequences for DHHC-3 were 5′-ACGGGAATAGAACAATTGAAA-3′ (siRNA-1), 5′AACATTGAGCGGAAACCAGAA-3′ (siRNA-2), 5′-AAAGGAAATGCCACTAAAGAA-3′ (siRNA-3), and 5′-CTACGTGTATAGCATCATCAA (siRNA-4). Our nonsilencing DNA target was 5′-AATTCTCCGAACGTGTCACGT-3′. siRNA duplexes were formed according to the manufacturer's protocol. For cell transfection, EA.hy.926 cells were grown to 30% of confluence in 6-well plates. 150 μl of Opti-MEM-1 medium (Invitrogen) was incubated for 5 min with siRNA sequences (50 nM final concentration). In addition, 150 μl of Opti-MEM-1 was incubated with 5 μl of Oligofectamine (Invitrogen) for 5 min. The two solutions were combined and incubated at room temperature for 30 min and then added to cells (1.5 ml of total volume) for 6 h. After transfection, cells were supplemented with 1.5 ml of DME with 10% FBS and antibiotics. Total RNA was extracted with Trizol reagent (Invitrogen). cDNA was synthesized from 5 μg using SuperScript first-strand synthesis system for RT-PCR (Invitrogen). A 1-μl aliquot of the reverse-transcription reaction was then used for subsequent PCR amplification with specific primers. Each 25-μl PCR contained 1 μl of the reverse-transcription reaction, 1 mM dNTPs (Roche), 20 pmol of each primer, and 1.25 U of DNA polymerase (QIAGEN). The primers sequences used were as follows: GAPDH (available from GenBank/EMBL/DDBJ under accession no. ), 5′-CCACCCATGGCCAAATTCCATGGCA-3′ and 5′-TCTAGACGGCAGGTCAGGTCCACC-3′; DHHC-2 (accession no. ), 5′-CGCCATCCAGCTGTGCATAGTG-3′ and 5′-GAGCAGTGATGGCAGCGATCTG-3′; DHHC-3 (accession no. ), 5′-TGTTTGTAAGCGGTGCATTCGG-3′ and 5′-TTGGTCTGGCGTGGCAAAGG-3′; DHHC-7 (accession no. ), 5′-TGCGATGGGAAGGGATGAAGTC-3′ and 5′-GGCGTTTGGCTTCTTCGTGTG-3′; DHHC-8 (accession no. ), 5′-TCAAACCCGCCAAGTACATCCC-3′ and 5′-ACGCCCGATGCAGTTGTTGAC-3′; and DHHC-21 (accession no. ) 5′-AAGCGTTCCCATCACTGCAGC-3 ′and 5′-GAACTCGCAGTGGTTGCCTCTG-3′. Each cycle of PCR consisted of denaturation at 90°C for 1 min, primer annealing at 60°C for 1 min, and primer extension at 72°C for 2 min. PCR products were separated on a 2% agarose gel and stained with ethidium bromide. Quantitative real-time PCR was performed by using iQ SYBR green supermix on iCycler real-time detection system (Bio-Rad Laboratories). The results are expressed as mean ± SD. Statistical comparisons between groups were done by the test, using the Statgraphics Plus 5.0 program (Statistical Graphics Corp.). Fig. S1 shows the colocalization between DHHC-CRD enzymes (DHHC-2, -3, -7, and -8) and GM-130, a Golgi marker. Fig. S2 demonstrates the colocalization between eNOS and different acyl transferases (DHHC-2, -3, -7, and -8). Online supplemental material is available at .

The mitochondrion is central for pathways of ATP production, metabolite synthesis and degradation, lipid metabolism, and iron–sulfur cluster assembly (). Mitochondrial dysfunction contributes to a broad range of neural and muscular diseases, including the X-linked disease Barth syndrome (BTHS; , , 2004). BTHS is characterized by cardiac and skeletal myopathies, delayed growth until puberty, and cyclic neutropenia. The disease presents in infants and, if undiagnosed, is frequently fatal because of cardiac failure or sepsis. The human () gene, located on Xq28 and expressed at high levels in cardiac and skeletal muscle, was recognized as mutated in BTHS patients (). To date, ∼28 different mutations resulting in single amino-acid changes in the Taz protein have been identified in BTHS patients (a comprehensive database of mutations is available on the Barth Syndrome Foundation website []). Aside from mutations resulting in either complete loss of Taz protein expression or expression of a severely truncated Taz, a biochemical explanation for the defect associated with any identified BTHS point mutation has not been provided. Bioinformatics studies showed that tafazzin was similar to acyltransferases, suggesting that BTHS might be caused by an acyltransferase deficiency (). Analysis of fibroblasts derived from BTHS patients demonstrated decreased steady-state levels of the mitochondrial-specific phospholipid cardiolipin (CL), although the biosynthetic rate of CL was normal (). CL is hypothesized to obtain its final composition of fatty acyl groups via a remodeling process (). According to this model, the final step in CL biosynthesis occurs when newly synthesized CL is deacylated to form monolyso-CL (MLCL) and subsequently reacylated with polyunsaturated fatty acyl chains, forming mature CL. Because the CL contained in patient samples was deficient in the mature tetralinoleoyl form of CL, which is the predominant form in normal cardiac muscle (), the defect associated with BTHS was suggested to occur during the process of CL remodeling. Yeast contains an orthologue of and has proven effective as a BTHS model (; ; ). Yeast lacking () arrest growth at 37°C in ethanol media and have decreased CL content. Importantly, as observed in patient samples, the predominant mature CL acyl species of wild-type (wt) cells (C18:1 and C16:1) are replaced with immature, saturated fatty acids. In addition, MLCL, which is the predicted intermediate in the remodeling cycle, accumulates in the yeast strain. Epitope-tagged Taz1p constructs have been localized to mitochondria (; ) and suggested to reside in the outer membrane (OM), facing the IMS (). However, this localization has not been confirmed for endogenous Taz1p and is difficult to reconcile with CL enrichment in the inner membrane. Interestingly, a recent study in the yeast suggests that Taz1p might additionally function as a lyso-phosphatidylcholine (PC) acyltransferase (). In fact, analyses of yeast and samples derived from BTHS patients have shown that PC, in addition to CL, is altered (; ; ). Therefore, the true functions of Taz1p, and, thus, the molecular basis for the pathologies observed in BTHS patients, are at present unknown. In this study, yeast was used to determine if Taz1p does localize to mitochondria; and, if so, in which submitochondrial compartment it resides. The detailed subcellular and submitochondrial localization of Taz1p presented herein shed new insight into the mechanism of Taz1p function. Moreover, a group of authentic BTHS point mutations that occur in an identified membrane anchor of Taz1p are characterized and provide the first molecular explanations for any of the numerous mutations identified to date that are linked to this important human disease. Strikingly, mutations in one membrane anchor result in two distinct biochemical fates for the characterized mutant tafazzins, which are all, nonetheless, nonfunctional. Specifically, for three of the characterized mutants, nonfunctional Taz1p is mislocalized to the matrix side of the inner membrane, indicating that its target lipid localizes to the IMS-sided leaflets of the membrane. Surprisingly, the fourth characterized BTHS Taz1p mutant localized appropriately within the mitochondrion, but, presumably because of an altered association with membranes, assembled into aberrant complexes. Thus, proper Taz1p sorting and assembly is critical for Taz1p activity, and the defect associated with a cluster of BTHS patients is caused by the missorting or misassembly of the mutated Taz1p. To gain insight into Taz1p functions, we sought to confirm the mitochondrial localization of endogenous Taz1p and, if endogenous Taz1p is a mitochondrial resident, to determine its submitochondrial localization. To this end, we raised a polyclonal antiserum in rabbits against the recombinant, full-length yeast HisTaz1 protein (Fig. S1, available at ). The specificity of the resulting antiserum was confirmed by immunoblot analysis of whole-cell extracts derived from a wt strain and an isogenic strain in which the endogenous gene was deleted (). The polyclonal antisera detected Taz1p in wt, but not , extracts at the predicted molecular mass of 44 kD (Fig. S1). To determine the subcellular localization of Taz1p, a series of differential centrifugations were performed on dounced homogenates derived from the wt strain, and the resulting fractions were analyzed by immunoblotting with antisera specific for assorted organellar markers (). Taz1p did not cofractionate with the ER marker Sec62p, the ribosomal marker α-nascent polypeptide–associated chain, or the cytosolic marker hexokinase. Based on cofractionation with the ADP/ATP carrier (AAC), Taz1p was demonstrated to be a mitochondrial resident. Additionally, we determined the subcellular localization using immunoelectron microscopy. In representative sections (), Taz1p localized to mitochondria, associating with the inner membrane (IM; , arrowheads) and, occasionally, with the OM (, arrows). Nuclei and cytoplasmic matrix were devoid of significant labeling. In the strain, Taz1p was not detected (not depicted). Because Taz1p is hypothesized to function as an acyltransferase, we investigated the nature, if any, of the membrane association of Taz1p with wt mitochondria. Sonication (Fig. ), the matrix (Mas1p), and proteins that are integrally or peripherally associated with either the mitochondrial OM or IM. That Taz1p remained largely in the pellet fraction after sonication provides evidence that it is membrane associated, but does not reveal the nature of this membrane association. To determine if Taz1p associates with mitochondrial membranes peripherally through electrostatic interactions, mitochondria or mitoplasts, which were generated by placing mitochondria in hypoosmotic conditions resulting in OM rupture (a process termed osmotic shock), were washed with either 1 M KCl or 0.5 M NaCl (Fig. S2). was released after osmotic shock, independent of the addition of high salt. In contrast, cytochrome , which is known to be peripherally anchored to the IM through electrostatic interactions with CL and/or phosphatidylglycerol (), was only released after a high-salt wash of mitoplasts, but not of intact mitochondria. Taz1p, like the integral membrane protein AAC, remained in the pellet fraction after every tested treatment. To distinguish between a peripheral and an integral membrane association, mitochondria were incubated with 0.1 M NaCO at increasing pH; integral membrane proteins remain in the pellet after centrifugation, whereas peripheral membrane and soluble proteins release into the supernatant. Taz1p and Tim23p fractionated similarly, suggesting that Taz1p may be an integral membrane protein (). Importantly, the observation that the peripherally associated cytochrome was released at every tested pH further confirms the conclusion that Taz1p is not a peripheral membrane protein. To gain insight as to which mitochondrial membrane Taz1p associates with, we used a fractionation technique that allows the separation of IM, OM, and so-called contact sites, which are areas where the IM and OM are connected (). Sonicated submitochondrial particles were separated over a linear sucrose gradient, fractions were collected from bottom to top (, fraction 1 and 16, respectively), and equal amounts of protein derived from each fraction were analyzed by immunoblot (). ), contact sites (intermediate density; contains detectable IM and OM markers), and OM (light density; enriched in OM45 [], porin, and Tom70p) were identified after quantitation of the immunoblots. Interestingly, Taz1p was present in each of the three peaks, indicating that it is localized in all three mitochondrial membrane compartments. Therefore, Taz1p is a new member of an emerging class of mitochondrial-resident proteins that have a dual localization to the IM and OM. Because Taz1p is a nonperipherally associated membrane protein, we used a variety of TM prediction programs to identify potential membrane-spanning domains (Table S1, available at ). The programs did not predict the same regions, but two stretches, 26–46 and 215–232, were consistently predicted in the 381–amino acid protein. Collectively, these different prediction methods suggest that Taz1p may contain from zero to two membrane-spanning domains. To decipher which, if any, of these potential TM domains exist, three Taz1p constructs were generated with an epitope tag placed at either termini (MycTaz and TazHA for N- or C-terminal–tagged Taz1p, respectively) or between the two putative TM domains (TazMycTaz; ). After transformation in the strain (Fig. S3, available at ), each construct rescued a growth defect of the strain on galactose medium at 37°C (Fig. S3), as well as Taz1p function, as demonstrated by the lack of accumulation of MLCL in these strains (). That each construct localized to mitochondria indicates that the immediate N or C termini are not required for mitochondrial targeting. Lastly, the membrane association of each construct in mitochondria was investigated by alkali extraction (). Quantitation of the alkali extraction profiles demonstrated that each construct behaved in a manner indistinguishable from endogenous Taz1p (). Insight into the topology of Taz1p was obtained by ascertaining the proteinase K accessibility of endogenous Taz1p () and the three tagged Taz1p constructs () in intact mitochondria (, lanes 2–5), osmotically shocked mitoplasts (, lanes 6–9), and 0.1% Triton X-100–solubilized mitochondria (, lanes 10–13). As expected, Tom70p, which is an OM protein, was readily degraded by proteinase K added to intact mitochondria (, lanes 2–5); Tim54p, which is an integral IM protein facing the IMS, was accessible to added proteinase K after OM rupture by osmotic shock (, lanes 6–9); and α-ketoglutarate dehydrogenase (KDH), which is a matrix resident, was only degraded by proteinase K when the mitochondria were solubilized with Triton X-100 (, lanes 10–13). Interestingly, 2–3 fragments (, gray arrows) could be detected in wt-derived samples after each experimental condition in the absence of added proteinase K (, lanes 2, 6, and 10). Each of these bands is not detected in samples prepared from mitochondria (, background bands highlighted with asterisks). The same three fragments, migrating slightly slower because of the appended epitope tags, are detected in MycTaz- () and TazMycTaz- (), but not in TazHA- (), derived samples, implying that these fragments are generated through the removal of increasingly larger portions of the C terminus. Upon further characterization (unpublished data), these fragments are generated during the TCA precipitation step used to completely inactivate proteinase K. Therefore, the presence of these bands reflects the persistence of full-length Taz1p at the end of an indicated incubation; and, conversely, their absence reflects a loss in full-length Taz1p caused by proteinase K–mediated degradation. No diminution in the full-length Taz1p signal was observed upon addition of increasing amounts of proteinase K to intact mitochondria (, A and C–E, lanes 3–5). In wt mitoplasts, the addition of low amounts of proteinase K results in a decrease in detectable full-length Taz1p and the appearance of an ∼27-kD fragment (, lanes 7 and 8, white arrow). Addition of high concentrations of proteinase K results in the loss of the ∼27-kD Taz1p fragment and the appearance of a novel ∼24-kD protected band (, lanes 8–9, black arrow). It is worth noting that the full-length Taz1p detected after addition of proteinase K to mitoplasts reflects the proportion of mitochondria remaining intact after the osmotic shock reaction. To detect the Taz1p fragments, overexposed images of the Taz1p immunoblots are presented (for lighter exposure see ). Importantly, neither of these bands is detected in samples prepared from (), MycTaz (), or TazMycTaz () mitoplasts (lanes 6–9), demonstrating that these fragments are generated by the removal of at least the N-terminal 155 amino acids of Taz1p (the first amino acid downstream of the integrated tag in TazMycTaz). In TazHA mitoplasts (), the addition of increasing amounts of proteinase K results in the sequential appearance and disappearance of an ∼29-kD fragment. Thus, the final protected ∼24-kD fragment observed in wt mitoplasts (, black arrow) results from the proteinase K–mediated removal of the C-terminal ∼27 amino acids of Taz1p. Interestingly, 0.1% Triton X-100 stabilizes a core structure of Taz1p, which is ∼27 kD and resists degradation, even at 100 μg/ml proteinase K (, lane 13, white arrow). This band is not detected in samples prepared from - (), MycTaz- (), or TazMycTaz-solubilized () mitochondria (lanes 11–13). In TazHA-solubilized mitochondria, a ∼29-kD band is readily detected at 10 μg/ml proteinase K. However, upon addition of 10× more proteinase K, this fragment is much fainter, suggesting that the appended C-terminal HA tag is not included in the final 0.1% Triton X-100–stabilized Taz1p core structure. Collectively, both termini of Taz1p are exposed to the IMS. Moreover, given that the banding profile for the MycTaz and TazMycTaz constructs are identical and that Taz1p is not a peripheral membrane protein, we conclude that Taz1p is a so-called integral interfacial protein, associating with mitochondrial membranes by protruding into, but not through, the lipid bilayer (). To date, 28 distinct mutations resulting in single amino acid changes in Taz1p have been identified in BTHS patients, 21 of which occur at residues that are either conserved or identical in the yeast orthologue. Interestingly, a cluster of four such conserved BTHS mutations reside within the second predicted TM domain, amino acid residues 215–232 of Taz1p (). This predicted TM domain is located within the Triton X-100–stabilized, tightly folded Taz1p structural element and is an attractive candidate for mediating the integral interfacial association of Taz1p with membranes. To test the hypothesis that yeast Taz1p residues 215–232 are an integral interfacial membrane anchor and that the BTHS mutants occurring within the orthologous region of human Taz1p inactivate this function, each of the four mutations were modeled in yeast Taz1p and individually expressed in the yeast strain. Compared with yeast transformed with wt Taz1p (WT Taz1p), the expression of the BTHS mutations was either drastically (V223D, V224R, and I226P) or slightly (G230R) reduced (). Moreover, none of these BTHS mutants rescued the growth defect of the yeast strain (not depicted) or prevented the accumulation of MLCL, which is a hallmark of loss of Taz1p activity (). As three of the four Barth syndrome mutations occur at conserved, but not identical, residues in the yeast orthologue, a “humanized” yeast Taz1p was generated, containing the human residues at all three positions. Importantly, the humanized yeast Taz1p was expressed at levels similar to wt Taz1p expressed in yeast (), rescued the growth defect of the strain (not depicted), and prevented the accumulation of MLCL (). Thus, all four BTHS mutations, when modeled in yeast Taz1p, are nonfunctional. One possibility for the inability of each of these mutations to rescue Taz1p function is that they fail to localize properly to mitochondria. However, all four mutations were exclusively localized to mitochondria (), demonstrating that this cluster of BTHS mutations does not result in the inactivation of a mitochondrial targeting signal in Taz1p. The membrane association of each of the mutants was investigated. Surprisingly, all four BTHS mutants retained the ability to associate with mitochondrial membranes based on their continued presence in the pellet fraction after sonication (). Identical to wt Taz1p, high-salt washing of intact mitochondria or osmotically swollen mitoplasts failed to strip any of the four mutants off of the mitochondrial membranes (unpublished data). However, when the membrane association of each of the mutants was assessed by alkali extraction, all four BTHS mutants were significantly more extractable by 0.1 M NaCO, pH 10.9 and 11, than wt Taz1p (, B and C, red arrows). Therefore, whereas each of the BTHS mutants retains some capacity to associate with mitochondrial membranes, the nature of that membrane association is altered. This is consistent with the hypothesis that residues 215–232 of yeast Taz1p represent a membrane anchor. To determine if the altered membrane association of each of the BTHS mutants resulted in a different submitochondrial localization, the compartment in which each mutant resides was assessed using a proteinase K protection assay (). Intriguingly, the three BTHS mutants occurring in the middle of the postulated membrane anchor are mislocalized to the mitochondrial matrix because they are not susceptible to protease digestion during osmotic shock (, red arrows); rather, the mutants are only digested by protease when Triton X-100 is added to disrupt the mitoplasts. Thus, for these three BTHS mutants, the failure to rescue Taz1p function in the yeast strain is explained by a mislocalization within the mitochondrion. In contrast, the BTHS mutation occurring more toward the edge of the predicted membrane anchor, G230R, is resident to IMS-facing mitochondrial membranes, similar to wt Taz1p. Given that the G230R mutant displayed an altered membrane association and was unable to rescue Taz1p function in the yeast strain, the assembly of the G230R Taz1p mutant into macromolecular complexes was assessed by blue native–PAGE after solubilization of mitochondria with 1.5% (wt/vol) digitonin. Importantly, wt Taz1p overexpressed in the yeast strain provided an identical profile of Taz1p complexes as endogenous Taz1p, with the expected increase in intensity of each detected complex; thus, overexpression of Taz1p, per se, does not alter its complex assembly. Specifically, wt Taz1p migrated on blue native gels as a broad smear ranging from ∼45–140 kD, with three larger 160-, 220-, and 280-kD complexes evident (, black, green, and blue arrows, respectively); critically, all of these Taz1p complexes were not detected in mitochondrial extracts derived from either yeast or yeast transformed with empty vector (, Vector Alone). In stark contrast to the three mislocalized mutants, the G230R Taz1p mutant migrated as a broad and intense smear from ∼45 to 400 kD; however, a discrete and unique 460-kD complex was also observed (, red arrow). Thus, a single point mutation that alters the membrane association of Taz1p, but not its submitochondrial localization, results in the inappropriate assembly of G230R Taz1p into aberrant protein complexes or, instead, freezes G230R Taz1p into protein complexes that are normally dynamic and transient in nature. In conclusion, these data demonstrate that Taz1p residues 215–232 are, in fact, an integral interfacial membrane anchor and provide the first mechanistic explanations for a series of BTHS mutations. In this study, using a new anti-Taz1p antiserum, we have demonstrated that endogenous Taz1p is a normal resident of mitochondria, consistent with the hypothesis that it functions as a CL acyltransferase. Moreover, we show that Taz1p associates with all mitochondrial membranes facing the IMS. The conclusion that Taz1p associates with the inner leaflet of the OM and the outer leaflet of the IM and, thus, effectively lines the IMS is based on three separate observations. First, Taz1p was only susceptible to added proteinase K upon osmotic shock of the OM, demonstrating that it resides within the IMS. Second, Taz1p was associated with both the IM and OM, as assessed by immunoelectron microscopy (). Third, Taz1p was localized to the IM, OM, and contact sites after separation of these compartments using a linear sucrose gradient (). Our conclusion that Taz1p is localized to the IM, as well as to the OM, contrasts with recent results in which it was concluded that an epitope-tagged Taz1p was found exclusively in association with the OM (), which is a localization that is difficult to reconcile with the vast enrichment of CL in the IM. However, this conclusion was drawn although their IM and OM resolved in immediately adjacent fractions. In addition, a recent paper describing the proteome of purified OM vesicles failed to identify Taz1p (). That Taz1p was identified in another proteomic study using whole mitochondria () indicates that Taz1p can be identified in a proteomics-based approach and that the failure to identify it in the OM may reflect its relatively low abundance in this compartment. Is it surprising that Taz1p is localized on both the IM and OM of mitochondria? As only two other mitochondrial proteins, Mgm1p and Fzo1p, have been demonstrated to have this dual-membrane topology (; ), the simple answer is yes. CL is highly enriched in the IM, and the expectation was that Taz1p would, as a CL acyltransferase, reside in the IM. However, CL has been additionally detected in both contact sites and the OM (; ), although the proportion of CL associated with these two membrane compartments relative to the IM is, at present, controversial. Quite possibly, our determination that Taz1p resides on the IM and OM of wt mitochondria reflects the relative distribution of its putative target, CL. However, recent work has suggested that, in yeast, Taz1p may additionally function as a lyso-PC acyltransferase (). That this might not be a yeast-specific phenomenon is suggested by the observation that the molecular composition of PC and phosphatidylethanolamine (PE) is altered in specimens from BTHS patients (; ). Moreover, a pathway of CL remodeling identified in rat liver and human lymphoblast mitochondria was recently described, in which PC and PE acted as the acyl donor for either CL or MLCL (). Importantly, this transacylation pathway was decreased in lymphoblasts derived from BTHS patients. PC constitutes approximately half of all mitochondrial phospholipids and is slightly enriched in the OM, relative to the IM (). Additionally, the OM has long been known to be enriched in a lyso-PC acyltransferase activity (; ). Therefore, if Taz1p were to act as an acyltransferase for both CL and PC, then the observed dual localization might reflect the relative distribution of PC between the IM and OM. As Taz1p was not removed by high salt and exhibited an alkali extraction profile distinct from the peripherally associated cytochrome and similar to the integrally associated Tim23p, it was expected that Taz1p is an integral membrane protein. Instead, a series of functional epitope-tagged constructs allowing the unambiguous determination of the membrane topology of Taz1p revealed that Taz1p does not contain any TM domains (up to two TM domains were predicted by the different programs). Our conclusion is supported by the fact that the regions of Taz1p on either side of the first potential TM domain both face the IMS, as does the extreme C terminus of Taz1p. Thus, we concluded that Taz1p is a monotopic integral interfacial membrane protein, which is an emerging class of membrane proteins that includes the alternative oxidase of plants and prostaglandin H2 synthase-1 (; ). Members of the monotopic integral interfacial membrane protein class are proposed to associate with membranes by protruding into, but not completely through, a lipid bilayer (). Worth briefly considering is that although alkali extraction, in combination with sonication and high-salt washes, can clearly identify peripheral membrane proteins, it cannot distinguish between classical integral membrane proteins, such as Tim23 with four TM domains, and nonclassical membrane proteins, such as Taz1p, which associate with membranes presumably through TM-like loops into the lipid bilayer. An attractive candidate for mediating the integral interfacial association of Taz1p with the membrane was the second region (residues 215–232) predicted to be a TM domain. The importance of this domain in Taz1p function was first suggested by the observation that a cluster of four BTHS mutations occur within the orthologous region of human Taz1p. Modeling each of these mutations in yeast Taz1p results in a loss of Taz1p function. Consistent with the assignment of residues 215–232 as an interfacial membrane anchor, each of these BTHS mutants exhibited an altered association with mitochondrial membranes (). Perhaps the most interesting aspect of this cluster of mutations was that there were two different consequences of mutations within this defined region. Three of the BTHS mutants were mislocalized to the mitochondrial matrix. As each of these mutations occurs in the middle of the predicted membrane anchor, it is tempting to speculate that they result in the inactivation of a stop–transfer signal that normally prevents the transport of Taz1p across the IM. Implicit in this observation is that the import of Taz1p into mitochondria normally involves an interaction with one of the translocases of the IM. This lends further weight to our conclusion that Taz1p includes the IM as one of its resident compartments. Thus, BTHS can be caused by Taz1p missorting within the mitochondrion (). The fourth BTHS mutant, G230R, localized appropriately to membranes lining the IMS, but assembled into abnormal complexes or abnormally stable complexes. Because this mutation occurs near the edge of the membrane anchor and involves the acquisition of a positive charge, we suggest that the membrane anchor is pulled partially out of the membrane by interactions between the positively charged Arg and negatively charged phospholipid headgroups (). Therefore, although the stop–transfer activity of this region is intact, the association of G230R with mitochondrial membranes facing the IMS is altered, leading to aberrant complex formation and loss of Taz1p function. Finally, the conclusion that Taz1p is a monotopic, integral, interfacial membrane protein that lines the IMS indicates that the acyltransferase activity of Taz1p is mechanistically performed in the context of only those membrane leaflets facing the IMS (). Thus, the final distribution of remodeled CL and/or PC within mitochondrial membranes would require trafficking between leaflets of a bilayer subsequent to Taz1p-mediated acylation. Future detailed investigations such as these will provide important insights into the mitochondrial dysfunction associated with BTHS and potential targets for treating this disease. Tafazzin was cloned into pBSK after PCR, using genomic DNA isolated from the wt GA74-6A strain as a template and 5′ and 3′ primers hybridizing ∼300 bp upstream of the predicted start translation and 430 bp downstream of the stop translation codon of . This construct, termed pBSK.Taz, acted as the template in all subsequent cloning procedures involving tafazzin. To generate Taz1p containing an N-terminal His tag, the entire open reading frame was cloned into the pET28a vector (Novagen) in frame and downstream of the His tag and thrombin cleavage site provided by the vector. HisTaz was induced in BL21-CodonPlus(DE3)-RIL (Stratagene) and purified under native conditions using Ni-agarose (QIAGEN) as per the manufacturer's instructions. Tafazzin with an in-frame C-terminal HA tag was generated using a 5′ primer that hybridized ∼300 bp upstream of the start translation codon and a 3′ primer containing a stop translation codon, the sequence for the HA tag, and the 19 bps immediately upstream of the endogenous stop codon. To generate tafazzin with an in-frame N-terminal Myc tag, overlap extension () was performed using a primer set that placed a Kozak and Myc tag sequence in frame with the first predicted amino acid of tafazzin, still downstream of the same ∼300 upstream of the endogenous start site used in the previous construct. To insert a Myc tag between the two potential TM domains, overlap extension was performed using a primer set that inserted the first seven amino acids of the myc tag sequence between amino acids 154 and 155 of tafazzin, with the final two amino acids (Asp and Leu) of the Myc tag provided by amino acids 155 and 156 of tafazzin. The series of BTHS point mutations were also generated by overlap extension. Each construct was cloned into pRS425. The sequence of every construct was verified by sequencing. The sequences of all primers are available upon request. The wt parental yeast strains used were as follows: GA74-6A ( α, ) and GA74-1A ( ). The His1.5 ) strain was constructed by replacing the entire open reading frame of with the marker using a PCR-mediated one-step gene replacement strategy (). The His1.5 strain was transformed and selected as previously described () with the empty vector, pRS425, WT Taz1p, or the aforementioned epitope-tagged Taz1p or BTHS Taz1p mutant constructs, all cloned into pRS425. Antibodies were raised in rabbits using the yeast HisTaz. Yeast whole-cell extracts were prepared as previously described (). Most of the antibodies used in this work were generated in the Schatz laboratory (J. Schatz, University of Basel, Basel, Switzerland) or our laboratory and have been described previously. Other antibodies used were as follows: mouse anti-Sec62p (gift of David Meyers, University of California, Los Angeles, Los Angeles, CA), anti–β-Actin (Abcam Inc.), anti-HA (Covance Research Products, Inc.), anti-Myc (strain 9E10; ; obtained from the Developmental Studies Hybridoma Bank; developed under the auspices of the National Institute of Child Health and Human Development and maintained by the University of Iowa), and anti-Myc (clone 9B11; Cell Signaling Technology) monoclonal antibodies, and horseradish peroxidase–conjugated secondary antibodies (Pierce Chemical Co.). After resolution on 12% SDS-PAGE gels under reducing conditions, proteins were transferred to nitrocellulose membranes (Schleicher and Schuell BioScience) at 1 Amp for 60–75 min at room temperature. Immunoblots were performed exactly as previously described (). All of the presented images were captured on film. For quantitation of immunoblots, images were captured with a VersaDoc controlling a charge-coupled device camera (Bio-Rad Laboratories), and bands from two exposures per blot were quantitated with the affiliated Quantity One software. Formulas for specific calculations are presented in the appropriate figure legends. Statistical analyses were performed using SigmaStat 3.0 (Jandel Corp.). For the subcellular fractionation studies, wt yeast were grown at 30°C in YPEG medium containing 1% yeast extract, 2% tryptone, 3% glycerol, and 3% ethanol. The mitochondria isolated for all of the other experiments were derived from cultures grown at 30°C to an OD of ∼0.8–1 in rich lactate medium (1% yeast extract, 2% tryptone, 0.05% dextrose, and 2% lactic acid, 3.4 mM CaCl 2HO, 8.5 mM NaCl, 2.95 mM MgCl 6H0, 7.35 mM KHPO, and 18.7 mM NHCl). Mitochondria were isolated as previously described (). Subcellular fractions were collected through a series of differential centrifugations. The amount of protein in each fraction was determined using the BCA assay (Pierce Chemical Co.). Immunoelectron microscopy was performed as previously described (). In brief, cells were fixed in suspension for 15 min by adding an equal volume of freshly prepared 8% formaldehyde contained in 100 mM PO buffer, pH 7.4. The cells were pelleted, resuspended in fresh fixative (8% formaldehyde, 100 mM PO, pH 7.4), and incubated for an additional 18–24 h at 4°C. The cells were washed briefly in PBS and resuspended in 1% low-gelling–temperature agarose. The agarose blocks were trimmed into 1-mm pieces, cryoprotected by infiltration with 2.3 M sucrose/30% polyvinyl pyrrolidone (10,000 mol wt)/PBS, pH 7.4, for 2 h, mounted on cryopins, and rapidly frozen in liquid nitrogen. Ultrathin cryosections were cut on an ultramicrotome (UCT; Leica) equipped with an FC-S cryoattachment and collected onto formvar/carbon-coated nickel grids. The grids were washed through several drops of 1× PBS containing 2.5% fetal calf serum and 10 mM glycine, pH 7.4, and then blocked in 10% FCS for 30 min and incubated overnight in rabbit anti-Taz1p antibody. After washing, the grids were incubated for 2 h in 5-nm gold donkey anti–rabbit conjugate (Jackson ImmunoResearch Laboratories). The grids were washed through several drops of PBS, followed by several drops of ddHO. Grids were then embedded in an aqueous solution containing 3.2% polyvinyl alcohol (10,000 mol wt)/0.2% methyl cellulose (400 centipoises)/0.1% uranyl acetate. The sections were examined and photographed on a transmission electron microscope (EM 410; Philips) at 100 kV and images were collected with a digital camera (Megaview III; Soft Imaging System). Figures were assembled in Photoshop (Adobe), with only linear adjustment of contrast and brightness. For submitochondrial localization, with the exception of the epitope-tagged Taz1p strains, all studies were performed using at least two different batches of isolated mitochondria per strain. To assay membrane association by sonication, the pellet fraction after osmotic shock of 0.5 mg mitochondria was resuspended in 0.6 M sucrose, 3 mM MgCl, and 20 mM Hepes-KOH, pH 7.4, and sonicated for 3 × 10 s, with 30-s intervals in an ice bath, using a microtip attached to a Sonic Dismembrator 550 (Fisher Scientific) with the amplitude set at 3.5. After removal of unbroken mitoplasts by centrifugation at 20,000 for 10 min at 4°C, the submitochondrial particles were separated from soluble matrix components with an airfuge at 27 psi for 30 min at 4°C. High-salt washes were performed for 15 min at 4°C by the addition of 0.5 M NaCl or 1 M KCl to either intact mitochondria or mitoplasts after osmotic shock. Alkali extraction was performed essentially as previously described (), except that 0.2 ml of 0.1 M NaCO at the indicated pH was added to 0.2 mg mitochondria, and the pellet and supernatant fractions were separated with an airfuge at 27 psi for 15 min at 4°C. For experiments using intact mitochondria, mitochondria were incubated in 0.6 M sorbitol and 20 mM Hepes-KOH, pH 7.4. Osmotic shock was performed by incubating mitochondria for 30 min on ice in 0.03 M sorbitol and 20 mM Hepes-KOH, pH 7.4. Where designated, the indicated concentration of proteinase K ±0.1% (vol/vol) Triton X-100 was included. Any proteinase K remaining associated with the pellets was inactivated, as previously described (). The supernatants were TCA precipitated and the pellet and supernatant fractions resuspended in equal volumes of thorner buffer (10% glycerol, 8 M urea, 5% (wt/vol) SDS, 40 mM Tris, pH 6.8, 4 mg/ml bromophenol blue, and 5% β-mercaptoethanol). The separation of sonicated membranes over linear sucrose gradients was performed essentially as previously described (). In brief, 5 mg of mitochondria were osmotically shocked (1 mM EDTA added to swelling medium) for 30 min and then shrunk for 10 min on ice by the addition of sucrose to 0.45 M. Sonication was performed as before, except six cycles were performed. After removal of unbroken mitoplasts by centrifugation for 10 min at 20,000 at 4°C, the sonicated vesicles were harvested with an airfuge at 27 psi for 30 min at 4°C and the membrane-containing pellet was resuspended with 0.5 ml of 0.45 M sucrose, 10 mM KCl, 1 mM EDTA, and 5 mM Hepes-KOH, pH 7.4, loaded onto a linear sucrose gradient (1.8–0.85 M sucrose, 10 mM KCl, 1 mM EDTA, and 5 mM Hepes-KOH, pH 7.4; 4 ml total volume), and centrifuged in a SW41Ti rotor (Beckman Coulter) at 100,000 for 20 h at 4°C. Using a syringe plunger to control the flow, a hole was punched into the bottom of the tubes using an 18-gauge needle and ∼0.2-ml fractions collected in individual wells of a 96-well plate. The fractions were quantified using the Bio-Rad Protein Assay (Bio-Rad Laboratories). Starter cultures were diluted to an OD = 0.2 in 2 ml of yeast peptone dextrose (wt or yeast) or SC-Leu (all remaining tested strains) supplemented with 10 μCi/ml P and grown at 30°C for ∼24 h. After a wash with HO, the yeast pellets were resuspended in 0.3 ml MTE buffer (0.65 M mannitol, 20 mM Tris, pH 8.0, and 1 mM EDTA) supplemented with 1 mM PMSF, 10 μM leupeptin, 2 μM pepstatin A, and 10 μM chymostatin (the latter three were obtained from Sigma-Aldrich), transferred to an Eppendorf tube containing 0.1 ml glass beads, and disintegrated by vortexing on high for ∼30 min at 4°C. A crude mitochondrial fraction was sedimented after a low-speed spin at 250 to remove the glass beads and any remaining intact yeast by centrifugation for 5 min at 13,000 at 4°C. Phospholipids from equal amounts of labeled crude mitochondria, as determined by liquid scintillation, were extracted with 1.5 ml 2:1 chloroform/methanol by vortexing at room temperature for 1 h. 0.3 ml of ddHO was added, the samples were vortexed on high for 1 min, and the phases were separated by centrifugation at 1,000 rpm in a clinical centrifuge at room temperature. The upper aqueous phase was removed by aspiration and the organic phase washed with 0.25 ml 1:1 methanol/HO. After phase separation carried out as before, the lower organic phase was transferred to a new borosilicate tube and dried down under a stream of liquid nitrogen. Chloroform-resuspended samples were loaded onto silica gel TLC plates (Analtech) and resolved in 1D twice using chloroform/ethanol/HO/triethylamine (30:35:7:35), as previously described (). Labeled phospholipids were revealed using a K-screen and FX-Imager (Bio-Rad Laboratories), quantitation was performed using the affiliated Quantity One software, and statistical analyses were performed using SigmaStat 3.0 (Systat Software, Inc.). Detergent solubilization of mitochondria (5 mg/ml) was performed for 30 min on ice with 20 mM Hepes-KOH, pH 7.4, 10% glycerol, 50 mM NaCl, 1 mM EDTA, and 2.5 mM MgCl supplemented with 1.5% (wt/vol) digitonin (Biosynth International, Inc.) and protease inhibitors, as listed in phospholipid analyses; insoluble material was removed by centrifugation for 30 min at 20,000 at 4°C. ∼100 μg of solubilized material was analyzed by blue native gel electrophoresis on a 6–16% linear polyacrylamide gradient and Taz1p was detected by immunoblot after transfer to PVDF membranes. Fig. S1 shows that a rabbit anti-Taz1p antiserum specifically recognizes yeast Taz1p. Fig. S2 shows that Taz1p associates with mitochondrial membranes. Fig. S3 shows that the three epitope-tagged Taz1p constructs rescue growth of the yeast strain on YP-galactose at 37C. Fig. S4 shows the altered membrane association of the BTHS Taz1p mutants results in two fates; matrix mistargeting or aberrant complex assembly. Table S1 shows Taz1p transmembrane domain predictions. Online supplemental material is available at .

Deciphering cytokinesis is challenging because it involves >50 proteins in both fungi and animal cells (; ; ; ). In fission yeast () and cells (), a conserved set of core proteins follow similar temporal pathways to assemble a contractile ring of actin filaments and myosin II that pinches the dividing cell in two. Nevertheless, even in fission yeast, 25 yr of genetic analysis (; ; ) and measurements of the global and local concentrations of 28 of the proteins () have left many fundamental questions unanswered. One point of disagreement concerns the initial assembly of the contractile ring. A popular hypothesis is that the anillin-like protein Mid1p recruits a progenitor spot containing myosin II or formin Cdc12p and the pombe Cdc15 homology (PCH) protein Cdc15p to the division site, followed by extension of a leading cable from the spot around the circumference of the equator (; , ; ; ; ; ). On the other hand, we (; ) and others (; , ; ) observed that some contractile ring proteins first form a broad band of small puncta around the middle of the cell and then coalesce to form a contractile ring. The nodes in the broad band contain Mid1p, myosin II Myo2p and associated light chains, and IQGAP Rng2p, but it was not known whether Cdc15p and Cdc12p concentrate in nodes or if localization of all of the node proteins depends on Mid1p. The role of the actin-related protein 2/3 (Arp2/3) complex in cytokinesis has also been uncertain. The Arp2/3 complex nucleates branched actin filaments () and concentrates in actin patches, but not the contractile ring in fixed fission yeast (; ; ). cells do not require the Arp2/3 complex for cytokinesis (; ), and fission yeast with the conditional mutations and have defects in septation, but form normal actin contractile rings under restrictive conditions (; ). However, Pelham and Chang () detected GFP-tagged ARPC5 (the smallest subunit, Arc5p/Arc15p, of the Arp2/3 complex) in contractile rings of live fission yeast and showed that the Arp2/3 complex is required for actin polymerization in contractile rings of permeabilized yeast. Furthermore, presented evidence that the PCH protein Cdc15p recruits both formin Cdc12p (by direct interaction) and the Arp2/3 complex (by interaction with myosin I Myo1p, which is an activator of the Arp2/3 complex) to the contractile ring. Others observed myosin I Myo1p in the contractile ring (). We addressed these issues by fluorescence microscopy of fission yeast strains expressing functional fusion proteins from their normal chromosomal loci under control of their native promoters. We found that most of the nodes around the equator of G2/M cells contain the seven proteins associated with the assembly of the contractile ring, and that the presence of all of these proteins in nodes depends on Mid1p. We reproduced most of the reported observations on progenitor spots, but found that nodes marked with functional fusion proteins condense laterally into a contractile ring without forming a progenitor spot or extending a leading cable. We also found that Arp2/3 complex contributes to septation, but neither Arp2/3 complex nor its activators concentrate in contractile rings. Instead, they begin to accumulate in endocytic actin patches lateral to the contractile ring after its formation. Remarkably, Cdc15p completely transitions from interphase endocytic actin patches dependent on the Arp2/3 complex to the contractile ring dependent on formin Cdc12p during mitosis. Our first question was if the contractile ring forms by growth from a single progenitor spot containing both myosin II isoforms, Myo2p and Myp2p, and the myosin regulatory light chain, Rlc1p (; ; ; ). Time-lapse microscopy of cells expressing both functional GFP-Myo2p and a spindle pole body (SPB) marker, Sad1p-GFP, at their endogenous levels showed that the contractile ring assembles by lateral coalescence from a broad band of nodes before the onset of anaphase B ( and Video 1, available at ). The behavior of Rlc1p-GFP is similar to Myo2p, including the number of molecules present in nodes (). Myp2p is not a component of the broad band, but joins the contractile ring ∼10 min after it forms (; ). What is the nature of the progenitor spot? We confirmed that the partially functional YFP-Myp2p fusion protein (with YFP on the N terminus) formed spots that persist >50 min after ring constriction (). Even the functional fusion proteins Myp2p-YFP and Rlc1p–monomeric CFP (mCFP) expressed from their native promoters colocalized in one or more spots at 36°C, but rarely at 25°C (Fig. S1 A, available at ). These spots did not interfere with cytokinesis and contained no Myo2p, even at 36°C (Fig. S1 B). Myp2p remained in a spot after Myo2p condensed into a contractile ring (Fig. S1 B, 67 min). Thus, the Myp2p-Rlc1p spot does not contain Myo2p. The presence of Rlc1p in the progenitor spot depended on Myp2p. Myp2p-GFP formed a spot at 36°C in most cells lacking Rlc1p, but no Rlc1p-GFP spot appeared in a strain lacking Myp2p (). Cells lacking a Rlc1p spot formed a normal broad band and a contractile ring, and the ring constricted normally ( and Video 2, available at ). Myp2p-Rlc1p spots incorporated into the contractile ring at various times, but these spots could persist after the assembly of a contractile ring (, arrow). The movements of Rlc1p spots appeared to be independent of the movements of the nodes in the broad band (). Although the exact nature of the spot is still unknown, our data provide strong evidence that it is not essential for the formation of the contractile ring. Extensive observations of cells expressing functional fusion proteins show that at least five proteins (anillin Mid1p, myosin II and two associated light chains, and IQGAP Rng2p) form a broad equatorial band of nodes before condensing into a contractile ring (see Introduction). However, it was not clear if other proteins are present in the nodes or if each node contains all of these proteins. In particular, we were uncertain about formin Cdc12p because it is the least abundant of the known cytokinesis proteins in , with only ∼600 molecules per cell and 300 molecules in the contractile ring (). Thus, it is very difficult to detect Cdc12p by microscopy as a GFP fusion protein or with fluorescent antibodies without overexpression (; , ). To obtain a stronger signal, we fused three tandem copies of YFP or GFP to the C terminus of Cdc12p. These fusion proteins are functional, and the fluorescent intensities are three times those of Cdc12p fusion proteins with a single YFP or GFP (). These brighter fusion proteins enabled us to observe Cdc12p during the early stages of ring formation. Within 1 min after SPB separation at the onset of mitosis, Cdc12p concentrated in small nodes located in a broad band around the equator of the cell (, arrowheads). Subsequently, these nodes coalesced laterally to form a ring that constricted normally (Video 3, available at ). We confirmed these observations using synchronized cells expressing Cdc12p-3GFP (, B and C; and Video 4). Some nodes marked with Cdc12p formed linear arrays as they condensed into a ring, but not a single leading cable. In the presence of latrunculin A (Lat-A) to inhibit actin polymerization, or in cells depleted of profilin Cdc3p, Cdc12p nodes formed normally, but did not coalesce into a contractile ring (, Fig. S2, and Videos 5 and 6). In the absence of profilin, the actin filaments nucleated by formin Cdc12p cannot elongate from the fast-growing barbed ends, but grow very slowly from their pointed ends (). Thus, formin Cdc12p localizes to nodes in the broad band independent of actin filaments and profilin. Furthermore, Cdc12p-dependent polymerization of actin filaments is required for nodes to coalesce into a contractile ring. The localization of Cdc15p in the broad band of nodes was also in question. GFP-Cdc15p concentrates in patches near the ends of the cell during interphase and in the contractile ring during cell division (; ; ). Under our conditions, GFP-Cdc15p first concentrated in many small nodes around the equator, followed by condensation of these nodes into a continuous contractile ring ( and Video 7, available at ), rather than forming a spot that initiates a leading cable. The forming ring of GFP-Cdc15p initially had many gaps, but the gaps filled in () as more Cdc15p molecules were recruited (). Cdc15p also localized to a broad band of nodes in cells arrested at G2/M in Lat-A (, H and J; ). To determine if each node in the broad band contains all seven proteins known to accumulate in nodes, we observed cells expressing pairs of contrasting fluorescent fusion proteins in wild-type () or cells (). To facilitate some observations (), cells were treated with Lat-A to prevent nodes from condensing into compact rings. We observed fluorescent fusion proteins in cells arrested at G2/M and then released into Lat-A for two reasons. First, by arresting cells at G2/M and then releasing them, we enriched for cells forming broad bands. Second, Rng2p, Cdc15p, and Cdc12p normally accumulate in nodes over a period of 10 min after myosin II, followed quickly by condensation of the nodes into a contractile ring (). Thus, Myo2 normally colocalizes in nodes with Rng2p, Cdc15p, or Cdc12p for just a few minutes, unless Lat-A prevents condensation of the nodes. The weaker fluorescence of CFP and higher background autofluorescence in the CFP relative to the YFP channel are also challenges for imaging some pairs of proteins. We observed that most nodes contained anillin-like Mid1p, conventional myosin II heavy chain Myo2p and light chains Cdc4p and Rlc1p, IQGAP Rng2p, PCH protein Cdc15p, and formin Cdc12p (). Approximately 70% of nodes in wild-type cells contained Myo2p, Mid1p, Rng2p, and Cdc15p (). In cells treated with Lat-A, a similar or even higher fraction of nodes contained all seven proteins (). Nearly all nodes marked with Myo2p also contained myosin light chains Rlc1p and Cdc4p, anillin-like protein Mid1p, and IQGAP Rng2p (). A majority of nodes marked with Myo2p also contained the PCH protein Cdc15p and formin Cdc12p. Most nodes marked by Cdc15p also contained Cdc12p (), but the Cdc15p signal was stronger in some nodes because of the fivefold excess of Cdc15p over Cdc12p (). The lack of signal from α-actinin Ain1p-YFP (a late arriving component of contractile rings) in the broad band () indicated that the signals from the other proteins do not arise from artifacts of microscopy or physiology in the background. The dependence of Cdc12p and Cdc15p on Mid1p for localization in nodes had not been established, so we compared the behavior of these proteins in and backgrounds. We synchronized cells expressing tagged Cdc12p and Cdc15p by incubation at 36°C and release to 23°C into Lat-A to prevent condensation of nodes into a contractile ring. In the presence of Mid1p, Myo2p, Cdc12p, and Cdc15p formed broad bands of nodes ( and L–O). Myo2p persisted in the nodes for ∼100 min during incubation in Lat-A, and then dispersed. Cdc15p and Cdc12p formed aggregates during long incubations in Lat-A. In the absence of Mid1p, none of Myo2p, Cdc12p, or Cdc15p concentrated in a broad band of nodes (). Myo2p remained diffusely spread in the cytoplasm, but Cdc12p formed aggregates faster in the absence, rather than in the presence, of Mid1p (Fig. S3, available at ). Thus, Mid1p is required for node formation. Studies in the literature differ regarding the participation of the Arp2/3 complex and its activators in contractile ring function. One study reported the Arp2/3 complex in the contractile ring (), and two studies presented evidence that assembly of contractile ring filaments depends on the Arp2/3 complex (; ). Other genetic and cellular studies presented evidence that the Arp2/3 complex is not required for actin-ring assembly (; ; ). Our new experiments show that the Arp2/3 complex and its activators contribute to septation, but not assembly of the contractile ring. Our time-lapse microscopy and 3D reconstructions using functional fusion proteins expressed from their chromosomal loci, under the control of their native promoters, clarify the early steps in the assembly of the contractile ring (). We found that the anillin-like protein Mid1p initiates the assembly as a broad band of nodes. Mid1p exits from the nucleus and specifies a broad band of ∼75 nodes around the cell equator. Mid1p first recruits dephosphorylated type II myosin Myo2p with the light chains Cdc4p and Rlc1p, as well as recruiting IQGAP Rng2p to nodes in this broad band before the onset of mitosis (; ). Myo2p is insoluble at physiological ionic strength (; ), so it may assemble into minifilaments like other cytoplasmic type II myosins. The role of Rng2p in the process is unclear. PCH protein Cdc15p and formin Cdc12p are recruited to the nodes at the onset of mitosis. Formin Cdc12p dimers initiate the nucleation of unbranched actin filaments and remain attached to their barbed end (, ; ). We assume that nodes attach to the inner surface of the plasma membrane and that a unitary node consists of ∼20 anillin Mid1p, 20 dimers of myosin II (Myo2p and light chains Cdc4p and Rlc1p), 20 IQGAP Rng2p, 20 PCH protein Cdc15p, 2 dimers of formin Cdc12p, and ∼250 of the essential light chain Cdc4p in excess of the binding sites on Myo2p (; ). Myo2p heads interact with actin filaments (), Myo2p tails interact with each other () and Mid1p (), Cdc4p interacts with IQ motifs of Rng2p (), and the N termini of Cdc15p and Cdc12p interact with each other (). Given these ratios and interactions, we expect nodes to have an organized macromolecular structure and orderly assembly (). However, the physical interactions among the node proteins are not yet established. Mid1p, but not polymerized actin, is required for the localization of Myo2p, Rng2p, Cdc15p, and Cdc12p to the nodes (; ). Cdc4p is required for the localization of Myo2p () and Rng2p () to the nodes. Cdc12p is not required for localization of Myo2p () and Rng2p () to the nodes. Without the nodes in cells, some cells fail to assemble a contractile ring, and some cells make misplaced and defective rings belatedly (). Thus, it seems that nodes contribute to both the proper positioning and efficient assembly of the contractile ring. Detection of formin Cdc12p in equatorial nodes is crucial for proposing a mechanism to transform the broad band into a contractile ring. Formins nucleate actin filaments at the division site (; ) and remain attached to their barbed ends so that they may also anchor the filaments in nodes. Because actin subunits add to barbed ends associated with Cdc12p, we presume that the pointed ends of these filaments radiate from nodes, which are separated by ∼0.7 ± 0.2 μm. Actin filaments >0.7 μm long might encounter myosin II in adjacent nodes. Movement of myosin II toward the barbed end of an actin filament anchored in an adjacent node might pull nodes together as they coalesce into a compact ring (). This hypothesis is consistent with the fact that assembly of actin filaments from nodes depends on Cdc12p and profilin Cdc3p and that actin filaments are required for nodes to coalesce laterally into a sharp, compact ring around the cell equator. Our lateral contraction model shares some features with models of cytokinesis in animal cells (; ), which postulate gradients of cortical tension that move the components to the equator where tension is highest (), and with the observation that poorly organized actin filaments spread over the medial cortex become aligned into parallel bundles around the equator ( ; ). Our evidence suggests that the fission yeast contractile ring assembles from a broad band of nodes () rather than by extension of a leading cable around the circumference of the cell from a single myosin progenitor spot (; ) or a Cdc12p-Cdc15p spot (; , ; ). We and others have also observed many proteins in one or more spots in fission yeast, including formin Cdc12p (; , ), PCH protein Cdc15p (), conventional myosin II Myo2p (; ), myosin II regulatory light chain Rlc1p (), unconventional myosin II Myp2p (), and α-actinin Ain1p (). These spots may be nonfunctional aggregates because they appear if proteins are overexpressed, nonfunctional, tagged with weakly dimeric GFP, or expressed at a high temperature. Alternatively, spots might be a storage form of inactive protein, as observed for inactive Cdc2-cyclin B in the cytoplasm of starfish oocytes (; ). The dissolution or incorporation of spots into the contractile ring during mitosis might result from the depletion of the inactive molecules in the cytoplasm. We favor the first possibility although we cannot rule out the latter because Cdc12p has been detected in a spot using an antibody (). Assembly of actin patches in fungi during endocytosis (, ) depends on the Arp2/3 complex to nucleate branched actin filaments (). On the other hand, assembly of actin filaments for the contractile ring in fission yeast depends on formin Cdc12p and profilin Cdc3p (; ; ). Formin Cdc12p nucleates unbranched actin filaments that grow at their barbed ends in the presence of profilin in vitro (, ; ). Contractile ring actin filaments appear to be unbranched in animal cells (; ; ), but less is known about their structure in yeast. Our experiments also add to evidence that the Arp2/3 complex does not participate in the assembly of the contractile ring. First, the strongest evidence is the presence of YFP-actin in actin patches mediated by Arp2/3 complex but not in the contractile ring. The high concentration of YFP-actin in patches () established that it could add to the free barbed end of filaments nucleated by the Arp2/3 complex. If the Arp2/3 complex nucleated any actin filaments in the contractile ring, we would observe more YFP-actin fluorescence in the ring than in the surrounding cytoplasm. A simple explanation for the lack of YFP-actin fluorescence in the contractile ring is that formin Cdc12p might nucleate all of the filaments and cannot add YFP-actin to growing barbed ends. Formins have a bias against actin with a fluorescent dye conjugated to Cys374 () located near the N terminus in the actin structure. Our observations on live cells suggest that Cdc12p does not use YFP-actin to polymerize contractile ring actin filaments. Second, well after the assembly of the contractile ring, actin patches containing the Arp2/3 complex and its activators accumulate lateral to the contractile ring, but not in the contractile ring. Third, strains with a conditional mutation in Arp3p () or lacking activators of the Arp2/3 complex ( and ) assemble Myo2p nodes and compact rings at the same rate as wild-type cells, but then progress through subsequent steps in cytokinesis slightly slower. The PCH protein Cdc15p crosses over between endocytic actin patches and the contractile ring. During interphase, Cdc15p concentrates with Arp2/3 complex in some, but not all, endocytic patches at the ends of cells (; ). At the onset of mitosis, Cdc15p leaves the endocytic actin patches and concentrates in nodes with myosin II and formin Cdc12p that condense into the contractile ring. Available evidence is consistent with the proposal that formins, rather than the Arp2/3 complex, nucleate most or all contractile ring actin filaments in other eukaryotes. Arp2/3 complex is not required for the actin-ring assembly in budding yeast () or (), and is not required for cytokinesis in (; ). On the other hand, formins are essential for actin-ring assembly and cytokinesis in fungi and animals (; ; ; ). Table S1 (available at ) lists the strains used in this study. We constructed the strains by PCR-based gene targeting (; ) and standard genetic methods (). For constructing strains tagged with 3GFP and 3YFP by gene targeting, plasmids pFA6a-3GFP-kanMX6 and pFA6a-3YFP-kanMX6 were used as templates in PCR to amplify DNAs for transformation. The plasmids were obtained by cloning triple GFP or YFP (a gift from W.-L. Lee and J. Cooper, Washington University, St. Louis, MO) into the pFA6a vector (). All strains except where noted expressed fusion proteins under the control of their native promoters and from their normal chromosomal locus. We tested the functionality of each fusion protein in previous papers (; ; ; ). We restreaked all cells, except for the strains JW816 and JW1288, from frozen stocks on YE5S plates and grew colonies at appropriate temperatures for 2–4 d. We inoculated cells from these colonies into 5–15 ml YE5S liquid medium in 50-ml baffled flasks and grew exponential cultures at densities of 1–10 × 10 in the dark at 25°C with shaking at 200 rpm for 36–48 h before observations and/or experiments. Strains JW816 (with a YFP-actin plasmid) and JW1288 ( under the control of repressible promoter) were grown in EMM medium. Some cultures contained 100 μM Lat-A to inhibit actin polymerization (, ). Cells were concentrated by centrifugation at 4,500 rpm for 5–10 s and resuspended in growth medium. 10 μl of cells were mounted directly onto a slide (for some single time-point experiments) or onto a thin layer of 100–150 μl YE5S or EMM5S medium containing 25% gelatin (G-2500; Sigma-Aldrich) and 0.1 mM n-propyl-gallate (for time-lapse experiments), sealed under a coverslip with Valap, and observed at 23°C, except where noted. We observed live cells in growth chambers () with inverted microscopes (IX-71; Olympus) equipped with a 60×/1.4 NA objective (Plan-Apo; Olympus) and appropriate filters (DIC, CFP, FITC, and YFP) or equipped with a 100×/1.4 NA objective (Plan-Apo) and a spinning-disk confocal system (UltraView RS; Perkin Elmer) with excitation by a 442-nm helium cadmium laser or 488- or 514-nm argon ion lasers. All images were acquired using cooled charge-coupled device cameras (ORCA-ER; Hamamatsu). Digital images and kymographs were processed with Image J software (National Institutes of Health; ), as previously described (). Some raw images from the confocal system were opened using the UltraView Opener plug-ins (provided by A. Isaacson and B. Mohler, University of Connecticut, Farmington, CT). Node compositions were quantified by stepping through the Z sections. YFP- and CFP-tagged proteins were considered in the same nodes if the fluorescence signals in the nodes were obviously stronger than the cytoplasmic background and the outlines of the nodes overlapped. The images for Video 5 and some images in (D and the max projection in E) were deconvolved with AutoQuant software (AutoQuant Imaging, Inc.) Fig. S1 shows myosin II spots, Fig. S2 shows that profilin Cdc3p is not required for concentration of formin in nodes, and Fig. S3 depicts formin aggregates. 10 videos are included to show that contractile ring assembly starts from a broad band of nodes (Videos 1 and 2), that the formin Cdc12p and the PCH protein Cdc15p localize to a broad band of nodes (Videos 3–7), and that YFP-actin, Arp2/3 complex, and its activators localize in dynamic actin patches, but do not concentrate in the contractile ring (Videos 8–10). Table S1 lists fission yeast strains. Online supplemental material is available at .

Skeletal muscle is composed of multinucleated myofibers that form through the process of myogenesis. During myogenesis, myoblasts must exit the cell cycle and subsequently undergo differentiation and cell–cell fusion to form myofibers in vivo or myotubes in vitro. Myoblast fusion follows an ordered set of cellular events, including cell migration, adhesion, and membrane fusion (). Myoblast fusion is important not only for skeletal muscle formation during development but also for the postnatal regeneration and growth of skeletal muscle. Mammalian myoblast fusion occurs in two phases (). Initially, myoblasts fuse with one another to form small, nascent myotubes. Additional myoblasts subsequently fuse with nascent myotubes, leading to the formation of large, mature myotubes. Although several molecules regulating the first phase of fusion have been identified, few molecules specifically regulating the fusion of myoblasts with nascent myotubes are known (). Molecules implicated to function during the second stage of fusion include secreted proteins and membrane bound proteins, as well as transcription factors. Follistatin (), prostaglandin F2α (), and interleukin-4 (IL-4; ) are secreted by muscle cells and enhance the growth of nascent myotubes. Prostaglandin F2α–mediated growth is dependent on the transcription factor NFATC2 (nuclear factor of activated T cells, cytoplasmic, calcineurin-dependent 2; ), and NFATC2 regulates expression of IL-4 (). IL-4 is secreted by a subset of nascent myotubes and acts on unfused cells, leading to their recruitment and fusion with nascent myotubes. In addition, an unknown secreted factor is responsible for mammalian target of rapamycin's actions in regulating myoblast–myotube fusion (). Membrane bound proteins are also important, as myoferlin, a protein localized to the intracellular region of the plasma membrane, is required for the formation of large myotubes (). Finally, the lectin wheat germ agglutinin inhibits the second stage of fusion in vitro (), suggesting that carbohydrate binding proteins likely play an important role during this phase of fusion. The mechanisms by which these molecules regulate the second stage of myoblast fusion have not been identified. The mannose receptor (MR) is a 175-kD type 1 transmembrane protein that binds a variety of soluble and cell surface glycoproteins (; ; ; ; ) and is one of four members of the MR family of proteins (). The extracellular region of MR consists of three types of domains: an N-terminal cysteine-rich domain that confers MR's ability to bind sulfated sugars (), a region of fibronectin type II repeats responsible for binding collagen (; ; ), and eight carbohydrate recognition domains, providing terminal mannose, fucose, N-acetylglucosamine, and glucose binding ability in a calcium-dependent manner (). MR is an endocytic receptor and contains a 45-amino-acid cytoplasmic region thought to be responsible for receptor internalization (). MR is expressed in a variety of tissues and has been proposed to function in serum glycoprotein clearance, antigen transport and presentation, and immune cell recognition of foreign microbes (; ; ; ). Several lines of indirect evidence have suggested that terminal mannose residues or MR may function in cell fusion. MR expression increases in hematopoietic precursors undergoing differentiation and peaks during cell fusion to form osteoclasts or multinucleated giant cells (). High mannose mannan, which binds MR with high affinity, inhibits the fusion of macrophages during multinucleated giant cell formation in vitro (). In addition, the mannose binding compound pradimicin and an inhibitor of glucosidase I, an enzyme required for high mannose oligosaccharide expression, prevent the fusion of hematopoietic precursor cells during osteoclast formation (; ). MR also functions in cell–cell adhesion, as antibodies that recognize MR inhibit lymphocyte adhesion to endothelial cells in vitro (). IL-4 signaling regulates MR expression in several cell types (; ; ; ; ). Because IL-4 is a known regulator of myoblast fusion and because carbohydrate binding proteins have been implicated in fusion, we hypothesized that MR may have an important function during myogenesis. Here, we demonstrate that MR is required for myoblast fusion with nascent myotubes in vitro and for proper skeletal muscle growth in vivo. We also provide the first evidence that MR plays an important role in cell motility, as MR cells have impaired migratory speed during myoblast fusion in vitro. In addition, we show that the collagen uptake is impaired in MR cells and that MR is required for directed cell migration during myotube growth. Importantly, these data identify a novel function for MR during skeletal muscle growth and have a broad implication for MR regulation of cell motility. To determine whether MR is expressed in muscle cells during fusion, myoblasts were induced to differentiate by switching to differentiation media (DM) for 0, 24, or 48 h. After 24 h in DM, myoblasts fused to form small, nascent myotubes, and after 48 h, large myotubes had formed (). RT-PCR analyses revealed that MR mRNA levels increased after the onset of myoblast fusion and remained elevated at 48 h (). RT-PCR analyses of myogenin expression, a marker of myogenic differentiation (), demonstrated that the increase in MR expression was concurrent with the onset of differentiation. Immunostaining of muscle cells after 24 h in DM with an antibody against the intracellular portion of MR () revealed that MR protein was present in both mononucleated cells and nascent myotubes (). No immunostaining was present in MR muscle cells (), indicating the specificity of the antibody. IL-4 signaling regulates MR expression in several cell types (; ; ; ; ). IL-4–mediated regulation of MR expression in muscle cells was assessed in two experiments. First, nascent myotubes were treated with recombinant IL-4 for 24 h. RT-PCR analyses indicated that MR mRNA levels increased in myotubes treated with IL-4 (). Conversely, MR mRNA expression was reduced in myotubes deficient of the IL-4 receptor (, IL-4Rα). Together, these data suggest that IL-4 signaling regulates MR expression in fusing myoblasts. To test the hypothesis that MR is involved in myoblast fusion, we examined the ability of myoblasts derived from wild-type (WT) or MR mice () to form myotubes in vitro. After 20 or 48 h in DM, cells were immunostained with an antibody against embryonic myosin heavy chain (eMyHC; ), marking the cytoplasm of differentiated muscle cells and clearly defining the nuclei of myotubes. After 20 h in DM, MR myoblasts fused to form small myotubes indistinguishable from WT myotubes. However, by 48 h in DM, WT myoblasts formed large myotubes, whereas MR myotubes remained small. The impaired growth of MR myotubes could arise from several factors, including defects in proliferation, differentiation, or fusion. To assess the requirement of MR in myoblast proliferation, WT and MR myoblasts were pulsed for 1 h with BrdU. The percentage of BrdU cells was similar in WT and MR cells, indicating that MR is not required for myoblast proliferation (Fig. S1 A, available at ). To determine whether MR myoblasts underwent impaired or delayed differentiation, we assessed expression of two markers of myogenic differentiation. Immunoblots were performed to examine myogenin expression at 16 h in DM, before myoblast fusion. WT and MR cells expressed similar levels of myogenin (), demonstrating that early stages of myogenic differentiation were not disrupted in MR cells. The percentage of nuclei found in eMyHC cells after 48 h in DM was not decreased in MR cells (), indicating that MR is not required for the later stages of myogenic differentiation. In addition, similar numbers of nuclei were present in WT and MR cultures after 48 h in DM (Fig. S1 B), indicating that cell survival is not disrupted in MR cells during differentiation and fusion. To determine whether MR myoblasts form small myotubes as a result of defects in myoblast fusion, two types of fusion analyses were performed. The fusion indices were calculated as the percentage of nuclei located in myotubes (≥2 nuclei) after 48 h in DM and were similar for WT and MR cells (), indicating that MR cells do not have a general defect in myoblast fusion. The number of nuclei contained within WT and MR myotubes was next quantified (). After 20 h in DM, MR myotubes contained the same mean number of nuclei as WT myotubes, indicating that the first stage of myoblast fusion is not disrupted in MR cells. After 48 h in DM, however, MR myotubes contained significantly fewer nuclei than WT myotubes. Importantly, MR cells do not form small myotubes as a result of delayed myoblast fusion, as the number of nuclei in MR myotubes remained low, even after 72 h in DM. These data suggest that MR is required for the second stage of myoblast fusion, during which myoblasts fuse with nascent myotubes (). Our in vitro data establish a role for MR during myoblast fusion. To determine whether MR plays a functional role in skeletal muscle in vivo, we examined myofiber size in WT and MR muscles. The tibialis anterior (TA) muscles were collected from adult WT and MR mice, and sections were stained with hematoxylin and eosin (H&E; ). As confirmed by cross-sectional area (XSA) analyses (), MR myofibers were significantly smaller than WT myofibers. In addition, WT muscles contained a higher percentage of large myofibers, whereas MR muscles contained a higher percentage of small myofibers (). Myonuclear number analyses were performed (; ) on TA muscle sections to determine whether MR myofibers contain fewer myonuclei than WT myofibers, as was observed in myotubes in vitro. MR myofibers contained significantly fewer myonuclei than WT myofibers (), suggesting that the reduced XSA of MR myofibers is at least partially due to a decrease in myonuclear number (). XSA analyses were also performed on WT and MR soleus muscles to ensure that the reduced myofiber size was not specific to the TA. Mean myofiber XSA was also significantly reduced in MR soleus muscles (). However, the number of myofibers in MR soleus muscles was not significantly different than WT (). Together, these data suggest that MR is required for developmental muscle growth or maintenance in vivo. To examine MR function specifically in skeletal muscle growth, we analyzed myofiber growth in WT and MR mice after muscle injury. BaCl was injected into the TA muscles of adult mice to induce injury (; ). After 5, 7, and 14 d of regeneration, muscles were collected, sectioned, and stained with H&E (). XSA analyses revealed that WT and MR myofibers were similar in size at early stages of muscle repair (5 d after injury), but MR myofibers were impaired in growth at later stages (7–14 d after injury; ). By 14 d of regeneration, both WT and MR myofibers had returned to their respective uninjured size. These data provide further evidence for the requirement of MR function during the later stages of muscle growth. MR may function in mononucleated cells and/or nascent myotubes during the second stage of myoblast fusion. If MR functions in mononucleated cells, MR mononucleated cells should not be recruited to fuse with WT nascent myotubes. To test this hypothesis, WT nascent myotubes were cocultured with MR mononucleated cells in DM for 24 h (). Before coculture, each cell population was stained with a fluorescent dye (). After coculture, myotubes were analyzed for the presence of both fluorescent dyes. Coculture of WT nascent myotubes with WT mononucleated cells resulted in 77% of myotubes containing both fluorescent dyes (). In contrast, coculture of WT nascent myotubes with MR mononucleated cells resulted in only 37% of myotubes containing both fluorescent dyes, indicating that MR mononucleated cells are impaired in their ability to fuse with nascent myotubes. To determine whether MR functions in nascent myotubes, MR nascent myotubes were cocultured with WT mononucleated cells. After coculture, ∼63% of myotubes contained both fluorescent dyes. These results are not statistically different from WT/WT coculture, suggesting that MR function is not also required in nascent myotubes. To confirm the requirement of MR function during the second stage of fusion, we cocultured MR nascent myotubes with MR mononucleated cells. As expected, MR deficiency led to a significant reduction in myoblast fusion with nascent myotubes, as only 32% of myotubes contained both fluorescent dyes. Together, these data suggest that mononucleated cells are the primary site of MR function during their fusion with nascent myotubes. To determine whether the impaired ability of MR mononucleated cells to fuse with nascent myotubes results specifically from a loss of MR function, MR expression was restored in MR muscle cells via retroviral infection (). Coculture of MR mononucleated cells infected with a MR retrovirus (RV) significantly increased the ability of these cells to fuse with WT nascent myotubes compared with MR mononucleated cells infected with a control RV (). Together, these results indicate that MR is required for proper fusion of mononucleated cells with nascent myotubes. We hypothesized that MR may regulate the second stage of myoblast fusion by influencing cell–cell adhesion or cell motility. Cell–cell adhesion assays indicated that MR muscle cells were not defective in their ability to adhere with one another in suspension (unpublished data). To determine whether MR regulates muscle cell motility, we performed time-lapse microscopy of WT and MR cells undergoing fusion in vitro. After 0 or 24 h in DM, cell movements were recorded every 5 min for 3 h. The paths of individual mononucleated cells were tracked, revealing that WT cells migrated farther than MR cells (). Additionally, the mean velocity of MR cells was reduced 23% compared with WT cells after 24 h in DM (), with a greater percentage of WT cells migrating at high velocities compared with MR cells (). Importantly, retroviral-mediated MR expression in MR cells significantly increased the mean cell velocity compared with MR cells infected with a control RV (). To ensure that retroviral infection of myoblasts does not alter cell motility, we assessed the migration of control or RV-infected WT cells after 24 h in DM. The motility of RV-infected cells was not disrupted (Fig. S2, available at ), suggesting that the differences in cell velocity shown in are due to variability between sets of cell isolates and not the infection process. The mean velocity of MR cells before the first stage of myoblast fusion (0–3 h in DM) was not significantly different from WT (). These data demonstrate the requirement of MR for efficient motility of myogenic cells during their fusion with nascent myotubes. The decreased velocity of MR cells during myoblast fusion may result from a defect in random or directed cell migration. To distinguish between these possibilities, we tested the ability of MR cells to respond to a chemotactic gradient. If MR is required for a directional response of muscle cells to a chemoattractant during myotube growth, we reasoned that such a factor should be present in conditioned media from nascent myotube cultures. Dunn chemotaxis chambers were used to establish a gradient of conditioned media, and the migratory response of muscle cells was observed over 3 h by time-lapse microscopy. The paths of individual cells were tracked, and the final location of each cell in relation to its origin was determined. Directional data were summarized in circular histograms, and statistical tests revealed that WT but not MR cells migrated up a gradient of conditioned media (). Conditioned media also contains chemokinetic properties not dependent on MR, as the velocity of both WT and MR cells increased 1.2–1.5-fold in the presence of a conditioned media gradient (). The mean velocity of MR cells was significantly lower than WT cells in the presence of control or conditioned media, confirming that MR is required for efficient motility of muscle cells in addition to functioning in directed migration. The MR family member Endo180 plays a role in both directed and random cell migration (; ). Endo180 is thought to facilitate cell motility via clearance of collagen, a component of the ECM. Degradation of the ECM is an important step in facilitating cell migration during tissue development, regeneration, and homeostasis (). Recently, MR was shown to bind collagen, most likely through its fibronectin type II repeats (; ). To determine whether MR facilitates collagen clearance in muscle cells, we performed uptake assays with I-labeled type IV collagen. Differentiating MR muscle cells internalized significantly less collagen than WT cells (). Together, these results demonstrate that MR functions in directed migration of muscle cells and suggest that MR facilitates cell motility by internalizing collagen during myotube growth. Skeletal muscle formation, growth, and regeneration rely on the fusion of mononucleated myoblasts with one another and with existing myofibers. Myoblast fusion is dependent on a series of cellular events, including myoblast differentiation, migration, adhesion, and membrane breakdown. Disruption of any of these processes may inhibit myoblast fusion. The molecular pathways regulating myoblast fusion in mammals are largely unclear. Here, we show that MR, a type I transmembrane protein, is required for the normal fusion of myoblasts with nascent myotubes. MR plays an important role in muscle cell motility, as MR myoblasts migrate at reduced velocity during myotube growth and directed migration up a chemoattractant gradient is ablated. In addition, collagen uptake is impaired, suggesting a role for MR in ECM remodeling during cell migration. Myoblast fusion in mammals occurs in two phases (). Initially, myoblasts fuse with one another to form small, nascent myotubes. Subsequently, myonuclear accretion occurs through the fusion of additional myoblasts with nascent myotubes. Our data indicate that MR is required during the second stage of myoblast fusion. Two lines of evidence suggest that MR is not required during the first phase of myoblast fusion. First, at early stages of myotube formation in vitro, WT and MR nascent myotubes contained similar numbers of nuclei (). Subsequently, WT myotubes continued to accumulate nuclei through additional rounds of myoblast fusion, whereas MR myotubes did not. Second, early phases of regeneration in vivo were similar in WT and MR muscles, but MR muscles were defective in later stages of muscle regeneration (). Furthermore, the myofibers of adult MR-null mice were significantly reduced in XSA compared with WT myofibers. Importantly, the reduced myofiber XSA of MR mice correlated with a decrease in myonuclear number (). These data indicate that the MR is also required for proper developmental myofiber growth or maintenance in vivo. Thus, MR is necessary for the fusion of myoblasts with nascent myotubes/myofibers both in vitro and in vivo. Cell mixing experiments demonstrated that MR function is required in myoblasts, as MR myoblasts were deficient in their ability to fuse with nascent myotubes. This defect was due specifically to the loss of MR, as retroviral-mediated MR expression in MR myoblasts restored their ability to fuse with nascent myotubes (). However, MR protein was present in both myoblasts and nascent myotubes at 24 h of differentiation in vitro (). This discrepancy between expression and function may be explained if levels of cell-surface MR protein are regulated differentially in myoblasts and myotubes. Members of the MR family of proteins are constitutively recycled from the plasma membrane, and estimates have been made that only ∼10–30% of total MR protein is present on the cell surface at any point in time (). What is the cellular mechanism by which MR acts in myoblasts to regulate the second stage of myoblast fusion? We hypothesized that MR may function in myogenic cell–cell adhesion or cell migration. MR has previously been implicated in adhesion of leukocytes to human lymphatic endothelium via interaction with the cell-adhesion molecule -selectin (). However, MR myogenic cells were capable of adhering to one another in suspension-based assays (; ; ; ). In contrast, MR myoblasts displayed decreased velocity and distance of migration during myotube growth (). Importantly, restoration of MR expression via retroviral infection of MR cells significantly increased the velocity of MR cells, indicating that MR is required for efficient myoblast migration. As expected, the migration of MR cells was not disrupted at early times in DM, as MR mRNA levels were very low () and protein levels were undetectable (not depicted) at the initiation of differentiation. These data are in agreement with the findings that MR is not required for the first phase of fusion. Interestingly, a protein related to MR, Endo180 (also referred to as urokinase-type plasminogen activator receptor–associated protein, or UPARAP) is required for efficient motility of fibroblasts, suggesting that members of the MR family of proteins may share a common role in facilitating cell migration (; ). These data provide the first evidence that MR plays a role in cell motility. Cell migration during tissue development and remodeling involves both a directed cellular response to chemoattractant factors and the breakdown of the ECM (; ). Degradation of the ECM by extracellular proteolytic enzymes facilitates cell motility, whereas chemotaxis involves the movement of cells to a specific location in response to directional signals. Our data reveal that MR is required for the directed migration of muscle cells up a conditioned media gradient (). We propose that nascent myotubes secrete factors necessary for the directed migration of myoblasts during fusion and that MR is required for the directional response of cells to at least one of these factors. The factors responsible for MR-dependent chemotaxis are unknown. A chemoattractant may bind the extracellular region of MR directly. Engagement of the MR by an extracellular ligand may initiate an intracellular signaling cascade necessary for providing directional cues to the cell. However, no characterized signaling domains have been identified in the MR cytoplasmic tail. Alternatively, MR may act as a coreceptor for a chemoattractant. For example, Endo180 interaction with the GPI-anchored urokinase plasminogen activator receptor is required for directed cell migration up a urokinase plasminogen activator gradient (). The mechanism by which MR mediates directed muscle cell migration is currently under investigation. In addition to impaired directional migration, MR cells migrated at a reduced velocity during myoblast fusion () and in the presence of control or conditioned media (). These data suggest that MR may also facilitate the random motility of muscle cells. Endo180 is thought to facilitate the motility of fibroblasts via clearance of the ECM component collagen (). MR has recently been shown to bind several forms of collagen, and internalization of collagen IV by macrophages is dependent on the presence of MR (; ). Our results revealed that MR muscle cells were impaired in the uptake of type IV collagen (). However, unlike MR macrophages, collagen uptake was not ablated in MR muscle cells. Endo180, which is expressed in fusing muscle cells (unpublished data), may enable the uptake of collagen in the absence of MR. We hypothesize that MR regulates cell motility by facilitating collagen clearance by muscle cells. Although MR cells migrated at a reduced velocity, their migration was not ablated, suggesting that additional migratory signals are functioning in the absence of MR. The migration of muscle precursor cells during embryonic development and postnatal regeneration is essential to the formation and maintenance of mammalian skeletal muscle. A variety of molecules, including growth factors, cytokines, chemokines, ECM components, proteolytic enzymes, and intracellular signaling proteins have been implicated in cell migration. Hepatocyte growth factor (HGF) and its receptor, c-Met, are required for the migration of muscle precursor cells from the dermomyotome to the limbs (). Several growth factors, including HGF, bFGF, PDGF A and B, LIF (leukemia inhibitory factor), TGF-β, and IGF-1 (insulin-like growth factor I) induce myoblast migration in vitro (; ; ). The cytokines TNF-α and IFN-γ and the chemokine RANTES also enhance myoblast migration in vitro (; ). ECM components such as laminin () and proteoglycans () as well as extracellular proteolytic enzymes, including matrix metalloproteinases and calpain (; ), influence myoblast motility. Studies of intracellular signaling pathways involved in myoblast migration indicate that HGF induces myoblast migration via activation of Ras and phosphatidylinositol 3-kinase and their downstream effectors (; ). The precise relationship among these various molecules during myoblast migration remains unclear. Our results suggest, but do not directly prove, the importance of myoblast migration for myotube growth. Further roles for MR may contribute to the defect in the fusion of MR myoblasts with nascent myotubes. For example, MR may regulate cell–cell interactions among myogenic cells. MR is known to bind a variety of glycosylated proteins in other cell types and may aid in the recognition of myoblasts and myotubes by interacting with a ligand or ligands on the surface of opposing cells. Identification of MR ligands in skeletal muscle will provide further insight into the mechanisms by which muscle growth is regulated. Understanding the molecular pathways involved in myoblast migration, adhesion, and fusion is important in designing treatments for impaired muscle growth associated with age, disease, and atrophy. In addition, promotion of cell fusion may aid in cell therapy protocols using exogenous stem cells (; ). MR mice produced on the 129vJ × C57BL/6 background and backcrossed to C57BL/6 mice for seven generations were provided by M. Nussenzweig (The Rockefeller University, New York, NY; ). Additional MR mice were generated by homozygous matings. Control age- and sex-matched C57BL/6 mice were purchased from Charles River Laboratories. Adult mice between 8–12 wk of age were used for all studies. All animals were handled in accordance with the institutional guidelines of Emory University. Primary myoblasts were derived from the hindlimb muscles of adult female WT or MR mice as previously described with the exception of a percoll gradient (; ). In brief, muscles were minced mechanically and digested with 0.1% pronase (Calbiochem) in DME containing 25 mM Hepes at 37°C with slight agitation for 1 h. The muscles were further dissociated by trituration and passed through a 100-μm filter. Cells were suspended in growth media (GM; Ham's F10, 20% FBS, 5 ng/ml bFGF, 100 U/ml penicillin G, and 100 μg/ml streptomycin) and grown on collagen-coated dishes in a humidified 5% CO incubator at 37°C. Primary cultures were enriched for myogenic cells to >99% purity using the preplating technique as described previously (). To induce differentiation, cells were plated on dishes coated with entactin–collagen IV–laminin (E-C-L; Upstate Biotechnology) in GM and shortly thereafter switched to DM (DME, 1% Insulin-Transferrin-Selenium-A supplement [Invitrogen], 100 U/ml penicillin G, and 100 μg/ml streptomycin). For analysis of MR mRNA expression, WT and IL-4Rα were derived and grown as described previously (). In experiments using exogenous cytokines, vehicle or 10 ng/ml recombinant mouse IL-4 (R&D Systems) was added to cells after 24 h in DM and RNA was isolated 24 h later. RNA was isolated from primary muscle cells using TRIzol reagent (Life Technologies) according to the manufacturer's instructions. Reverse-transcriptase reactions were performed using 2.5 μg of total RNA. cDNA was amplified using Expand High Fidelity PCR system (Roche) with primers specific for MR (available under GenBank/EMBL/DDBJ under accession no. ; sense, 5′ AGTGATGGTTCTCCCGTTTCCTAT; antisense, 5′ TGACTGCCCACCATTCTTGTTTAT) or myogenin (accession no. ; sense, 5′ AGCGGCTGCCTAAAGTGGAGAT; antisense, 5′ GGACGTAAGGGAGTGCAGATTGTG). All primer pairs spanned intron and exon boundaries to control for any contaminating DNA in RNA samples. MR cDNA was amplified by incubation at 94°C for 5 min, followed by 35 cycles of 94°C for 30 s, 54°C for 30 s, and 72°C for 30 s, and terminating at 72°C for 5 min, generating a 390-bp amplicon. Myogenin cDNA was amplified by incubation at 94°C for 5 min, followed by 25 cycles of 94°C for 30 s, 60°C for 30 s, and 72°C for 30 s, and terminating with 72°C for 5 min, generating a 266-bp amplicon. Amplicons were separated by electrophoresis in a 1% agarose gel and visualized with ethidium bromide. RT-PCR analysis of 18S ribosomal RNA was included as a control for each sample using QuantumRNA 18S primers (Ambion). For detection of MR protein by immunofluorescence, WT and MR primary myoblasts were differentiated for 24 h and subsequently fixed in 3.7% formaldehyde for 10 min. Cells were then incubated in block buffer (PBS containing 0.25% Triton X-100 and 5% donkey serum) for 1 h, followed by incubation with a polyclonal antibody recognizing the cytoplasmic tail of MR (provided by A. Regnier-Vigouroux, Deutsches Krebsforschungszentrum, Heidelberg, Germany; ) diluted 1:500 in block buffer for 1 h. After several washes in PBS + 0.2% Tween 20 (PBS-T), the cells were incubated with biotin-conjugated donkey anti–rabbit IgG (Jackson ImmunoResearch Laboratories) diluted 1:500 in block buffer for 1 h. Cells were washed with PBS-T and subsequently incubated with streptavidin-horseradish peroxidase diluted 1:250 in block buffer for 30 min. The Tyramide Signal Amplification green reagent (NEN Life Science Products) was used to visualize antibody binding. Fluorescence images were acquired using a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with a 0.3 NA 10× Plan-Neofluar objective (Carl Zeiss MicroImaging, Inc.) and camera (QImaging) with OpenLab 3.1.4 (Improvision). Cells were stored in PBS at room temperature for all image acquisition. Images were assembled using Photoshop 7.0 (Adobe) software and were not modified with the exception of equal adjustments in size, brightness, and contrast. Primary myoblasts from WT and MR mice were seeded on E-C-L–coated 6-well dishes at a density of 2 × 10 cells/well in GM. Cells were allowed to adhere to the dish for ∼1 h before switching to DM. After 20 or 48 h in DM, cells were fixed in 3.7% formaldehyde for 10 min and subsequently immunostained with an antibody against eMyHC (F1.652; Developmental Studies Hybridoma Bank) as described previously (). The differentiation index was determined by dividing the total number of nuclei in eMyHC-positive cells by the total number of nuclei counted. The mean number of nuclei per myotube was determined by dividing the total number of nuclei in myotubes (≥2 nuclei) by the total number of myotubes counted. The fusion index was determined by dividing the total number of nuclei in myotubes by the total number of nuclei counted. At least 100 myotubes and 500 nuclei per condition were analyzed for each assay. To assess myogenin expression, three independent WT and MR cell isolates were differentiated for 0 or 16 h and were subsequently lysed in RIPA-2 buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% NP-40, 0.5% deoxycholic acid, and 0.1% SDS) containing protease inhibitors (Mini complete; Roche) for 10 min on ice. Lysates were spun at 21,000 for 15 min at 4°C. Protein concentration was determined using the Bradford assay (Bio-Rad Laboratories), and 25 μg of total protein was separated by SDS-PAGE. After transfer to a polyvinylidene difluoride membrane (Millipore), myogenin protein was detected using a mouse monoclonal antibody (F5D; Developmental Studies Hybridoma Bank) diluted 1:10 in block buffer as described previously (). Membranes were stained with Coomassie (Bio-Rad Laboratories) to confirm equal loading. TA and soleus muscles were collected from adult male mice ( = 5–6) as described previously (). Serial 14-μm sections were collected along the entire length of each muscle and stained with H&E. Histological analyses were performed on sections collected from similar regions of each TA muscle and the belly of each soleus muscle. Two images were captured from each section, and Scion Image 1.63 (Scion Corp.) was used to determine the XSA of 50–100 myofibers per field. All photography was performed on a microscope (Axioplan; Carl Zeiss MicroImaging, Inc.) with a 0.3 NA 10× Plan-Neofluar objective (Carl Zeiss MicroImaging, Inc.) equipped with a charge-coupled device camera (Carl Zeiss MicroImaging, Inc.). Pictures were assembled using Photoshop 7.0 and were not modified other than adjustments of size, color levels, brightness, and contrast. In vivo myonuclear number analyses were performed as described previously (). In brief, sections of TA muscles from WT and MR mice ( = 5–6) were immunostained with an antibody against dystrophin (MANDYS8; Sigma-Aldrich) to visualize the sarcolemma of myofibers and mounted in Vectashield mounting media containing DAPI (Vector Laboratories) to stain nuclei. Nuclei within dystrophin-positive sarcolemma were counted for 50–100 myofibers, and the number of nuclei was expressed per 100 myofibers. To analyze muscle growth during regeneration, injury was induced in the TA muscles of WT and MR mice ( = 5–6) by injection of 50 μl of 1.2% BaCl diluted in PBS with a 27-gauge needle (; ) along the length of the muscle. Muscles were collected 5, 7, or 14 d after injury, and XSA of centrally nucleated regenerating fibers was assessed as described above. A retroviral vector encoding full-length MR (provided by L. Martinez-Pomares, Queen's Medical Center, Nottingham, UK; ) and a control vector (pFB-neo; Stratagene) were used to produce infectious retroviral supernatants as described previously (). Primary WT and MR myoblasts were subjected to two rounds of infection (), and infected cells were selected by growing cells with 50 μg/ml of Geneticin (Invitrogen) in GM. Cell mixing experiments were performed as described previously with minor modifications (). Primary myoblasts were grown at low density (0.5 × 10 cells per well of a 6-well plate) or high density (2 × 10 cells per well of a 6-well plate) in DM for 24 h to generate differentiated mononucleated cells or nascent myotubes, respectively. Mononucleated cells were incubated with CellTracker Orange CMTMR (5-(and-6)-(((4-chloromethyl) benzoyl) amino) tetramethylrhodamine) (Invitrogen) diluted to 2.5 μM in DM, and nascent myotubes were incubated with CellTracker Green CMFDA (5-chloromethyl-7-hydroxycoumarin; Invitrogen) diluted to 0.5 μM in DM for 10 min at 37°C. Cells were washed twice with PBS, trypsinized, mixed at equal cell number, and plated to give a final cell number of 2 × 10 cells per well of a 6-well E-C-L–coated plate. After 24 h in DM, the cells were fixed for 10 min in 3.7% formaldehyde. The presence of dual label was analyzed in 50–100 myotubes with ≥3 nuclei. Mixing experiments were performed in triplicate using WT and MR myoblasts from three independent cell isolates. Primary myoblasts were seeded on E-C-L–coated 35-mm plates at a density of 2 × 10 cells per plate in GM. After allowing cells to adhere for ∼1 h, cells were switched to DM. At 0 or 24 h in DM, 25 mM Hepes was added to the cultures and cells were transferred to a microscope stage heated to 37°C. Cell migration was visualized using a Axiovert 200M microscope with a 0.3 NA 10× Plan-Neofluar objective (Carl Zeiss MicroImaging, Inc.), and images were recorded (QImaging camera and OpenLab 3.1.4 software) every 5 min for 3 h. Cell velocities were calculated in micrometers per hour using ImageJ software by tracking the paths of individual mononucleated cells. Cell migration assays were performed for each genotype using three independent cell isolates. The mean velocities of 45–50 cells (∼15 cells per isolate) were pooled and analyzed for statistical significance as described (see Statistics). Permanox plastic cell culture slides (Nunc) were cut into 6-cm squares, and an ∼1-cm region of each slide was coated with E-C-L for 1 h at 37°C. Primary myoblasts were then seeded at a density of 5 × 10 cells per slide in GM. Cells were allowed to adhere for 1 h, and GM was replaced with DM. The low density at which the cells were plated ensured that cells underwent myogenic differentiation with limited cell fusion. After 24 h in DM, the Dunn chamber was assembled as described previously (, ). DM that had been conditioned by differentiating primary muscle cells for 24 h was collected before chamber assembly and supplemented with 25 mM Hepes. To set up gradient experiments, both concentric wells of the chemotaxis chamber were filled with control DM (supplemented with 25 mM Hepes), and the slide containing differentiating cells was inverted onto the chamber to cover both wells. The slide was sealed onto the chamber with a hot 1:1:1 mixture of paraffin wax, beeswax, and petroleum jelly, leaving a small slit of the outer well open. DM was removed from the outer well and replaced with control or conditioned media, and the slit was sealed. After allowing the gradient to establish for 30 min at 37°C, a small region over the annular bridge was visualized and cell migration was analyzed by time-lapse microscopy as above (see Cell migration assays). Statistical analyses of directional data were performed to assess the chemotactic response of the cells as described previously (). Each cell path was converted to a trajectory originating from on an x-y axis. A horizon distance for each condition was established by determining the distance passed by 50% of the cells in a straight line from their starting point. The horizon method is designed to assess the directionality of cell movement without influence from differences in cell motility. Cells that fail to reach the horizon distance were excluded from directional analysis. A trajectory angle for each cell was calculated as the direction of each cell from its starting point to the point at which the cell crossed the horizon distance. The directional data were summarized as circular histograms in which the area of each sector represents the proportion of trajectory angles located within each 18° interval. The Rayleigh test for unimodal clustering was applied with P < 0.05 as the criterion for rejecting the null hypothesis of uniform distribution. Where unimodal clustering was observed, a mean direction and 95% confidence interval were calculated. Statistical analysis was performed using Oriana 2.0 (RockWare). Dunn chamber assays were performed using three independent cell isolates. Directional analyses were performed using at least 15 cells per assay for a total of 45 cells. Collagen internalization assays were performed as described previously (). Type IV collagen (Calbiochem) was labeled with I via Iodogen (PerkinElmer), resulting in a specific activity of 88 μCi/mg. Primary myoblasts were differentiated for 24 h as described above (see Differentiation and fusion assays). After 24 h in DM, 1 nM I-collagen was added to the cells. After a 4-h incubation at 37°C, the medium was removed and cells were washed twice with PBS to remove unbound collagen. Cells were treated with 0.2% type I collagenase (Worthington) diluted in 0.05% trypsin and 0.53 mM EDTA (Invitrogen) to lift cells and cleave cell surface bound collagen. In pilot experiments, 0.2% type I collagenase treatment released >95% of cell surface bound collagen. The detached cells were centrifuged at 1,000 for 5 min, and the radioactivity of the supernatant (cell surface released collagen) and pellet (internalized collagen) was measured using a gamma counter (1470 WIZARD; Wallac). To determine significance between two groups, comparisons were made using tests. Analyses of multiple groups were performed using a two-way analysis of variance with Bonferroni's posttest. Statistical analyses were performed using GraphPad Prism 4.0 (GraphPad) for Macintosh or SigmaStat 2.03 (SPSS). For all statistical tests, a confidence level of P < 0.05 was accepted for statistical significance. Fig. S1 shows that cell proliferation and cell survival are not disrupted in MR cells. Fig. S2 demonstrates that retroviral infection does not alter myoblast motility. Online supplemental material is available at .

The creation of a precise morphology in which a neuron generates multiple dendrites and one long axon is essential for the formation of neuronal circuitry. The establishment of axon–dendrite polarity is an important feature of neurons (). The primary cultured hippocampal neuron is an established model for the characterization of neuronal polarity (). Cultured hippocampal neurons extend several minor neurites after plating, which remain indistinguishable in stages 1 and 2, after which one of them develops into an axon at stage 3. In contrast, the others develop into dendrites (; ). Local activity of the phosphatidylinositol (PI) 3-kinase–Akt–glycogen synthase kinase 3β (GSK-3β) pathway is required for both the establishment and maintenance of neuron polarity in these neurons (, ; ; ; ; ). A recent study suggested that polarized growth occurs before neurites are formed (). PI 3-kinase is activated at the tip of the newly specified axon to stimulate Akt kinase (; ). Activated Akt then phosphorylates and inactivates GSK-3β, turning neurites to axons (, ; ; ; ; ). Furthermore, active Akt is found in the soma and axon terminus but not in other neurites, and the expression of constitutively active Akt leads to the formation of multiaxons (; ). Therefore, activation of Akt in the axon is critical for axon formation (). However, the mechanism through which the asymmetrical activation of Akt is established remains unknown. Protein degradation by the ubiquitin (Ub)–proteasome system (UPS) is important for the regulation of many cellular functions, including cell cycle, growth, and polarity (; ; ; ; ). In response to various stimuli, the UPS, which involves the sequential action of Ub-activating enzymes (E1), Ub-conjugating enzymes (E2), and Ub ligases (E3), can be activated, resulting in the conjugation of Ub to the lysine residues of proteins (; ). Those proteins tagged with poly-Ub are then degraded by the proteasome complex. Because Akt stability in different types of cells is regulated by the UPS (; ; ; ; ), it is possible that the asymmetrical activation of Akt is caused by its selective distribution mediated by the UPS. In this study, we have examined the role of the UPS in neuronal polarity and found that selective degradation of Akt by the UPS in dendrites is required for generating neuronal polarity. To test whether the UPS is involved in neuronal polarity, we first examined the effect of UPS inhibition on axon–dendrite specification in cultured hippocampal neurons. As shown in , UPS inhibition by MG132 and lactacystin, two agents known to inhibit the proteasome, led to the loss of neuron polarity and formation of multiple axons. The percentages of neurons with no axon, a single axon, or multiple axons were 7.33 ± 1.15, 83.33 ± 1.15, and 9.33 ± 2.31%, respectively, in neurons treated with DMSO, whereas the percentages were 9.00 ± 4.58, 31.33 ± 2.31, and 59.67 ± 6.81%, respectively, in neurons treated with MG132 ( = 100; three experiments; ). Similarly, lactacystin dramatically reduced the number of neurons with a single axon and increased the number of neurons with multiple axons (). Furthermore, expressing K48R-Ub, a dominant-negative form of Ub known to inhibit the UPS (), markedly reduced the number of neurons with a single axon and increased the number of neurons with multiple axons, whereas expressing a control vector or the wild-type (WT) Ub did not affect neuron polarity ( = 100; three experiments; ). UPS inhibition also increased the number of axons and extended or maintained the mean length of axons (Fig. S1, A–D; available at ). These results suggest that the UPS is critical for the formation of neuronal polarity. To test whether the UPS is required for both the establishment and maintenance of neuronal polarity, we treated neurons with MG132 before (from 12 to 48 h after plating) or after (from 48 to 96 h) the establishment of neuronal polarity (most of the neurons have established polarity 48 h after plating; Fig. S1 E). As shown in ( = 100; three experiments; E and F), treatment with MG132 before or after the establishment of neuronal polarity dramatically increased the number of neurons with multiple axons. These results indicate that the UPS is also required for maintenance of neuronal polarity. To further support this notion, we transfected neurons with WT-Ub or K48R-Ub 48 h after plating. The expression of K48R-Ub but not WT-Ub led to the formation of multiple axons in the neurons in which the polarity had been previously established ( = 100; three experiments; Fig. S1, F and G). Although these axons were Tau1 and synapsin I positive, it is not clear whether they were functional. We then examined the uptake and release of FM4-64 dye, which is an assay used to show the functionality of axons in synaptic vesicle recycling (). As shown in Fig. S1 (H and I), FM4-64 dye was taken up only by axon termini but not by dendrites in the neurons treated with DMSO or transfected with WT-Ub. Moreover, in the presence of 45 mM KCl, the dye was taken up by the axons formed after treatment with MG132 or after transfection with K48R-Ub. The dye was then released in the presence of 90 mM KCl (Fig. S1, H and I). Therefore, these axons were functional in synaptic vesicle recycling. Collectively, these results suggest that UPS was required for both the establishment and maintenance of neuronal polarity. To identify the molecules degraded by the UPS during the formation of neuronal polarity, we examined the protein levels for P110 (a subunit of PI 3-kinase), Akt, the phosphorylated form of Akt (p-Akt), GSK-3β, and the phosphorylated form of GSK-3β (p–GSK-3β) in the neurons treated with MG132 and lactacystin from 12 to 48 h after plating. As shown in , protein levels of Akt in the treated cultures were greatly increased as compared with those in the control. Similarly, in the neurons treated with MG132 and lactacystin, the levels of p-Akt and p–GSK-3β were also markedly increased as compared with those in control cultures (). However, the two inhibitors did not affect the protein levels of P110 or GSK-3β (). Consistent with these results, the protein levels and activity of Akt but not those of P110 and GSK-3β were greatly enhanced by transfection with K48R-Ub (). The aforementioned results point to the possibility that Akt is degraded by the UPS during the formation of neuronal polarity. It has been known that ubiquitination of a protein is a critical step for its degradation by the UPS. We then examined whether Akt was ubiquitinated in the stage 1–3 neurons. As shown in , Akt was found in complexes precipitated by the antibody against Ub, and the levels of Akt in these complexes were increased in stage 2 neurons compared with those in stage 1 and 3 neurons (). In contrast, P110 and GSK-3β were not ubiquitinated in stages 1–3 (Fig. S1 J). Furthermore, treatment with MG132 or lactacystin from 12 to 24 h after plating greatly increased the level of the ubiquitinated form of Akt (Fig. S1 K). These results indicate that Akt was ubiquitinated and that the level of the ubiquitinated form of Akt was increased in stage 2 neurons and by UPS inhibition. This possibility was further supported by the reciprocal immunoprecipitation experiments in which complexes that were immunoprecipitated by anti-Akt antibody were Western blotted with anti-Ub antibody (). Together, these results suggest that Akt but not P110 and GSK-3β is preferentially degraded by UPS during the formation of neuronal polarity. We then examined the localization of Akt and GSK-3β by immunostaining. Akt was present in both the soma and multiple minor processes of stage 2 neurons. In stage 3 neurons, however, Akt was preferentially found in the soma and axonal tips (). In contrast, GSK-3β was found in all neurites in both stage 2 and 3 neurons (). Statistical analysis of the density of immunofluorescence obtained through scanning the tips of >90 neurons in stage 3 revealed that the density (normalized by the mean density in dendritic tips) of Akt in axonal tips was five times higher than that in dendritic tips, whereas the density of GSK-3β between axonal and dendritic tips was not different (). Moreover, the densities of p-Akt and p–GSK-3β in the axonal tips were four to six times higher than those in dendritic tips (Fig. S2, A–C; available at ). Therefore, the distribution of Akt was changed from symmetric to asymmetric as neurons progressed from stage 2 to 3. This shift from a symmetric to asymmetric distribution of Akt points to the possibility that Akt was selectively degraded in the dendrites. To directly observe whether Akt is degraded in dendrites, we transfected neurons with Akt–photoactive (PA) GFP, which encodes Akt fused to a PAGFP (Fig. S2 D; ). The intensities of fluorescence in neurons transfected with either Akt-PAGFP or PAGFP were monitored for 4 h after photoactivation, which was initiated 36 h after transfection. The relative intensities of fluorescence between axonal and dendritic tips in the neurons expressing PAGFP were similar (P > 0.05; = 10; ). In contrast, in the neurons expressing Akt-PAGFP, the fluorescence intensity in the dendrites was much weaker than that in axons (Fig. S2 E), and the relative intensity of fluorescence in dendritic tips after photoactivation gradually decreased compared with that in axonal termini ( = 10; ). At the end of observation, the relative intensity of the fluorescence for Akt-PAGFP in axonal tips was 21 times higher than that detected in dendritic tips ( = 10; ). Moreover, treatment with MG132 prevented the decrease in the intensity of Akt-PAGFP in dendrites ( = 10; ). Together with the immunostaining results, these findings indicate that preferential degradation of dendritic Akt was mediated by the UPS. We then examined whether UPS inhibition affects Akt/p-Akt localization. As shown in and Fig. S3 A (available at ), treatment with MG132 and lactacystin led to the presence of Akt/p-Akt in all neurites and consequently converted these neurites into axons. The effects of these inhibitors on the redistribution of Akt/p-Akt was further confirmed by transfection experiments. Expressing K48R-Ub resulted in the presence of p-Akt/Akt in most of the neurites, whereas expressing WT-Ub did not affect the distribution of these molecules ( and Fig. S3 B). In contrast, the distribution of GSK-3β was not affected by MG132 or by expressing K48R-Ub to inhibit the UPS (Fig. S3, C and E). However, inhibiting the UPS resulted in the expression of p–GSK-3β in the terminals of most neurites (Fig. S3, D and F). Because p–GSK-3β is a downstream target of p-Akt, these results suggested that the redistribution of p–GSK-3β from an asymmetric to symmetric distribution was caused by a similar change in the distribution of p-Akt after inhibition of the UPS. To examine whether local inhibition of the UPS affects neuronal polarity, we applied MG132 to a neurite in a stage 2 neuron and stained the neuron with antibodies against Tau1 or Akt. As shown in Fig. S3 G, the neurite treated with MG132 for 4 h was enriched in Akt staining and became longer and Tau1 positive, indicating that this neurite was turned into an axon in response to local inhibition of the UPS. Therefore, UPS inhibition prevented neurite terminals from losing p-Akt/Akt and p–GSK-3β, resulting in the formation of multiple axons. It has been known that protein degradation can be regulated by posttranslational modifications (; ). The phosphorylation state of Akt affects its stability in different cell types (; ; ; ; ). To test whether Akt phosphorylation regulates its degradation by the UPS in hippocampal neurons, we examined the ubiquitination state of Akt in stage 2 neurons in response to the treatment with wortmannin, a PI 3-kinase inhibitor also known to suppress Akt phosphorylation. Wortmannin treatment greatly increased the level of ubiquitinated Akt (). The neurons treated extended only several short processes (), which is consistent with a previous study (). Moreover, the protein levels of Akt () and p-Akt (Fig. S4 A, available at ) in all neurite terminals were reduced to undetectable levels. Similarly, p–GSK-3β was absent from all neurites after the application of wortmannin (Fig. S4 A). However, the distribution of GSK-3β was not affected (). Treatment with MG132 reversed the wortmannin-induced Akt loss from neurite terminals, but this treatment reversed neither the loss of p-Akt from neurite terminals ( and Fig. S4 A) nor the effect of wortmannin application on polarity (Fig. S4 B), indicating that Akt activity is required for axon formation. Together, these results suggest that the inhibition of Akt phosphorylation results in its degradation mediated by the UPS. To directly study the effect of inhibiting Akt phosphorylation on its degradation in live cells, we monitored the changes in the fluorescence intensity of Akt-PAGFP in neurons in response to wortmannin. As shown in (A and B), the relative fluorescence intensity found in the axons between the neurons treated with wortmannin and those treated with DMSO was similar (P > 0.05; = 10). In contrast, wortmannin treatment in the first 80 min after photoactivation greatly reduced the relative intensity of fluorescence in dendrites (). After 2 h, the relative intensity of fluorescence in dendritic tips in both control and wortmannin-treated groups was ∼20% of the intensity found immediately after photoactivation. Immunostaining with Akt/p-Akt antibodies further revealed that treatment with wortmannin for 2 h markedly decreased the level of p-Akt in the neuron but did not affect the levels of Akt in axon terminals (Fig. S4 G). Because wortmannin inhibits PI 3-kinase to suppress Akt phosphorylation, our results suggested the possibility that the inhibition of Akt phosphorylation accelerates its degradation in dendrites but does not affect its protein level in axons. Together, our results suggest that the inactive form of Akt was preferentially degraded in dendrites. To test whether the activated form of Akt is resistant to the degradation mediated by the UPS (Fig. S5 A, available at ), we transfected the neurons with WT Akt and myristoyl (myr)-Akt, a constitutively active form of Akt (; ), and examined the ubiquitination of Akt 2 d later. As shown in , WT Akt was highly ubiquitinated, whereas myr-Akt was barely ubiquitinated. These results further suggested that the active form of Akt in neurons is resistant to degradation mediated by the UPS. We then examined whether the expression of myr-Akt affects the polarized distribution of Akt to induce the formation of multiple axons. Neurons transfected with GFP-pcDNA3 or GFP–myr-Akt were stained with an antibody against Akt 4 d after plating (Fig. S5 B). In the neurons expressing GFP, Akt was found mostly in the soma and axon terminals. In contrast, in the neurons expressing myr-Akt, Akt was found in the entire cell (Fig. S5 B). Moreover, the percentages of neurons with no axon, a single axon, and multiple axons were 4.67 ± 2.31, 81.67 ± 2.89, and 13.67 ± 0.58%, respectively, in the neurons transfected with GFP. In contrast, the percentages were 5.33 ± 3.06, 55.33 ± 4.58, and 39.33 ± 4.51%, respectively, in the neurons transfected with WT Akt and 5.67 ± 2.08, 26.33 ± 7.77, and 67.67 ± 10.21%, respectively, in neurons treated with myr-Akt ( = 100; three experiments; ). Additionally, myr-Akt was more potent than WT Akt to induce the formation of multiple axons in the neurons after the initial establishment of polarity ( = 100; three experiments; ). These results demonstrated that the expression of constitutively active Akt eliminated the polarized distribution of Akt and disrupted neuronal polarity. The PI 3-kinase pathway plays instructive roles in cell polarization and migration during chemotaxis (; ) and also regulates axon–dendrite specification in hippocampal neurons (; ). Activation of PI 3-kinase at the terminals of neurites triggers a sequential response, including the stimulation of Akt, GSK-3β, and CRMP-2, leading to the formation of neuronal polarity (, ; ; ). We report that Akt degradation within dendrites by the UPS is required for both the establishment and maintenance of neuronal polarity. Specifically, local instability of Akt/p-Akt in dendrites mediated by the UPS led to axon–dendrite polarity. Several findings support this conclusion. First, Akt was increasingly ubiquitinated in neurons grown from stage 1 to 2, whereas p110 and GSK-3β were not ubiquitinated ( and Fig. S1 J). Second, Akt was present in the axon but not in the dendrites of stage 3 neurons (). Third, UPS inhibition restored the presence of Akt in all neurites and destroyed neuronal polarity (Fig. S3). Lastly, disruption of polarized Akt distribution by the expression of constitutively active Akt prevented the formation of neuronal polarity (). Thus, the preferential localization of Akt to the axon controls the local activity of the Akt–GSK-3β pathway, leading to the formation of neuronal polarity. It has been known that the polarity of epithelial cells (Mv1Lu cells) is regulated by the UPS (; ; ). We have shown that neuronal polarity is also controlled by the UPS. Furthermore, in these epithelial cells, the overexpression of Smurf1, an Ub ligase, induces the specific degradation of RhoA to affect cell polarity. Smurf1 is recruited by PKCζ to cell protrusions, where it controls the local level of RhoA to regulate cell polarity and protrusion formation (). It is plausible that a specific Ub ligase, which controls Akt degradation, may be recruited to dendritic tips, leading to local Akt degradation. In neurons and other cell types, Akt can be degraded by the UPS in response to different stimuli (; ; ; ; ). Furthermore, depending on cell type, both active Akt and Akt may be degraded. An important finding in this study is that blocking Akt phosphorylation by wortmannin promoted its degradation ( and ), and inhibiting the UPS resulted in the maintenance of Akt activity in neurite terminals, which consequently became axons (). Consistent with this, the inhibition of a receptor tyrosine kinase, which activates PI 3-kinase, inhibits the establishment of neuronal polarity (). We also found that in dendrites of the neurons expressing Akt-PAGFP, Akt degradation was accelerated by inhibition of its phosphorylation. However, in axons, the level of Akt was not affected by wortmannin after the establishment of polarity (). Although the explanation of the differential response to wortmannin of Akt in axons and dendrites is not clear, a possible interpretation is that an E3 ligase specific for Akt may be selectively activated in dendrites but not in axons. It is important to note that there was more ubiquitinated Akt in stage 2 than that in stage 1 and 3 neurons (). Stage 2 is a critical period for the initiation of neuron polarity. In this stage, all neurites exchange phases of elongation and retraction, and all neurites have the potential to become axons (; ). During this frequent alternation, the growth and retraction of the Akt level in the neurite terminals may be rapidly regulated by the UPS. After the establishment of neuronal polarity, Akt was enriched in axon terminals but not in dendrites, and disruption of the asymmetrical distribution of Akt by UPS inhibition or by Akt activation throughout the cell led to the formation of multiple axons (, E and F; and Fig. S1, F and G). Therefore, the polarized distribution of Akt as a result of its localized degradation in dendrites as compared with axons was also required to maintain neuronal polarity. It has been shown that protein transport also affects protein localization to regulate neuronal polarity (). It is therefore possible that dendritic instability of Akt arises not from local degradation but through differential protein transport. However, several lines of evidence indicate that dendritic instability of Akt was not caused by the inefficiency of Akt transportation. First, Akt but not P110 and GSK-3β was ubiquitinated ( and Fig. S1 J), and blocking Akt phosphorylation increased its ubiquitination (). Second, UPS inhibition restored Akt/p-Akt presence in all neurites and suppressed the formation of neuronal polarity ( and Fig. S3, A and B). Third, activation of Akt-PAGFP in whole cells did not reveal the differential transportation of Akt between axons and dendrites (). Lastly, the expression of constitutively active Akt prevented the polarized distribution of Akt and inhibited the formation of neuronal polarity (). Some polarity decision molecules such as mPar3/mPar6/aPKC and Rap1B/Cdc42 are present in most neurites in stage 2 and are then redistributed exclusively to axons in stage 3 neurons during the formation of neuron polarity (; , ). The stage-dependent change in the distribution of these molecules similar to that of Akt is important for the establishment of neuronal polarity. However, it is unclear how their localizations are regulated. We found that mPar3 and αPKC were ubiquitinated during the formation of neuronal polarity (Fig. S5 D), suggesting that regulation of their localized redistributions may be through the same mechanism that results in the redistribution of Akt. In the dendrites of mature hippocampal neurons, local degradation of postsynaptic proteins mediated by the UPS plays an important role in synaptic plasticity and spine morphology (; ). An important implication of our findings is that local protein degradation mediated by the UPS is also essential for the establishment of normal morphology in early stages of neuronal development. In summary, Akt degradation in dendrites mediated by UPS was required for neuronal polarity, and this degradation was regulated by its phosphorylation state. The following primary antibodies were used: mouse monoclonal antibody against Ub; goat anti-Tau1 or -P110; rabbit anti-PKCξ (C-20; Santa Cruz Biotechnology, Inc.); rabbit antibody against GFP; synapsin and mouse monoclonal anti-GFP, -Tuj1, -MAP2, or -Tau1 antibodies (Chemicon); rabbit anti–p-Akt (Ser473), p–GSK-3β (Ser9), anti-Akt, or –GSK-3β antibodies (Cell Signaling Technology); mouse anti-myc or -HA antibodies (Sigma-Aldrich); and rabbit anti-Par3 antibody (Upstate Biotechnology). The following secondary antibodies were used: donkey Cy5-, FITC-, and Rhodamine Red-X–conjugated antibodies against mouse, rabbit, or goat IgGs (Jackson ImmunoResearch Laboratories) and Texas red– or FITC-conjugated and AlexaFluor488- or -546–conjugated goat anti–mouse or rabbit IgGs (Invitrogen). HRP-conjugated anti–mouse or rabbit secondary antibodies and all materials for Western blotting were purchased from GE Healthcare. FM4-64 dye was obtained from Invitrogen. MG132, lactacystin, and all other reagents were purchased from Sigma-Aldrich. Rat primary hippocampal neurons were prepared as previously described (). In brief, hippocampi dissected from embryonic day (E) 18 rats were digested with a mixture of proteases at 37°C for 15 min and dissociated with a pipette in MEM containing Earle's salts with 15% FBS, 0.5% glucose, 1 mM sodium pyruvate, and 25 μM glutamine. Neurons were plated onto glass coverslips coated with poly--lysine at a density of 100–200 neurons/mm. Neuronal cultures were incubated at 37°C with 5% CO. After 1 h, the medium was changed to neurobasal medium (with B27 supplement and 0.5 mM glutamine). Before the establishment of polarity, neurons were transfected with different constructs using nucleofector (Rat Neuron Nucleofector Kit; Amaxa Biosystems). After the establishment of polarity (48 h after plating), neurons were transfected using the calcium phosphate. In brief, the constructs were mixed with 250 mM CaCl and an equal volume of 2× Hepes-buffered saline (274 mM NaCl,10 mM KCl, 1.4 mM NaHPO, 15 mM -glucose, and 42 mM Hepes, pH 7.06). The DNA–calcium complex was incubated for 20 min and added to the neurons in DME without glutamine. After transfection, neurons were washed three times with DME, incubated for 1 h, and transferred to the original medium for 3 d. His6-myc–WT-Ub and His6-myc–K48R-Ub constructs were gifts from E. Burstein (University of Michigan Medical School, Ann Arbor, MI). HA WT-Akt, K179M Akt, and myr-Akt constructs were gifts from A. Bellacosa (Fox Chase Cancer Center, Philadelphia, PA). PAGFP-N1 construct was a gift from J. Lippincott-Schwartz and G.H. Patterson (National Institute of Child Health and Human Development, Bethesda, MD). Neurons cultured on coverslips were washed three times with PBS and fixed with 4% PFA in PBS containing 0.4% sucrose at 4°C for 30 min. The fixed neurons were washed, incubated with 0.5% Triton X-100 in PBS for 5 min, and blocked with 10% FBS in PBS for 1 h at room temperature. Neurons were probed with the primary antibodies at 4°C overnight and washed three to six times with 0.05% Tween-20 in PBS. They were then incubated with the secondary antibodies at room temperature for 1 h and washed three to six times with 0.05% Tween-20 in PBS. All antibodies were diluted with PBS containing 10% FBS. Axons are defined as Tau1-positive/MAP2-negative neurites with a mean length of >120 μm 4 d after plating or neurites with a length twice that of other neurites at 2 d in culture (). Dendrites are defined as MAP2-positive/Tau1-negative neurites. Images were acquired by fluorescent microscopes (LSM510 Axiovert 200M [Carl Zeiss MicroImaging, Inc.]; or E600 FN Neurolucida system [Nikon]). Neuronal morphology was analyzed using the Physiology software of LSM510 (Carl Zeiss MicroImaging, Inc.). In the PAGFP experiments, neurons were cotransfected with PAGFP (or Akt-PAGFP) and RFP before plating. Photobleaching was performed using a two-photon microscope (LSM510 META NLO Axioskop 2 FS MOT; Carl Zeiss MicroImaging, Inc.) at 37°C with 5% CO (). Images were acquired every 10 min, and each cell was observed for 4 h. Data analysis was performed using MetaMorph software (Molecular Devices). Protein levels of Akt in the axonal and dendritic tips were estimated by normalizing the intensity of PAGFP/RFP fluorescence, which was calculated according to the equation (Fn − F0)/(F1 − F0), in which F0 is the PAGFP/RFP fluorescence 1 min before photoactivation, F1 is the PAGFP/RFP fluorescence right after photoactivation, and Fn is the PAGFP/RFP fluorescence of ( − 1) × 10 min. Total proteins were extracted using radioimmunoprecipitation buffer (25 mM Tris-HCl, pH 7.4, 150 mM KCl, 5 mM EDTA, 1% NP-40, 0.5% sodium deoxycholate, and 0.1% SDS), and protein concentrations were measured using a protein assay kit (Bio-Rad Laboratories). Immunoprecipitation was conducted by incubation of the cell extracts with primary antibodies (1:50) overnight at 4°C. Protein G or A Sepharose was then added and incubated for 4 h. The immunocomplexes were collected and washed with radioimmunoprecipitation buffer. For Western blots, proteins were denatured by boiling in sample buffer for 5 min, separated on 6–10% SDS PAGE, and transferred to polyvinylidene difluoride membrane. After blocked with 5% fat-free milk in PBS, the polyvinylidene difluoride membrane was probed with the indicated primary antibodies and HRP-conjugated secondary antibodies. The bands were visualized with an ECL system (GE Healthcare). The densities of the bands were determined by ImageQuant software (GE Healthcare). Neurons were incubated with 10 μM FM4-64 and 45 mM KCl for 1 min and washed with normal medium for 15 min. FM4-64 fluorescence was observed using an inverted microscope (LSM510 Axiovert 200M; Carl Zeiss MicroImaging, Inc.). Neurons were imaged again after destaining in 90 mM KCl for 5 min. Neurons on coverslips were perfused locally using a micropipette (tip opening of <1 μm) pointed to a specific region of the neuron (). Perfusion medium contained culture medium with 200 nM DMSO/wortmannin or 0.15 μm MG132. The micropipette was positioned near one neurite, and local perfusion was then performed for a period of 2–4 h using the PM8000-B eight-channel pressure injector system (positive pressure of 2 psi was applied at 2 Hz; World Precision Instruments). The images were acquired by microscopes (TE2000E; Nikon) at one image/2 min. The data were analyzed by MetaMorph software. Statistical analysis was conducted using the test. Group differences resulting in P values of <0.05 were considered statistically significant. Fig. S1 shows that UPS inhibition affects both the establishment and maintenance of neuron polarity. Fig. S2 shows the asymmetric distribution of p-Akt and p–GSK-3β in stage 3 neurons. Fig. S3 shows that UPS inhibition disrupts the asymmetric distribution of Akt and p–GSK-3β. Fig. S4 shows the effects of wortmannin on neurite growth, Akt degradation, and distribution. Fig. S5 shows that expressing myr-Akt disrupts the asymmetric distribution of Akt, and mPar3 and aPKC are ubiquitinated. Online supplemental material is available at .

Phosphatidylinositol (PI) 3-kinase signaling plays an important role in neuronal development (; ; ; ). It is essential for the NGF-induced neurite elongation of PC12 cells, as this process is inhibited by wortmannin, a relatively specific inhibitor of PI 3-kinase (). Also, overexpression or microinjection of active PI 3-kinase in PC12 cells is sufficient to induce neurite formation (; ). Recent technological developments made it possible to visualize the intracellular localization of PIP by the use of pleckstrin homology (PH) domain of PI--trisphosphate (PIP) interacting proteins such as Akt () and GRP1 () as PIP-specific probes. These studies revealed that PIP is highly enriched at the tip of growing neurites in PC12 cells (; ). Similarly, in cultured hippocampal neurons, PIP accumulates at the distal end of the longest neurite and induces the single longest neurite to develop into an axon (; ). The downstream signaling mechanism coordinating the impact of concentrated PIP at the tip of the neurite to form an axon has been extensively studied (; ; ; ; ; ). However, how PIP accumulates at the tip of neurite remains an intriguing, yet unresolved, question. One possible explanation is that the activity of PI 3-kinase and its upstream signaling mediators are concentrated at the tip of neurites, thereby producing PIP locally (). A possible positive-feedback loop in which PIP activates upstream signaling of PI 3-kinase has been proposed (). Alternatively, PIP might be transported from the cell body through neurites by the motor-dependent trafficking mechanism. Although historically PI 3-kinase activity and PIP production have been considered to occur at the plasma membrane (), significant evidence has emerged supporting the idea that PIP is also produced within intracellular membranes, such as the endocytosed vesicles containing activated receptors (). Therefore, an attractive hypothesis predicts that such PIP-containing vesicles, laden with its upstream and downstream signaling components, are transported to specialized destinations where the local signaling activity is most needed. Kinesin-family proteins are microtubule-based motor proteins implicated in the transport of diverse cargos (). Some of these kinesins recognize phospholipids as their cargo molecules. A kinesin-3 family protein (), Unc104/KIF1A has a PH domain located at the C-terminal tail, which directly interacts with PI--bisphosphate (PIP) and transports PIP-containing vesicles (; ). Another kinesin-3 protein, KIF16B, transports PI--phosphate (PI3P) through direct binding, via its C-terminal PhoX homology domain to PI3P (). We propose that guanylate kinase–associated kinesin (GAKIN; ; ), classified as a kinesin-3 protein, KIF13B (), transports PIP-containing vesicles through the interaction with an adaptor protein PIPBP/centaurin-α (; ). PIPBP has two PH domains, which specifically interact with PIP and has been implicated in the regulation of PIP signaling function (, ; , ). The PIPBP was originally identified in other species and named as centaurin-α (Hammonds-Odie et al.,1996). Centaurin-α/PIPBP was recently shown to exhibit Arf GAP (ADP-ribosylating factor GTPase activating protein) activity in vivo (). Our results provide the first evidence that PIP is transported by the motor-dependent mechanism. Using full-length bovine PIPBP as bait, we performed a yeast two-hybrid screen in HeLa cell cDNA library and isolated a partial cDNA of human GAKIN, a kinesin-like protein. Recently, in a collaborative study, human centaurin-α was used as a bait to isolate rat homologue of GAKIN/KIF13B from rat brain cDNA library (). We confirmed the interaction between PIPBP and GAKIN and mapped the binding site using the GST pull-down assay. Defined segments of GAKIN, covering the region originally isolated by the yeast two-hybrid screen, were expressed in 293T human kidney cells, and their biochemical interactions with GST-PIPBP immobilized on glutathione–Sepharose beads were tested. The construct spanning the motor-FHA domains (1–557) interacted specifically with GST-PIPBP but not with the GST control (). Deletion of the FHA region within this segment completely abolished the interaction (motor 1–486), and a short construct containing only the FHA domain (FHA 455–557) showed specific binding to PIPBP. To further test whether the observed biochemical interaction is direct, we incubated GST-PIPBP with thioredoxin (Trx) fusion protein of the GAKIN FHA domain immobilized on agarose beads. GST-PIPBP specifically interacted with Trx-FHA but not with the control Trx (). Together, these results demonstrate that the FHA domain of GAKIN mediates its direct interaction with PIPBP. To demonstrate the in vivo association between PIPBP and GAKIN, we performed coimmunoprecipitation experiments from COS-7 cells transiently expressing PIPBP and full-length GAKIN. The COS-7 cells were transfected with myc-PIPBP and FLAG-GAKIN, and an anti-myc antibody was used to immunoprecipitate the complex. Western blotting with anti-FLAG antibody detected GAKIN in the myc immunoprecipitate (, second lane). We also tested KIF13A, the closest homologue of GAKIN (), for its potential interaction with PIPBP. The FHA domains of KIF13A and GAKIN share 70% amino acid identity. Interestingly, however, we failed to detect any interaction of KIF13A with PIPBP in the coimmunoprecipitation experiments (, third lane). This result suggests that the interaction between PIPBP and GAKIN is highly specific. It is noteworthy that the FHA domain is a widely conserved module in the kinesin-3 family members (); however, its function is poorly understood in the context of motor proteins. As PIPBP specifically interacts with PIP via its PH domains (), we hypothesized that PIPBP mediates the physical linkage between GAKIN and PIP-containing vesicles, thereby facilitating the microtubule-dependent transport of the cargo vesicles containing PIP. To test this hypothesis, we first characterized the motor activity of GAKIN by the microtubule gliding assay. Recombinant GAKIN containing motor and FHA domains (1–557) was expressed in Sf9 cells using the baculovirus expression system, and purified protein was coated on the glass surface of the motility chamber. Microtubules were polarity labeled with Alexa 488 to highlight their minus ends, and their gliding movement was visualized using time-lapse fluorescence microscopy (). The smooth movement of microtubules toward their brightly labeled minus end was reproducibly observed in multiple trials, demonstrating that GAKIN is a plus end–directed microtubule motor. A histogram showing the velocity of labeled microtubules was generated from multiple tracings and is shown in . The mean velocity of moving microtubules was estimated to be 1.66 μm/s. Next, we tested whether PIPBP and GAKIN can form a macromolecular complex on the surface of PIP-containing liposomes. Purified recombinant GST-PIPBP and GAKIN (motor-FHA segment 1–557) were incubated with synthetic liposomes containing 10% PIP and 90% phosphatidylcholine (PC), and the liposome and soluble protein components were separated in the top and bottom fractions, respectively, by sucrose density gradient centrifugation. GAKIN was recovered from the top (liposome fraction) of the gradient when PIPBP and PIP liposomes were present (, top, lanes 5 and 6). In contrast, GAKIN was recovered only from the bottom (soluble protein fraction) of the gradient when control GST was used instead of GST-PIPBP (, top, lanes 3 and 4) or when 15% PIP liposomes were used instead of PIP (, top, lanes 7 and 8). The GST-PIPBP was consistently recovered from the top fraction with PIP liposomes (, bottom, lanes 5 and 6) and only in the bottom fraction when 10% PIP liposomes were used (Fig. S1, available at ). When 15% PIP liposomes were used as a negative control, only a trace amount (<0.5%) of GST-PIPBP was recovered in the top fraction, and most of the protein remained in the bottom fraction (, bottom, lanes 7 and 8). The lipid binding specificity of PIPBP was documented previously (Hammonds-Odie et al., 1996; ; ). Based on our Western blot results, the affinity of PIPBP to 10% PIP liposomes appears to be >100 times better than 15% PIP liposomes. As a negative control, GST was recovered only from the bottom fraction (, bottom, lanes 3 and 4). Together, our data are consistent with the notion that PIPBP is a specific binder of PIP and demonstrate that PIPBP can bind both GAKIN and PIP simultaneously, thereby serving as a molecular linker connecting GAKIN to PIP-containing lipid vesicles. We then examined whether the complex of GAKIN and PIPBP formed on PIP liposomes could support actual movement of the liposomes along microtubules. To demonstrate this possibility, we reconstituted the microtubule-dependent movement of liposomes using purified recombinant PIPBP and GAKIN. A dense microtubule lawn was prepared on the glass surface of the motility chamber using Alexa 488–labeled microtubules. To visualize liposome movement by fluorescence microscopy, liposomes were labeled using rhodamine-phosphatidylethanolamine (Rh-PE). Processive movement of PIP liposomes along microtubules was observed in the presence of GAKIN, PIPBP, and ATP ( and Videos 1 and 2, available at ). Mean velocity of liposome movement was ∼0.7 μm/s (). No processive movement was observed in the absence of GAKIN, PIPBP, or ATP or when PIP was used instead of PIP, demonstrating the specificity of the liposome movement (). This result suggests that GAKIN is capable of transporting PIP-containing vesicles via direct interaction with PIPBP. Next, we asked whether this PIP transport by GAKIN–PIPBP complex plays a functional role in vivo. As endogenous GAKIN and PIPBP can be coimmunoprecipitated from rat brain lysate (), we surmised that the neuronal cell might be the site where such transport is important. Therefore, we selected PC12 cells, the widely used cell line with neuronal properties, for further studies. In PC12 cells, PIP is highly enriched at the tip of growing neurites (; ) and its signaling is important for neurite outgrowth (; ). First, we examined whether PIPBP and GAKIN colocalize at specific sites in PC12 cells. When PC12 cells were cotransfected with myc-PIPBP and GFP-GAKIN and treated with NGF, both proteins significantly colocalized and accumulated at the tip of neurites (). Higher magnification images show that both proteins are found at the growth cone protrusions (). To visualize the localization of PIP, the PH domain construct of Akt (GFP-Akt-PH), which specifically interacts with PIP in the cells (), was used. As reported previously (), accumulation of GFP-Akt-PH was observed at the tip of growing neurites (). Endogenous GAKIN was also found concentrated at the tip of the neurites, where it colocalized with GFP-Akt-PH (). To test whether the observed colocalization of GAKIN, PIPBP, and PIP is the consequence of the transport of PIP by the GAKIN–PIPBP complex, we decided to use the dominant-negative (DN) approach by blocking specific interactions. The DN construct of GAKIN is a motor-deleted mutant of GAKIN, which can still bind to PIPBP via its FHA domain (unpublished data). Unlike full-length GAKIN, which accumulates only at the tip of neurites ( and ), the transfected DN-GAKIN in PC12 cells distributes along neurites and the cell body. The accumulation of DN-GAKIN is still evident at the tip (); however, it is not as intense and, consequently, the cell body and the entire length of neurites are visible. This distribution suggests the inability of DN-GAKIN to translocate its cargo to its final destination. Accumulation of PIP, as monitored by the GFP-Akt-PH detector, at the tip of neurites is significantly inhibited by the overexpression of DN-GAKIN (). The relative intensity of GFP-Akt-PH and anti-FLAG staining was measured and is shown in the graphical format ( and Fig. S2, available at ). Because the FHA domain of GAKIN is required for its interaction with PIPBP, two additional constructs lacking the FHA domain of GAKIN were made and used as negative controls (). In the Δ-FHA construct of GAKIN, the FHA domain was deleted from the DN-GAKIN mutant. Thus, the Δ-FHA construct starts with the membrane-associated guanylate kinase (MAGUK) binding stalk domain and extends up to the Cap-Gly domain. A chimera construct (FHA-chimera) of GAKIN was also made where its FHA domain was replaced with the FHA domain of KIF13A, the closest homologue of GAKIN, which does not bind to PIPBP. Thus the FHA-chimera contains the FHA domain of KIF13A and the MAGUK binding stalk to Cap-Gly domains of GAKIN. Unlike DN-GAKIN, neither of these constructs showed any effect on PIP localization (). These results suggest that PIP is transported to the distal end of neurites by a GAKIN-dependent mechanism. In cultured hippocampal neurons, the location of PIP was monitored by the GFP-Akt-PH detector and found highly enriched in the distal end of the longest neurite () in contrast to the diffused fluorescence observed for GFP ( and Fig. S2). This location of PIP is consistent with previously published studies (; ). Endogenous PIPBP was found colocalized with GFP-Akt-PH at the end of the longest neurite (). As was the case in the PC12 cells, overexpressed full-length GAKIN was found highly enriched at the distal ends of neurites, often at their multiple termini (, middle). Significant colocalization of GFP-Akt-PH with overexpressed GAKIN was observed, and the accumulation of GFP-Akt-PH appeared to be enhanced under these conditions (, middle). On the other hand, overexpression of DN-GAKIN caused the loss of enriched accumulation of GFP-Akt-PH at the tip of the longest neurite (, right). Similar to the pattern observed in the case of GFP-Akt-PH expression alone in control neurons, some staining of GFP-Akt-PH could be visualized at the end of short neurites in DN-GAKIN transfected cells. As before, the relative intensity was measured and is shown in (Fig. S2). We quantified the GFP-Akt-PH accumulation by scoring the cells with the relative intensity ratio between neurite tip and cell body (), and the representative images are presented (Fig. S4, available at ). The data show that upon overexpression of DN-GAKIN, a subpopulation of cells that show high accumulation at the tip (tip/cell body ratio >1) disappeared. In contrast, cells with a high accumulation ratio are increased when full-length GAKIN is overexpressed. These results suggest that the accumulation of PIP in hippocampal neurons is dependent on the activity of GAKIN. Also, we noticed that the cells overexpressing GAKIN and DN-GAKIN often lost the typical morphology of the well-differentiated axon-dendrite structure, leading to the formation of multiple, highly branched neurites (). In the case of full-length GAKIN, this could be attributed to the hyperaccumulation of PIP at the ends of multiple neurites. Generally, PIP without the overexpression of GAKIN would accumulate in the longest neurite observed in normally developing neurons. In the case of DN-GAKIN, the reduction of PIP accumulation in the longest neurite end might inhibit proper axon-dendrite specification. Therefore, we quantified the number of cells with axon-dendrite polarity based on the morphological observation (; ) upon overexpression of GAKIN and DN-GAKIN. It is noteworthy that a recent study examined the role of PIP in neuronal polarization and axon formation and established that the hippocampal neurons expressing low levels of the GFP-Akt-PH construct retained cell polarity (). Following this experimental strategy, we used a highly purified preparation of GFP-Akt-PH plasmid and selected neurons expressing low levels of GFP-Akt-PH. Similar to the GFP transfected control neurons, the GFP-Akt-PH transfected cells maintained cell morphology as established by the observation of Tau-1–positive neurites in individual neurons (Fig. S3, available at ). However, to further eliminate the possibility of deleterious effects of the GFP-Akt-PH detector on the neuronal polarity (), GAKIN and DN-GAKIN were cotransfected with GFP as a morphological marker. Representative images () and the quantification data () demonstrate that overexpression of GAKIN and DN-GAKIN caused the loss of neuronal polarity. To confirm that such an effect is mediated by the FHA–PIPBP interaction, the effect of the motor-FHA (1–557) domain of GAKIN was tested. When expressed in primary neurons, the motor-FHA domain showed diffuse distribution throughout the cell and did not accumulate at the neurite ends (Fig. S5), suggesting that it is not capable of transporting cargo in vivo. The motor-FHA domain suppressed neuronal polarity, whereas the motor domain alone (1–486, without the FHA domain) did not ( and Fig. S5). These results suggest the necessity of the FHA domain for the DN effect of GAKIN mutants on neuronal polarity. As a negative control, KIF13A had no effect on polarity. To further confirm this effect on neuronal polarity, transfected neurons were stained for tau, a marker for axon differentiation (; ). GFP-expressing control neuron exhibited a single long Tau-1–positive neurite, which is an axon (, left). However, neurons expressing full-length GAKIN or DN-GAKIN did not have Tau-1–positive neurites (). These results confirmed the loss of polarity observed by the morphological analysis. In this study, we demonstrate that a complex of PIPBP and GAKIN provides a novel mechanism for motor-mediated transport of PIP-containing vesicles along microtubules (). Previously, a model for the transport of PIP-containing vesicles by Unc104/KIF1A, an evolutionally conserved kinesin-like motor protein, was proposed (). In that model, a PH domain located at the C-terminal tail of Unc104 interacts with PIP specifically, thereby connecting the PIP vesicles to microtubule-dependent transport (, left). Our results propose a new model in which PIPBP serves as a molecular linker connecting PIP vesicles to a motor protein (, right). The PIPBP binding site in GAKIN was localized to a small region encoded by the FHA domain (). The FHA domain was originally discovered in transcription factors and DNA repair proteins and is now well established as the phosphothreonine binding protein domain (). All members of the kinesin-3 subfamily of proteins, including GAKIN, contain an FHA domain in close proximity to the N-terminal motor domain (). However, it is not known whether the FHA domain of kinesins recognizes phosphothreonine residues, as do other conventional FHA domains. Our direct binding experiments used bacterially expressed GST-PIPBP and GAKIN FHA domain, suggesting that phosphorylation was not required for the ligand recognition of FHA, although we could not rule out the possibility that the recombinant GST-PIPBP was phosphorylated in the bacterial host. Nonetheless, our results provide the first evidence indicating that the FHA domain of kinesin-3 family protein is a cargo binding domain. Using in vitro motility assays, previous studies have determined the velocity range of kinesins to be ∼0.1–1.5 μm/s (for review see ). We measured the velocity of the recombinant motor-FHA domain of GAKIN using a microtubule gliding assay in the presence of 1.0 mM ATP and ATP regeneration system, and it was determined to be ∼1.66 μm/s. Using a liposome motility assay, the velocity was determined to be ∼0.7 μm/s, which is slightly less than the velocity measured by the microtubule gliding assay. It may be speculated that in the microtubule gliding assay, multiple motor molecules contribute to the movement of each microtubule, whereas in the vesicle motility system, fewer motor molecules or possibly a single motor could be responsible for the movement of a single liposome. Overall, these values are consistent with the microtubule-dependent ATPase activity of the GAKIN motor domain that we reported previously (). These results indicate that the motor and FHA segment of GAKIN shows relatively fast anterograde activity; however, the motor activity of the full-length GAKIN remains to be established. GAKIN was originally discovered as the binding partner for human discs large tumor suppressor protein (hDlg), a member of the MAGUK superfamily (). As a scaffolding protein with multiple protein–protein interaction motifs, hDlg is proposed to link specific cargo vesicles to GAKIN for their transport to specialized membrane regions (). Its homologue, Dlg, is an important polarity determination factor in epithelial cells and asymmetric cell division of neuroblasts (). It is, therefore, possible that the transport of hDlg by GAKIN also has significant effects on cell polarity in general. However, in the current setting, we believe that the effect on the neuronal polarity is primarily regulated by GAKIN–PIPBP–mediated transport of PIP, mainly because significant enhancement of PIP accumulation that was observed upon overexpression of full-length GAKIN correlated with the loss of polarity and the motor-FHA construct, which binds to PIPBP but not to hDlg, exhibited DN effects on neuronal polarity. The precise relationship of two potential cargos of GAKIN, mediated by PIPBP and hDlg, will be a subject of future studies. Generally, PIP accumulates only at the tip of the longest neurite of a neuronal cell and, thus, the PIP-positive neurite receives the exclusive differentiation cue to develop into an axon. Our images of GFP-Akt-PH localization in the control neurons (, A and B, left) are consistent with this model, with the signal accumulating only in the single terminus of the longest neurite. When the full-length GAKIN is overexpressed in neurons, it tends to accumulate not only to the longest neurite but also at multiple termini of every growing neurite (). Consequently, the GFP-Akt-PH accumulation is detected in multiple termini of several neurites where GAKIN accumulates (). Our interpretation is that the accumulation of PIP in the longest neurite as well as at the ends of multiple neurites disrupts or delays the axon-specification signal, thus causing the loss of tau-positive (axon-bearing) neurons under these experimental conditions (). In differentiated neurons, how kinesin motors recognize axon from dendrites and transport axon-specific cargo to axons and dendrite-specific cargo to dendrites remains unresolved. However, in cultured hippocampal neurons that are yet to be fully differentiated, motors may not be able to distinguish one neurite from another, which appears to be the case for GAKIN. PIP, by accumulating in specialized membrane sites in the cell, is an important cellular polarity determination factor (). As is the case in hippocampal neurons, where the localized accumulation of PIP determines the axon-dendrite polarity, localization of PIP at the leading edge determines the internal polarity of neutrophil and during their chemotaxis (; ; ; ). How is the local gradient of PIP formed? One possible mechanism to initiate and maintain localized PIP is the positive-feedback loop formed between PIP and small GTPases, Rac and Cdc42 (; ). In neutrophils, exogenously added PIP activates endogenous PI 3-kinase, suggesting the existence and importance of such a positive-feedback loop (). It was suggested that such positive-feedback pathways operate in the growth cone of neuronal cells (). Therefore, it has been assumed that local production of PIP at the tip of neurites and the enhancement of the signal by positive feedback, which activates further production of PIP, are sufficient to explain the sustained local accumulation of PIP in neuronal cells (; ). How, then, does our proposed model of PIP transport fit into the current paradigm? Our results demonstrate that full-length GAKIN overexpressed in hippocampal neurons accumulated at the distal ends of neurites and PIP accumulation is enhanced at sites where GAKIN is accumulated. In contrast, overexpression of a related motor protein, KIF13A, did not significantly affect PIP accumulation. These results suggest that GAKIN specifically transports PIP itself or transports the factors that stimulate the production of PIP, to contribute to the local accumulation of PIP. From our present data, it is not possible to differentiate between these two possibilities at this stage. In any event, these two possibilities may not be mutually exclusive, as the transported PIP can stimulate the local production of PIP by the aforementioned positive-feedback mechanism. The transport of PIP-containing vesicles could move upstream and downstream components of the PIP signaling complex residing on the same vesicles as PIP. Overall, our data suggests that GAKIN-dependent transport contributes to the local accumulation of PIP at the tip of neurites, thus supporting the existence of a novel mechanism for the proposed transport of PIP in vivo. The conventional model of PIP as a second messenger describes that it is produced on the plasma membrane by a signaling complex formed by the activated receptors and is metabolized quickly by the lipid phosphatase activity such as PTEN (phosphatase and tensin homologue deleted on chromosome 10), thus eliciting only the transient response on site (). Can PIP be found on the intracellular vesicles that are transportable? Significant evidence has been reported that demonstrates the production of PIP at the intracellular membranes (). Also, in sympathetic neurons, locally administered NGF at the tip of axons leads to activation of PI 3-kinase at the cell body () by the retrograde transport of the internalized activated NGF receptor (; ; ). Therefore, it is reasonable to assume that in developing neurons a significant amount of PIP is continuously produced at intracellular membrane vesicles in the cell body. In neurons, which often extend exceptionally long processes, transport of specific cargo molecules by motor proteins is essential for the proper targeting of cellular components (). For example, signaling components downstream of PIP that regulate axonal specification, such as CRMP-2 (collapsin response mediated protein 2) and Par3/Par6/aPKC, are transported by kinesin-1 and -2, respectively (; ). We propose that the GAKIN–PIPBP complex transports PIP-containing vesicles from the cell body to the distal end of the neurites. Directed transport of PIP and its signaling components to the specialized membrane region contributes to the local enhancement of the signaling strength at the neurite ends, thus leading to a precipitous breakdown of the cellular symmetry. HEK293T and COS-7 cells were maintained in DME (Invitrogen) supplemented with 10% FCS (Hyclone). PC12 cells were cultured in DME supplemented with 10% FCS and 5% horse serum (Invitrogen). DNA transfection in COS-7, PC12 cells, and hippocampal neurons were performed using Lipofectamine 2000 (Invitrogen). HEK293T cells were transfected using the calcium phosphate method (). GFP- and myc-PIPBP (), GFP-GAKIN, and GFP-DN- GAKIN () were described previously. Other plasmids used to express tagged proteins were GFP and pEGFP-C1 (CLONTECH Laboratories, Inc.), Trx and pET32a (Novagen), and FLAG and pCMV-Tag2b (Stratagene). Corresponding cDNA fragments were cloned into vectors by conventional molecular biology methods. GFP-Akt-PH was a gift from T. Balla (National Institutes of Health, Bethesda, MD; ). Anti-GAKIN rabbit polyclonal antibody () and anti-PIPBP/centaurin-α rabbit anti-serum () were described previously. Anti-GAKIN mouse monoclonal antibody (4A05) was generated against recombinant protein of GAKIN (1414–1826). Commercial antibodies used are anti-myc (9E10; Sigma-Aldrich), anti-FLAG (M2; Sigma-Aldrich), anti-GFP (Invitrogen), HRP-conjugated anti-GST (Santa Cruz Biotechnology, Inc.), and Tau-1 (Chemicon). GST-PIPBP and GST were expressed and purified as described previously (). Trx-FHA protein of GAKIN, which also carries His- and S-protein tags, was expressed in BL21(DE3) cells and purified by Ni-NTA agarose (Novagen) chromatography. GAKIN His-motor-FHA (aa 1–557) was expressed in Sf9 cells using the Bac-to-Bac baculovirus expression system (Invitrogen). The protein was purified by Ni-NTA agarose. For immunoprecipitations, COS-7 cells were transfected with FLAG-GAKIN or -KIF13A along with myc-PIPBP. Cells were lysed in the IP buffer (10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1.0 mM MgCl, and 0.5% Triton X-100), and immunoprecipitations were performed using anti-myc monoclonal antibody (9E10). FLAG- and myc-tagged proteins were detected by Western blotting. For GST pull-down experiments, 293T cells were transfected with various GFP-GAKIN constructs and lysed in the IP buffer. GST-PIPBP or control GST proteins immobilized on glutathione–Sepharose 4B beads were used to pull down interacting proteins from the lysate. GFP-GAKIN proteins bound to the beads were detected by Western blot using anti-GFP antibody. Tubulin was prepared from cow brain () and fluorescently labeled using Alexa 488 carboxylic acid succinimidyl ester (Invitrogen) as described previously (). Microtubules were polymerized in the presence of 20 μl of unlabeled tubulin (2.5 mg/ml) and 5 μl of Alexa 488–labeled tubulin (0.5 mg/ml) using taxol (Calbiochem). Minus-end polarity-marked microtubules were prepared by a two-step polymerization protocol of labeled and unlabeled tubulins mixed at a defined ratio (; Nielsen et al., 2001). A flow chamber with a volume of ∼10 μl was prepared using two glass coverslips (Corning) separated by two parallel strips of double-sided tape. 0.5 mg/ml of recombinant GAKIN motor-FHA (1–557) protein in BRB80 (80 mM Pipes, pH 6.8, 2.0 mM MgCl, and 1.0 mM EGTA) supplemented with 0.2 mM Mg-ATP, 2.0 mM DTT, and 20% glycerol was incubated inside the flow cell for 10 min. After washing the flow cell with 10 μl of motility buffer (BRB80, 1.0 mM ATP, 20 μM taxol, 1.0 mg/ml casein, 50 mM glucose, 0.5 mg/ml glucose oxidase, 4.0 mM phosphoenolpyruvate, 20 μg/ml pyruvate kinase, and 0.14 mg/ml catalase), 10 μl of gliding buffer containing polarity-marked microtubules was introduced into the flow cell. The flow cell was sealed with petroleum jelly and visualized by fluorescence microscopy (TE2000E [Nikon]; Plan Apo 100× objective lens) at room temperature. Images were recorded by a camera (CoolSnapHQ; Roper Scientific) and MetaMorph software (Universal Imaging Corp.). Sequential photographs were taken every 1 or 5 s. Egg yolk PC, Rh-PE, synthesized unsaturated PIP, and synthesized unsaturated PIP (all from Avanti Polar Lipids, Inc.) were stored in organic solvents. 1.0 μmol PC was mixed with 1.0 or 10% (g/g lipids) PIP or 10 or 15% PIP and 0.05% Rh-PE and dried in a glass vial under the stream of dry nitrogen. The lipid film was further desiccated under high vacuum overnight, and lipids were rehydrated by the addition of BRB80 containing 5% sucrose and incubated at 70°C for 30 min with occasional mixing by vortex. Three freeze/thaw cycles were performed in the dry ice box and a 37°C water bath. A mini-extruder device with 250-μl syringes (Avanti Polar Lipids, Inc.) was used to extrude liposomes through a 100-nm pore polycarbonate filter (Whatman) at 70°C. Liposomes were stored in the dark at 4°C and used within 2 wk. Liposomes containing 10 or 15% PIP or 10% PIP (5.0 nmol) were incubated on ice for 30 min with 70 pmol GST-PIPBP and 30 pmol His-motor-FHA. BRB80 containing 2.0 M sucrose was added to the incubation reaction to bring the final sucrose concentration to 1.6 M, and this mixture was overlaid with cushions containing 1.4, 0.4, and 0.25 M sucrose in the same buffer in a centrifuge tube (7 × 20 mm). After centrifugation at 40,200 rpm at 4°C for 30 min in a rotor (Type 42.2Ti; Beckman Coulter), the 0.25/0.4 M fraction (top fraction) and the loading fraction (1.6 M sucrose; bottom fraction) was collected and analyzed by SDS-PAGE followed by Western blotting with anti-GAKIN polyclonal antibody and anti-GST antibody. For anti-GST Western blot, the loading ratio of top/bottom fraction was 5:1. For anti-GAKIN Western blot, the ratio was 25:1. GST-PIPBP loaded in the input lane was 15 ng. The His-motor-FHA amount was 20 ng. A lawn of Alexa 488–labeled microtubules was prepared on the motility chamber to visualize the movement of rhodamine-labeled liposomes by fluorescence microscopy. To ensure the attachment of microtubules on the glass surface, the flow cell was first coated with recombinant His-GAKIN-CT (aa 1414–1826) that contains a microtubule binding CAP-Gly domain and binds avidly to microtubules in vitro. A similar approach was used before to immobilize microtubules on the glass surface coated with either anti-tubulin of poly--lysine (). After incubation of 0.125 mg/ml His-GAKIN-CT (in BRB80 containing 0.1 mg/ml BSA) in the flow cell for 10 min at room temperature, fluorescent-labeled microtubules (0.1–0.2 mg/ml in BRB80 supplemented with 10 μM taxol and 1.0 mM GTP) were introduced and incubated for 30 min at room temperature in a humidified chamber. The flow cell was washed twice with BRB80 supplemented with 5.0 mg/ml casein, 10 μM taxol, and 1.0 mM GTP and used immediately for the motility assay. Liposomes of various compositions were incubated with 0.15 nmol GST-PIPBP on ice for 30 min, and stable liposome–PIPBP complex was separated from unbound PIPBP by ultracentrifugation using sucrose gradient as described (see Liposome flotation assay). The top fraction of the gradient containing the liposome–PIPBP complex was mixed with 0.5 mg/ml of His-motor-FHA in the motility buffer and introduced into the flow cell containing a lawn of microtubules. The final concentration of liposomes was ∼50 μg/ml. Time-lapse imaging was captured as described for the microtubule gliding assay. Sequential photographs were taken every 1 s for 30 s. Imaging of the fluorescently labeled microtubules was captured for each frame and overlaid with the imaging of liposomes. The velocity of each moving liposome was measured using MetaMorph software by tracing the position of each liposome for every second. Videos were made using the MetaMorph software. Individual experiments were repeated more than three times for each sample (nine times for 1% PIP-containing liposomes and six times for 10% PIP-containing liposomes). PC12 cells were seeded onto coverslips (Fisher Scientific) coated with 0.01% poly--lysine (Sigma-Aldrich), stimulated with 5.0 ng/ml NGF for 36 h, and fixed with 4% PFA in PBS at room temperature for 15 min. Cells were then washed with PBS and permeabilized by PBS supplemented with 0.1% Triton X-100. Hippocampal neurons were fixed at day 3 in vitro with 4% PFA. Primary antibodies were anti-myc (1:2,000), anti-FLAG (1:2,000), anti-GAKIN monoclonal antibody (1:1,000), and anti-PIPBP rabbit serum (1:500). Secondary antibodies were Alexa 594–conjugated anti-mouse and anti-rabbit antibodies (Invitrogen). The intensity of GFP-Akt-PH was observed by fluorescence microscopy using a 40× magnification setting. The expression of several GAKIN constructs was detected by staining with an anti-FLAG antibody. The number of neurites with and without the accumulation of GFP-Akt-PH fluorescence was manually counted. At least three independent experiments were performed for each construct (>70 cells and >300 neurites were observed for each experimental condition). Methods for preparing the hippocampal cell cultures were essentially the same as those described previously (). Hippocampi were dissected from mice at embryonic day 16.5, and dissociated cells were plated in plating medium (DME supplemented with 10% FCS and 5% horse serum) onto glass coverslips coated with poly--lysine. After neurons attached to the substrate, the medium was exchanged to neuronal culture medium (MEM [Invitrogen] with B27 supplement [Invitrogen], 0.6% glucose, and 1 mM sodium pyruvate [Sigma-Aldrich]). Around 18 h after plating, neurons were transfected with various DNA constructs using Lipofectamine 2000. Neuronal polarity was assessed by determining the percentage of neurons with a single long process that was at least twice as long as the other processes. For each construct, at least three independent experiments were performed. The test was used for the statistical significance. Fig. S1 shows that GAKIN and PIPBP form a complex on PIP liposomes. Fig. S2 shows that relative fluorescence intensity was measured to show the accumulation of PIP at the tip of the neurite in PC12 cells and in neurons. Fig. S3 demonstrates that low-level expression of GFP-Akt-PH detector does not affect neuronal polarity. Fig. S4 shows the effect on PIP accumulation by overexpression of full-length GAKIN and DN-GAKIN in hippocampal neurons. Fig. S5 depicts the distribution of several GAKIN constructs expressed in hippocampal neurons. Videos 1 and 2 show that GAKIN transports PIP liposomes via PIPBP in vitro. Online supplemental material is available at .

Polarity, an essential property of eukaryotic cells, allows yeast cells to bud and mate, epithelial cells to form apical and basolateral surfaces, neurons to form synapses, fibroblasts to heal wounds, and leukocytes to crawl to sites of infection. These behaviors require the orientation of polarity toward external cues that are detected by cell surface receptors, which trigger a complex interplay between Rho GTPases and the actin and microtubule cytoskeletons. At the cell's leading edge, this interplay often depends on phosphatidylinositol-3,4,5-tris-phosphate (PIP3), a membrane lipid, and creates positive feedback loops (; ). In differentiated HL60 (dHL60) cells, a neutrophil-like cell line, polarity is mediated by two divergent and competing sets of signals, both triggered by a single species of receptor (). A tripeptide chemoattractant, f-Met-Leu-Phe (fMLP), triggers frontness (protrusive filamentous actin [F-actin] in pseudopods) by stimulating receptor-mediated activation of a trimeric G protein, Gi, which in turn initiates a signaling cascade in which positive feedback loops linking PIP3, Rac, and F-actin create robust pseudopods (; ; ). fMLP stimulates backness (contractile actomyosin) by inducing the receptor-dependent activation of G12 and G13, which promote the activity of RhoA, a Rho-dependent kinase (p160–Rho-associated coil-containing protein kinase [ROCK]), and myosin II (). In a polarized cell, RhoA-dependent backness confines frontness to pseudopods (), whereas Rac-dependent frontness reciprocally constrains backness to the cell's trailing edge (). The ability of frontness and backness to inhibit one another locally helps to explain the segregation of these two responses in a polarized cell. However, it does not explain how uniformly applied fMLP elicits the formation of a single stable front rather than many in the absence of any spatial cue. Formation of a single stable front is similarly elicited by applying a uniform stimulus to many other cells: yeast, neurons, and amoebae form one shmoo tip (), one axon (), or one pseudopod (; ), respectively. In this study, we report that fMLP cannot elicit the formation of a single stable pseudopod in dHL60 cells treated with isoform-selective inhibitors of phosphatidylinositol 3′-kinases (PI3Ks). Inhibition of PI3Ks causes these cells to form pseudopods that are multiple, weak, and transient, leading to the loss of persistent migration and impaired chemotaxis. We also show that PIP3 stabilizes polarity in two ways: first, by locally enhancing Rac activity to stabilize frontness at the leading edge; and second, by stimulating the activation of Cdc42, which promotes RhoA-dependent backness at the trailing edge, thereby preventing the formation of multiple pseudopods. To explore the roles of PIP3 in controlling polarity and chemotaxis, we assessed the effects of compounds that inhibit different subsets of four class I PI3K isoforms expressed in dHL60 cells (; ; ; and unpublished data). From a wide range of PI3K-inhibiting compounds, we chose five that show distinct patterns of isoform selectivity (Table S1, available at ) in vitro (, ): IC87114 (selectively inhibits δ), TGX-115 (β and δ), PI-103 (α > β and δ), PIK-90 (α, γ, and δ), and PIK-93 (α and γ > δ). None of the five compounds is excluded from intact dHL60 cells (supplemental Results), and all are more potent and selective than classical PI3K inhibitors (). Of these compounds, PIK-90 and -93 are unique in their potencies for inhibiting PI3Kγ in vitro (Table S1), for completely inhibiting the fMLP-stimulated phosphorylation of Akt, a kinase downstream of PIP3 (phosphorylated Akt [pAkt]; Fig. S1 and Table S2), for preventing the accumulation in pseudopods of the fluorescent PIP3 probe pleckstrin homology domain (PH)–Akt-YFP (Fig. S2 A), and for impairing polarity and chemotaxis (). These results are consistent with observations (; ; ) in PI3Kγ knockout mice. Consequently, we used these two inhibitors to effectively inhibit PI3Kγ, although contributions from other isoforms to these responses cannot be completely ruled out (supplemental Results). Treatment of dHL60 cells with PIK-90 or -93 impairs consolidation and stability of the leading edge formed after treatment with uniform fMLP, whereas the other three compounds do not (). None of the compounds significantly reduces the fMLP-dependent accumulation of total F-actin (Fig. S2 B), but PI3Kγ inhibition by treatment with PIK-90 or -93 alters its localization, as shown by the multiple pseudopods in . Multiple F-actin–containing pseudopods are twice as frequent in PIK-90– or -93–treated cells 3 min after exposure to uniform fMLP relative to control cells or cells treated with the three other compounds (Table S3, available at ). Time-lapse microscopy () revealed that PIK-90 treatment destabilizes pseudopods of living cells exposed to uniform fMLP. Control cells (, left) typically polarize, form a single pseudopod, and crawl efficiently in one direction for several minutes; in contrast, the leading edges of PIK-90–treated cells (, right) persist for a short time (1–2 min) and retract, to be replaced by a leading edge at another site (Videos 1 and 2). Consequently, unlike control cells stimulated with uniform fMLP, PIK-90–treated cells fail to migrate persistently in one direction. In fMLP gradients, PIK-90 and -93 substantially reduce the chemotactic index (CI; the ratio of a cell's displacement in the correct direction to the actual length of its migration path), whereas inhibitors selective for other PI3K isoforms do not (). In addition, PI3Kγ inhibition by either PIK-90 or -93 triples the cells' turning frequency, but the other three inhibitors have no effect (). PIK-90–treated cells migrate in jerky trajectories that are marked by frequent turns and less persistent orientation toward the source of attractant, which is in contrast to the straighter paths of control cells or cells treated with IC87114 (Fig. S2 C). Nonetheless, the PIK-90–treated cells interpret the fMLP gradient correctly in that their wavering trajectories accomplish net migration in the up-gradient direction (Fig. S2 C). These results differ from our previous observations () with LY294002 or wortmannin in that dHL60 cells treated with either of these nonselective PI3K inhibitors showed poorly developed pseudopods and moved hardly at all (). We suspect that this difference reflects the reported inhibition by the two nonselective agents of lipid and protein kinases that are distinct from PI3Ks (supplemental Results; ). Using pull-down assays or immunoblots of extracts of fMLP-treated cells (), we asked how frontness signals are affected by PIK-90 and latrunculin B (LatB), a toxin that blocks the formation of actin polymers by sequestering monomeric actin (). LatB strongly reduces the fMLP-dependent accumulation of Rac-GTP and modestly reduces activation of the Rac- and Cdc42-dependent kinase p21-activated kinase (PAK), which is assessed by the accumulation of phosphorylated PAK (pPAK). LatB modestly reduces fMLP-dependent pAkt accumulation as described previously () but has no effect on the accumulation of Cdc42-GTP. These effects presumably reflect the interruption of F-actin's participation in the PIP3–Rac–F-actin–dependent feedback loop in pseudopods (; ; ) and suggest that Cdc42 activation is not subject to that feedback. Note that shows results at a single time point (1 min); careful time course analysis showed that all four responses (Cdc42-GTP, Rac-GTP, pAkt, and pPAK) peak at 1 min in control cells and in cells treated with LatB or PIK-90 (not depicted). Effects of PIK-90 () show a different pattern: the complete inhibition of fMLP-dependent pAkt and Cdc42-GTP accumulation, modest inhibition of pPAK accumulation, and severe inhibition of Rac activation. We suspect that the small residual Rac activation seen in PIK-90–treated cells suffices to account for the fMLP-dependent elevation of normal amounts of F-actin (Fig. S2 B) and the formation of transient but multiple fMLP-stimulated pseudopods (; and Table S3). In keeping with this idea, the expression of the dominant-negative Rac mutant (Rac-N17) inhibits pseudopod formation in PIK-90–treated cells (). This result, of course, does not completely rule out a Rac-independent contribution to pseudopod formation because the Rac mutant could affect mechanisms that are distinct from the activation of endogenous Rac. Consistent with the observed modest inhibition of pPAK, the p21-binding domain (PBD) of PAK fused to CFP (PAK-PBD-CFP) still translocates to the periphery of PIK-90–treated cells that are treated with fMLP (). This probe was shown previously to reflect the localization of activated Rac in dHL60 (). shows that fMLP stimulates translocation of PAK-PBD-CFP to the entire cell periphery at 1 min of PIK-90–treated cells, whereas at 3 min, the probe accumulates at multiple sites, which correspond to multiple leading edges. The same is true of cells in which Cdc42 signals downstream of Cdc42-GTP are inhibited by the expression of ΔC-WASp. Combined treatment with PIK-90 and LatB reduces the activation of both Rac and PAK to virtually nil (). Compared with the effects of the two treatments alone, the reduction of pPAK accumulation caused by the combination of inhibitors is very much greater, suggesting that separate pathways dependent on PIP3 and F-actin normally converge to stimulate the maximal activation of PAK. With respect to the accumulation of Rac-GTP, each treatment substantially reduces the fMLP response, but the two together appear to induce even greater inhibition. fMLP promotes RhoA activation by a pathway that requires PIP3 and the activation of Cdc42 (). fMLP treatment of intact neutrophils induces RhoA to associate with the particulate fraction of extracts from these cells (); PIK-90 treatment reduces the basal RhoA content of particulate fractions and prevents the response to fMLP (). The same effects are seen with a second assay () that is based on the quantitative assessment of fluorescence resonance energy transfer (FRET) signals triggered by GTP occupying the guanine nucleotide-binding pocket of a recombinant RhoA biosensor (). As we previously reported (), fMLP increases RhoA FRET in dHL60 cells. PIK-90 lowers basal RhoA FRET and prevents fMLP from elevating it (). (For representative RhoA FRET images of individual cells corresponding to the averaged values shown in , see Fig. S3; available at ). In contrast, we previously reported () that LY294002 appeared to increase basal RhoA activity. This effect may reflect the documented () effects of LY294002 on kinases other than PI3Ks (for further discussion, see supplemental Results). PIK-90 treatment prevents the fMLP-dependent activation of Cdc42 (), and Cdc42-GTP mediates RhoA activation in response to fMLP (). Expression of a constitutively active Cdc42 mutant, Cdc42-V12, significantly increases the mean RhoA FRET of cells that were not exposed to fMLP (). In addition, the expression of ΔC-WASp prevents fMLP from elevating the RhoA FRET signal (). In addition to increasing RhoA activity (), Cdc42-V12 in dHL60 cells prevents fMLP from inducing ruffling and pseudopod formation, as previously described () and confirmed in . In retrospect, we overlooked the potential importance of this surprising Cdc42 effect, which differs strikingly from the Cdc42-stimulated formation of the filopodia seen in fibroblasts (). Now we find that Cdc42-V12 inhibits ruffling and pseudopod formation because it induces the activation of a RhoA-dependent pathway that inhibits frontness responses. Indeed, morphologic effects of Cdc42-V12 are reversed by treating cells with Y27632 to inhibit p160-ROCK (), a kinase that links RhoA to the phosphorylation of myosin light chains and actomyosin contraction. Although only 1/10 Cdc42-V12–expressing cells made ruffles all around the cell at 1 min and formed an F-actin–containing leading edge at 3 min in response to fMLP, 9/11 cells expressing the mutant protein did so after exposure to Y27632 (unpublished data). The morphologic effect of constitutively active Cdc42 () closely mimics the reported () effect of constitutively active RhoA, which was also reversed by exposure to Y27632. Inhibiting Cdc42 with ΔC-WASp has an effect opposite to that of Cdc42-V12; that is, like PIK-90 and -93, it reduces fMLP-stimulated RhoA activity (). This inhibition of RhoA is paralleled by effects on fMLP-stimulated polarity and chemotaxis, which also resemble effects of the PI3K inhibitors. We reported previously () that dominant-negative Cdc42 (Cdc42-N17) or ΔC-WASp cause cells to form multiple transient pseudopods in response to fMLP (). In the course of this study, we repeated these experiments and obtained virtually identical results (unpublished data). In addition, Cdc42-N17–expressing cells migrated toward an fMLP-containing micropipette in jerky trajectories marked by multiple turns (unpublished data); these trajectories closely resembled those of cells treated with PIK-90 (Fig. S2 C). We infer from these results that the fMLP-dependent activation of Cdc42, like the accumulation of PIP3, normally stimulates backness signals. Because the loss of fMLP-dependent RhoA activation in PIK-90–treated cells is accompanied by the reduced activation of Rac (), we used Rac mutants to ask whether Rac-GTP plays a role in activating RhoA. Expression of a constitutively active Rac mutant, Rac-V12, failed to elevate RhoA biosensor activity in unstimulated cells (), and Rac-N17, a dominant-negative mutant, did not inhibit the fMLP response (). These results are directly opposite to those seen with Cdc42-V12 and ΔC-WASp, respectively. We infer that Rac activation is neither sufficient nor necessary for RhoA activation. Finally, we reported () that G12 and G13, which are trimeric G proteins, mediate fMLP stimulation of backness based on observations with constitutively active and dominant inhibitory mutants of these proteins. In accord with these findings, the expression of dominant-negative G12/13 prevents fMLP from activating the RhoA biosensor (). Because the multiple fMLP-dependent pseudopods of G12/13-inhibited cells accumulate prominent PH-Akt-GFP fluorescence (), they presumably also accumulate PIP3 and Cdc42-GTP in response to fMLP, although such responses could not be quantitated. Therefore, we suspect that RhoA activation requires simultaneous positive inputs from G12/13 and the long-range pathway triggered by PIP3/Cdc42, although a strong Cdc42 signal (from Cdc42-V12) can override the requirement for G12/13. Cdc42 may also activate RhoA through G12/13 independently of the receptor, as suggested by the increased basal RhoA activity in cells overexpressing a constitutively active Cdc42 in the absence of fMLP. We proposed (; ) that dHL60 cells polarize and break their symmetry when fMLP stimulates competing Gi-dependent frontness and G12/13-dependent backness responses whose mutual incompatibility causes them to segregate into separate membrane regions. However, breaking symmetry is not enough. Cells treated with PIK-90 or -93 break their symmetry in response to fMLP quite easily but do not maintain a single persistent pseudopod and a single persistent back; consequently, they cannot migrate persistently in one direction in uniform fMLP or migrate efficiently up fMLP gradients (). Intuitively, we might have suspected that competing frontness and backness responses alone would not suffice to maintain stable polarity. Responses of approximately equal strengths would probably produce a shifting array of multiple transient fronts and backs, whereas the excess strength of one response might easily enable it to win the competition. For instance, positive feedback loops between frontness signals could generate pseudopods covering the entire cell surface. How, then, do neutrophils maintain stable polarity? Our experiments indicate that PIP3 and Cdc42, which were generated as part of the frontness response, exert a powerful stabilizing effect by strengthening pseudopods and by promoting long-range activation of Rho-dependent backness ( and ). Inhibiting PI3Kγ or Cdc42 activity disrupts this stabilizing effect, leading to the formation of multiple transient pseudopods. Normally, however, PIP3 and Cdc42 signals originating in a strong pseudopod effectively promote G12/13- and RhoA-dependent actomyosin-based contractility at the cell's opposite end, and the proportionately stronger back increases the likelihood of achieving stable asymmetry with a single robust pseudopod and one back. This scenario resembles local excitation global inhibition models (; ), which combine local positive feedback and a globally active diffusible inhibitor. For amoebae of , the local excitation global inhibition model (; ) proposes that PIP3 at the front promotes actin polymerization with positive feedback, whereas an unidentified signal generates a rapidly diffusible mediator that activates phosphatase and tensin homologue (PTEN), a PIP3 phosphatase, at the back of the cell. The Rho pathway is reported to enhance PTEN localization at the trailing edge of mouse neutrophils (). However, this effect is probably not essential for stabilizing neutrophil polarity because we () and others () could not detect PTEN localization to the trailing edge of dHL60 cells, and neither excess nor the depletion of PTEN altered gradient sensing (). Just as actomyosin contraction constrains the pseudopod to a single location (), local effects of protrusive F-actin at the front of the cell reciprocally inhibit RhoA activation and actomyosin-based contraction (). How do PIP3 and Cdc42, which are generated at the leading edge, exert long-range positive regulation of RhoA outside pseudopods ()? Long-range regulation involving PIP3 or Cdc42 has been described in other systems, but mechanisms are poorly understood. One precedent is the sharp segregation of Cdc42-GTP and RhoA-GTP into spatially separated concentric rings that surround and promote the closure of plasma membrane wounds in frog oocytes (). Active Cdc42 in this system is physically separate from active RhoA but is nonetheless required for RhoA to become active. A second precedent is the ability of PIP3 at the leading edge of amoebae to trigger a kinase cascade that promotes myosin II contraction at the back (). Instead of Cdc42 and Rho, the pathway depends on the PIP3-dependent activation of Akt/PKB and subsequent activation of PAKα, a PAK1 homologue located at the back of the cell. How the message moves from Akt/PKB at the front to PAKα at the back is unknown. One possibility in dHL60 cells is that PIP3 and Cdc42 lead to the generation of a phosphorylated protein or a cytosolic second messenger that diffuses rapidly from the front to the back. The putative diffusible mediator could augment activation of a Rho guanine nucleotide exchange factor or inhibit activity of a GTPase-activating protein that inactivates Rho. Alternatively, Cdc42 could also promote transport of a regulator from front to back via endocytosed vesicles and microtubules. A possible role for microtubules is consistent with the documented ability of Cdc42 to regulate the interaction of microtubules with the cell cortex (; ), with microtubule-dependent delivery of a Rho guanine nucleotide exchange factor to the plasma membrane of S2 cells (), and with microtubule-dependent localization of RhoA activity at the cleavage plane of frog oocytes (). Cdc42 has been reported to regulate actomyosin contraction via myotonic dystrophy kinase-related Cdc42-binding kinase independently of ROCK (). Such a role for myotonic dystrophy kinase-related Cdc42-binding kinase is unlikely to be quantitatively important in fMLP-treated dHL60 cells because inhibiting ROCK with Y27632 rescues the inability of Cdc42-V12–expressing cells to form pseudopods (). Why are pseudopods of PIK-90–treated cells weak and transient () despite the accumulation of quantitatively normal amounts of F-actin (Fig. S2 B)? The most likely explanation is that reduced PIP3 accumulation dramatically reduces the activation of Rac (), an essential positive regulator of actin polymerization. Our results also suggest that the loss of PIP3 inhibits the consolidation of Rac activity into one region of the cell periphery so that the remaining active Rac, which is detected by the localization of PAK-PBD-CFP, localizes in multiple transient pseudopods (). Maintaining strong pseudopods may require positive PIP3-dependent signals mediated by Cd42 as well as Rac. Indeed, the pseudopod defects of cells exposed to PIK-90 and -93 inhibitors () closely resemble those of cells expressing inhibitors of Cdc42 (): in both cases, pseudopods are not only multiple but are also transient and weak. One potential integrator of Rac and Cdc42 signals in pseudopods is PAK1, which can be activated by both GTPases, phosphorylates numerous cytoskeletal regulators, and promotes the formation of protrusive actin with the inhibition of contractile actomyosin (). In favor of this possibility, PAK1 in dHL60 cells appears to be regulated by a PIP3-dependent signal in addition to an F-actin–dependent positive feedback signal; either PIK-90 or LatB treatment alone only inhibits a portion of the pPAK response, but inhibiting both signals blocks the response almost completely (). Finally, it is also possible that Cdc42 augments pseudopod strength and stability by stimulating actin polymerization directly via Wiskott-Aldrich Syndrome proteins (). Antibodies against Akt, phosphorylated at T308 (pAkt) and PAK and phosphorylated at S199/204 (pPAK), were obtained from Cell Signaling Technology. Mouse monoclonal antibody against RhoA was purchased from Santa Cruz Biotechnology, Inc. Mouse antitransferrin receptor antibody was purchased from Zymed Laboratories. LatB and Y27632 were obtained from Calbiochem, and GST-PAK-PBD and antibodies against Rac (isoforms 1 and 2) were purchased from Pierce Chemical Co. Rhodamine-phalloidin was obtained from Invitrogen, and human fibronectin was obtained from BD Biosciences. Protease and phosphatase inhibitor cocktails and fMLP were obtained from Sigma-Aldrich. HRP-conjugated donkey anti–rabbit and anti–mouse IgG were purchased from GE Healthcare. Glass capillaries obtained from Frederick Haer Co. were converted into micropipettes with a puller (P-87; Sutter Instrument Co.). Compounds IC87114, TGX-115, PI-103, and PIK-93 were dissolved in DMSO; PIK-90 was dissolved in 50:50 DMSO/HO; all compounds were stored at −20°C. Final dilutions were performed in cell media, with a final DMSO dilution of 0.5% for PIK-90 and 1.0% for the other compounds. Myc-tagged V12-Cdc42 cloned in pTET7 was described previously (); to activate its expression, this construct was cotransfected with a plasmid constitutively expressing a tetracycline repressor VP16 fusion protein, as also described previously (). Myc-tagged Rac-V12 and Rac-N17 were purchased from the Guthrie cDNA Resource Center. ΔC-WASp () and PAK-PBD-CFP () were described previously. The RhoA biosensor construct used for transient transfection was described previously (). Quasi-stable expression of the RhoA biosensor or PH-Akt-YFP was accomplished by lentiviral transfection as described previously (). G12- and G13-DN constructs (residues 326–379 for Gα12 and residues 321–377 for Gα13) were gifts from Y. Takuwa (Kanazawa University, Kanazawa, Japan) and were described previously (). Transient transfection with the RhoA biosensor, Cdc42-V12, ΔC-WASp, Rac-V12, Rac-N17, or G12/G13-DN was previously described (); experiments were performed 4 h after transfection. Procedures for cultivation and DMSO-stimulated differentiation of HL60 cells have been previously described (). Live dHL60 cells were imaged in 1.5% human albumin in modified HBSS (mHBSS) at room temperature using an inverted microscope (Eclipse TE200; Nikon) equipped with a Photometrics cooled CCD camera (CE300; Roper Scientific) driven by DeltaVision software (Applied Precision). Pictures shown were taken with a 60× NA 1.40 oil planApo objective (Nikon), whereas tracking experiments were performed with a 20× NA 0.75 planApo objective (Nikon). Cell trajectories were tracked using SoftWorx software (Applied Precision). The CI was calculated as described previously (): CI = (AO − BO)/AB, where O is the position of the micropipette; A and B are the cell's position at time = 0 and time = 25 min, respectively; AO is the distance from A to O; BO is the distance from B to O; and AB is the actual length of the cell's curving migration path from HL60 A to B. Turns in a cell trajectory were quantified by approximating the cell's trajectory over successive 2-min intervals, with a vector derived from a least squares deviation fit to five successive positions of the leading edge (coordinates are assessed at 30-s intervals). Angles between each successive pair of vectors were computed; a turn was defined as a deviation of one 2-min vector from its predecessor by 60° or more in either direction. dHL60 cells were preincubated in suspension with IC87114, TGX-115, PI-103, PIK-90, PIK-93, or no treatment for 40 min, centrifuged for 5 min at 2,000 rpm at room temperature in a J6-B centrifuge (Beckman Coulter), resuspended in mHBSS containing the respective agent at the same concentration, allowed to stick to fibronectin-covered coverslips, and subjected to stimulation with a uniform concentration of 100 nM fMLP for 3 min. Cells were fixed in 3.7% PFA and stained with 10 units/ml rhodamine-phalloidin for 15 min. dHL60 cells were preincubated for 40 min with 40 μg/ml LatB, 0.5 μM PIK-90, or both, centrifuged for 5 min at 2,000 rpm at room temperature in a J6-B centrifuge (Beckman Coulter), resuspended in mHBSS (2 × 10 cells in 0.5 ml per condition), and stimulated or unstimulated for 1 min with 100 nM fMLP. The reaction was stopped by adding 0.5 ml of 2× lysis buffer at 4°C (1× lysis buffer: 25 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM MgCl, 1% NP-40, 1 mM DTT, and 5% glycerol, protease inhibitor, and phosphatase inhibitor cocktails). Subsequent steps were performed as described in the protocol attached to the Cdc42 and Rac pull-down kit (Pierce Chemical Co.). dHL60 cells with or without 40 min of pretreatment with 0.5 or 1 μM PIK-90 were stimulated for 0 or 1 min with 100 nM fMLP. Particulate fractions were prepared as described previously (). The level of RhoA in the particulate fraction was normalized to that of the transferrin receptor. This assay was first described by . For comparison, the normalized data were further standardized by setting the unstimulated control value to one. Cells quasi-stably expressing the RhoA biosensor were pretreated with or without 1 μM PIK-90 for 40 min, resuspended in 1.5% human albumin in mHBSS, and plated on fibronectin-coated coverslips before treatment with or without 100 nM fMLP for 3 min. Cells transiently cotransfected with the RhoA biosensor and Cdc42-V12, ΔC-WASp, Rac-V12-myc, Rac-N17-myc, or G12 and G13 dominant-negative mutants were subjected to the same procedure without exposure to PIK-90. Images were acquired at room temperature, and data were processed as described previously (). Supplemental material provides details about the selectivity of PI3K inhibitors and methods for the F-actin. Table S1 provides IC50 values for inhibiting class I PI3K isoforms in vitro. Table S2 provides IC50 values of five compounds for inhibiting pAkt accumulation in dHL60 cells. Table S3 provides data on how PI3Kγ inhibitors induce cells to form multiple leading edges in response to fMLP. Fig. S1 shows the effects of five PI3K inhibitors on the phosphorylation of Akt in dHL60 cells. Fig. S2 shows the effects of inhibiting PI3Kγ on PIP3 localization, accumulation of F-actin, and chemotaxis. Fig. S3 shows the subcellular distribution of the FRET signal. Video 1 shows YFP-actin–expressing cells in uniform stimulation, and Video 2 shows PIK-90–treated YFP-actin–expressing cells in uniform stimulation. Online supplemental material is available at .

Cell adhesion to the extracellular matrix plays an essential role during cell migration. Transmembrane integrins at focal adhesions (FAs) undergo cycles of matrix attachment, cytoskeletal recruitment, induction of contractile forces, and disassembly (). Nascent FAs form at or near anterior cell margins and mature into larger FAs under the middle and rear of the migrating cell. Some proteins, e.g., vinculin, are present for nearly the lifetime of a FA, whereas other proteins appear during maturation (). Although both nascent and mature FAs are associated with stress fibers containing F-actin, α-actinin, and bipolar myosin II filaments (), nascent FAs are responsible for most of the contractile forces (; ). Mature FAs retard the rate of cell translocation () and generate signals for cell survival and transcriptional activation (; ). FA disassembly/turnover is facilitated by Src family tyrosine kinases, adaptor proteins such as Crk, extracellular factor–regulated kinases (ERKs), selective proteolysis, and/or microtubules (; ) and may be enhanced in fast-moving cells, such as immune cells and many tumor cells, especially invasive carcinomas (; ; ). Reorganization of FAs stimulated by mechanical cues involves zyxin and associated proteins (; ). The appearance of zyxin at FAs coincides with the loss of strong traction forces (), which is consistent with a role during maturation (). Other proteins related to zyxin found at mature FAs are lipoma-preferred partner (LPP; ) and thyroid receptor–interacting protein (TRIP6; ), which is also called zyxin-related protein 1 (ZRP-1) and Opa-interacting protein 1 (; ). Each zyxin family member contains an N-terminal domain that promotes local actin filament assembly and a C-terminal domain with three LIM domains, which are zinc finger motifs found in many cytoplasmic and nuclear proteins (). The TRIP6 LIM domains bind to other proteins with LIM domains (), to endoglin/CD105, which is a component of the TGF-β receptor complex (), to the protein tyrosine phosphatase (PTP) nonreceptor type 13 (PTPN13, PTP1E, FAP-1; ; ), to Crk and the Crk-associated substrates (CAS) p130 and CASL/HEF1 (), and to the membrane-bound, G protein–coupled lysophosphatidic acid 2 (LPA) receptor (). The latter two interactions are potentiated by Src phosphorylation of Tyr-55 in TRIP6 (), suggesting regulatory interactions between the TRIP6 N and C termini. The TRIP6 C terminus also binds to class 1 PDZ motifs (; ), and the N terminus contains a nuclear export signal and transactivates gene expression (; ; ). TRIP6 has been implicated in the organization of actin filaments and in the control of cell migration with conflicting results. Knockdown of TRIP6 in human endothelial cells results in the disruption of cytoskeletal actin filaments (). Overexpression of TRIP6 in mouse fibroblasts slows cell movements (), whereas overexpression in SKOV3 ovarian carcinoma cells increases LPA-mediated cell migration and ERK activation (; ). Nothing is known about the role of TRIP6 in cell–substrate adhesion or FA dynamics, leaving open questions about cell type–specific factors and regulation of different steps during cell migration. In our exploration of membrane domains in highly motile cells, we have characterized a detergent-resistant plasma membrane fraction from neutrophils (; ). This fraction contains F-actin, α-actinin, myosin II, cholesterol-organizing proteins, heterotrimeric G proteins, and the Src family kinase Lyn. The most tightly bound peripheral membrane protein, called supervillin (SV), is also relatively enriched in most carcinoma cell lines (; ). SV binds tightly to F-actin and the S2 regulatory domain of myosin II, as well as to membranes (; ), and is implicated in both nuclear and membrane processes. Although SV contains functional NLS () and can transactivate androgen receptor–mediated gene expression (), ≥95% of total SV localizes with membranes in differentiated cells (; ). As the only membrane protein known to bind both myosin II and F-actin, SV may organize actin and myosin II filaments at the membrane and/or mediate actin-independent binding of myosin II to membranes (; ). SV sequences can induce F-actin cross-linking and bundling, myosin II mislocalization, and disruption of vinculin localization at ventral cell surfaces (; ). Alternatively spliced forms of SV, which are found in striated and smooth muscle (archvillin [AV] and smooth muscle AV [SmAV], respectively), are implicated in costamere organization, ERK signaling, contractility, and myogenic differentiation (; ). We now show that SV promotes a process leading to the loss of cell–substrate adhesion and large FA. We have also identified the region of SV responsible for this activity as aa 342–571 and characterized relevant SV342-571 binding partners. SV-mediated down-regulation of FAs involves binding to TRIP6 and, possibly, also to LPP. SV and TRIP6 negatively regulate large FAs, either by blocking maturation or by facilitating disassembly. This interaction may regulate cell–substrate adhesion through the recruitment of actin and/or myosin II to TRIP6-associated signaling networks and could play a role in FA signaling to the nucleus. EGFP-SV at low levels overlaps with vinculin at or near FAs on the basal surfaces of CV1 () and COS7 (Fig. S1, available at ) cells. This overlap is more pronounced at large, mature FAs in the center and posterior of the cell than at newly formed FAs at the cell periphery (Fig. S1, arrow). SV also localizes along associated stress fibers; this signal increases disproportionately with increasing amounts of EGFP-SV (). SV targeting to large FAs apparently reduces the number of these structures (). Large vinculin-labeled FAs are abundant in COS7 cells that overexpress EGFP (), but most large vinculin spots are lost in cells that overexpress EGFP-SV (). We find an ∼1:1 ratio of EGFP-SV to endogenous SV in lysates from preparations with ∼20% transfected cells (), indicating that the average level of EGFP-SV in the transfected cells is approximately five times greater than endogenous SV, for a total of approximately six times more SV than in untransfected cells. We normalized the amount of fluorescence in cells transfected with EGFP alone or EGFP-SV so that cells with approximately sixfold overexpression of SV exhibit ∼50% of maximal fluorescence intensity (). The number of large FAs (≥10 μm) in cells expressing EGFP-SV with ∼30–90% of maximal fluorescence are less than half those observed in cells expressing equivalent amounts of EGFP alone or in untransfected cells (). Thus, even an approximately threefold increase in SV levels down-regulates the number of large FAs. Cell–substrate adhesion also correlates inversely with SV levels (). After preparative FACS, COS7 cells expressing EGFP-SV contain ∼6.3-fold more SV than cells expressing EGFP (, top) and are significantly less adherent to fibronectin-coated coverslips (, bottom). Conversely, cells with ∼10% of endogenous levels of SV adhere more tightly to fibronectin (, bottom). Collectively, these results show an inverse correlation between SV levels and FA function, although any increases in the number and/or size of mature FAs after SV knockdown lack statistical significance (unpublished data). To identify SV sequences responsible for loss of FA integrity, we quantified the number of total and large FAs in COS7 cells expressing different EGFP-tagged SV constructs (). Full-length SV (, SV1-1792) reduces the number of large vinculin foci ( and ), but has relatively little effect on the number of the more abundant, smaller foci near cell peripheries (). The result is the absence of a statistically significant effect on the number of total FAs per cell (). The SV N terminus (SV1-830) decreases the number of both total and large FAs, whereas the SV C terminus (SV830-1792) exhibits effects similar to those of full-length SV (). Thus, a sequence with major impact on FA stability resides within the SV N terminus. Further dissection of N-terminal SV sequences identifies SV1-174 and SV342-571 as FA disruption sequences (). Both of these sequences significantly reduce the number of large FAs (), but only SV342-571 significantly reduces the total number of vinculin foci per cell (). COS-7 cells transfected with SV or SV N-terminal sequences also exhibit decreased numbers of large F-actin bundles (). Although full-length SV with its three F-actin–binding sites () increases the number of thin actin structures (), SV overexpression greatly decreases the number and size of large, straight actin bundles (, compare b with a). However, most SV-expressing cells still contain at least one large basal F-actin bundle (). Because most of these bundles exhibit periodic staining for α-actinin () and myosin II (unpublished data), we refer to them here as “stress fibers,” although they are smaller and less well organized than are stress fibers in fibroblasts. Consistent with their effects on FA structure (), SV1-830 and SV342-571 significantly decrease the number of COS7 cells containing at least one stress fiber, as compared with cells expressing EGFP alone or other EGFP-SV fragments (). SV1-830 and SV342-571 also decrease the number and thickness of stress fibers in CV1 cells (unpublished data). Thus, SV342-571 contains a site that is primarily responsible for the morphological effects of SV and SV1-830 on FAs and stress fiber organization and function. In support of this conclusion, SV342-571 reduces the contractility of CV1 cells ( and ). Significant percentages of cells expressing EGFP only or other EGFP-SV N-terminal sequences deform flexible silicon substrates, forming “wrinkles” ( and ). In contrast, wrinkles are absent from the areas around most CV1 cells expressing SV342-571 (), even cells near untransfected cells that are actively wrinkling the substrate, thus, demonstrating its local deformability (, arrow vs. arrowhead). Thus, SV342-571 contains sequences capable of reducing FA structure and function. We hypothesized that a previously unknown interaction mediated the effects of SV342-571 on FAs. The only known binding partner for SV342-571 is F-actin (), but F-actin also binds to SV171-342 and SV570-830 (), neither of which affects FA structure and function to the same extent ( and ). To identify candidate interactors with SV342-571, we undertook an undirected yeast two-hybrid screen. Because SV343-570, SV171-570, and SV343-830 all self-activated reporter gene expression (unpublished data), our bait plasmid encoded SV171-830. 11 prey proteins were identified and confirmed by directed yeast two-hybrid assays (). These proteins were further screened using baits encoding either SV171-343 or SV570-830. Only the longer plasmid (SV171-830) containing SV343-570 sequences interacted strongly with the prey proteins. Two interactors, TRIP6/ZRP-1 and Tctex-1/DYNLT1, account for a majority of the clones identified in these screens (). Ten independent clones encode all three C-terminal LIM domains of TRIP6. Another clone encodes the first two LIM domains of LPP, suggesting that two or more LIM domains are required for binding. Six independent clones encode full-length Tctex-1, which is a dynein light chain (DYNLT1). We confirmed binding interactions with TRIP6 and Tctex-1 using pull-down assays with GST fusion proteins containing SV343-571 (). V5-tagged TRIP6 LIM domains () and V5-tagged full-length Tctex-1 () copellet with glutathione–Sepharose beads containing GST-tagged SV171-830, SV343-571, SV171-571, and SV343-830 (, lanes 2, 5, 7, and 8). Neither V5-tagged protein cosediments with beads containing GST only or SV171-342 (, lanes 3 and 4). A low-affinity interaction with SV570-830 is observed for Tctex-1 (, lane 6). No proteins reactive with the anti-V5 antibody are detectable in control yeast extracts (, lane 9), confirming the binding specificities of TRIP6 and Tctex-1 for SV342-571. Binding of SV342-571 to TRIP6 or Tctex-1 is also observed in doubly transfected COS7 cells. Myc-tagged TRIP6 LIM domains and Tctex-1 cosediment with GST-tagged EGFP-SV342-571, but not with EGFP-GST alone (, lane 3 vs. 4). Notably, neither Myc-tagged LIM domains from zyxin nor Flag-tagged, full-length TRIP6 cosediments with EGFP-SV342-571-GST. Coimmunoprecipitation of full-length proteins was precluded by the inextractability of full-length SV (; ). The TRIP6 LIM domains and Tctex-1 bind directly to overlapping sequences within SV343-571 (). Purified, bacterially expressed hexahistidine (6×His)-tagged TRIP6 LIM domains () and 6×His-Tctex-1 () cosediment with glutathione–Sepharose beads containing prebound, purified GST-SV343-571 (, lane 2), but not with beads bound to GST alone (, lane 4). Increasing amounts of 6×His-tagged Tctex-1 compete with 6×His-tagged TRIP6 LIM domains for binding to a fixed, limiting amount of GST-SV343-571 (). Point mutagenesis provides further support for competition between TRIP6 and Tctex-1 for binding to SV343-571. Conversion of the highly conserved SV residues Arg-426 and Tyr-427 to alanines (RY/AA) coordinately reduces binding of GST-SV343-571-RY/AA to both TRIP6 and Tctex-1 by ∼70% (). Similarly, mutagenesis of these residues in EGFP-SV343-571 reduces this protein's effects on stress fiber and FA structure (Fig. S2, available at ), demonstrating the functional importance of the TRIP6/Tctex-1–binding site within SV342-571. To determine whether the SV interaction with TRIP6 and/or with Tctex-1 is involved in SV-mediated changes in FA structure and function, we colocalized these proteins in A7r5 () and COS7 () cells. Endogenous levels of the relatively abundant SmAV () partially colocalize with endogenous vinculin () and TRIP6 () in A7r5 cells on fibronectin (). The SmAV signal is a subset of that for vinculin and TRIP6 in FAs, but stops short of the labeling for these two proteins at cell edges. This result is consistent with a role for the SV– TRIP6 interaction in regulating adhesion, but is uninformative about Tctex-1 because we could not detect Tctex-1 in A7r5 cells (unpublished data). In doubly transfected COS7 cells, Flag-TRIP6, but not Myc–Tctex-1, partially colocalizes with EGFP-SV and vinculin (). Although COS7 cells contain endogenous TRIP6 and Tctex-1 (Fig. S3, available at ), the epitope-tagged versions of these proteins are required for immunofluorescence. Full-length EGFP-SV and Flag-TRIP6 colocalize in membrane surface extensions that lack strong vinculin staining and along F-actin bundles, as well as with vinculin at FAs (). In contrast, little or no colocalization is seen between TRIP6 and the primarily nuclear EGFP () or between Tctex-1 and either EGFP-SV or vinculin (), suggesting that TRIP6, rather than Tctex-1, mediates SV effects at FAs. In fact, TRIP6 may help recruit SV to FAs. Although SV is not required for the FA localization of TRIP6 (Fig. S4, available at ), full-length EGFP-SV lacking the TRIP6-binding site is largely absent from FAs (, SVΔ343-561). Unlike EGFP-SV (, arrowheads, and Fig. S1), EGFP-SVΔ343-561 is associated with branched and curved actin filaments that appear to associate with vinculin foci at only one end (, arrows). Blind scoring of the percentage overlaps of EGFP signals with vinculin at large FAs shows overlap at 49.7 ± 8.8% of the FAs in cells expressing EGFP- SVΔ343-561, compared with 74.2 ± 6.1% in cells expressing EGFP-SV (mean ± SEM; = 220 and 238 FAs from 15 cells each; P < 0.03). This decreased overlap with vinculin is probably an underestimate of TRIP6's effect on SV recruitment to FAs because the SV localization along filaments favors false positives. TRIP6 also modulates adhesion of human A549 lung carcinoma cells to fibronectin (). RNAi-mediated knockdowns of TRIP6 (hTR1, hTR2), but not Tctex-1 (hTx), significantly increase adhesion (). This effect is comparable to that observed upon SV knockdown (hSV) and is consistent with the reported inverse correlation between TRIP6 levels and the number of large FAs (). TRIP6 sequences also partially reverse losses of stress fibers and large FAs induced by SV sequences capable of binding to these proteins (). Exogenously coexpressed TRIP6 LIM domains, but not zyxin LIM domains or Tctex-1, reduce the effects of SV domains, EGFP-SV1-830 and EGFP-SV342-571, on stress fibers in COS7 cells (). Full-length TRIP6 partially rescues the effects of full-length EGFP-SV () and EGFP-SV1-830, although not that of SV342-571 (). The basis for the TRIP6-mediated rescue may be mislocalization of TRIP6–SV complexes because cells with high expression levels of both TRIP6 and EGFP-SV () exhibit increased colocalization of TRIP6 and SV in protrusions lacking vinculin (, arrows, vs. ). Rescued FAs contain TRIP6, but are largely devoid of EGFP-SV (, arrowheads). As expected, TRIP6 has no effect on the phenotype of point (SV-RY/AA) and deletion (SVΔ343-561) mutants of full-length SV that have reduced or no binding to TRIP6, respectively (). Consistent with the observations that multiple SV sequences affect FA structure (), full-length SV mutants deficient in binding to TRIP6 induce a loss of large FAs in both the presence and absence of TRIP6 (). This result emphasizes the importance of additional SV-binding partners. We show that SV down-regulates FA structure and function, and that the mechanism involves interactions with TRIP6. Decreased levels of either protein increase cell adhesion to fibronectin. Increased SV levels decrease cell adhesion, as well as the number of stress fibers and large, mature FAs. Although more than one region of SV deleteriously affects FAs, the SV sequence with the largest effect on FA structure and function is SV342-571, which binds directly to Tctex-1 and the C-terminal LIM domains of TRIP6. SV and TRIP6 colocalize at mature FAs, and optimal SV recruitment to FAs requires binding to TRIP6. TRIP6 and the TRIP6 LIM domains partially rescue disruptive effects of SV sequences on FAs and stress fibers. Specificity is indicated by the lack of effect of Tctex-1 or the zyxin LIM domains on SV phenotypes. Thus, binding to SV342-571 is necessary, but not sufficient, to reverse SV effects on FAs. The TRIP6 N terminus may shield the C-terminal LIM domains from SV342-571 in the absence of a regulatory signal, as has been proposed for other LIM domain proteins (; ). No direct interaction between SV sequences and the TRIP6 N terminus was detected in either yeast two-hybrid or pull-down assays. Nevertheless, full-length TRIP6 rescues the disruptive effects of longer SV proteins, implying the possibility of regulatory cross-talk between the TRIP6 N terminus and SV sequences other than SV342-571. Observations that zyxin influences motility, adhesion, and stress fiber formation (; ) are reminiscent of those observed upon overexpression of SV sequences. However, SV342-571 does not bind zyxin. In conjunction with other recent observations (), these results suggest that the members of the zyxin protein family have overlapping, but distinguishable, functions. The loss of adhesion induced by SV overexpression apparently represents a gain of function because this phenotype is opposite that observed after SV knockdown. An SV-induced negative effect on large FAs is supported by the localization of SV with large FAs, which are structures that undergo dynamic remodeling (; ). SV-mediated loss of large FAs also fits with the absence or reduced prevalence of large FAs and stress fibers in cells that contain relatively high amounts of endogenous SV, e.g., carcinomas and neutrophils (; ). Carcinomas and hematopoietic cells also express TRIP6 and/or LPP (; ), which is consistent with a physiological role for interactions with SV at dynamic FAs. Neutrophil FAs must be highly dynamic because they turn over rapidly during immune responses when little or no integrin is synthesized (). Although the rescue of the SV phenotype by the TRIP6 LIM domains may be attributable to a simple dominant–negative effect on TRIP6 function (), the rescue by full-length TRIP6 is more interesting. In cells that overexpress both TRIP6 and wild-type SV, both proteins mislocalize into cell protrusions, sequestering SV away from FAs. Full-length SV proteins deficient in binding to TRIP6 still reduce the number of mature FAs, although these SV mutants are largely absent from FAs; TRIP6 localization at residual (or new) FAs is essentially unaffected. Thus, the FA equilibrium is disturbed by SV mutants that either contain the TRIP6-binding site out of context or contain the other SV FA-targeting sequences in the absence of high-affinity binding to TRIP6. The simplest interpretation of these results is that TRIP6 and SV, together with other associated proteins, act during a FA assembly/disassembly cycle to control the rate of FA turnover. In this working model, TRIP6 helps recruit SV to FAs; SV then, directly or indirectly, either blocks later stages in FA maturation and/or increases the rate of FA turnover. When SV or TRIP6 levels are limiting, cell–substrate adhesion increases because FAs are locked “on.” Increasing amounts of the limiting protein decrease cell adhesion by increasing the rate of maturation of adhesive nascent FAs into less adhesive mature FAs and/or by increasing the rate of FA disassembly. In this model, TRIP6 overexpression helps restore the balance between FA assembly and disassembly in SV-overexpressing cells by (a) accelerating the formation of new FAs, (b) sequestering SV and proteins required for FA disassembly away from FAs, and/or (c) promoting the recycling of SV-associated proteins that are required for FA assembly. Overexpressed full-length SV that is deficient for binding to TRIP6 may disrupt FAs by interacting with other proteins involved in FA turnover in such a way that they become rate limiting for FA reassembly. This model is consistent with the conflicting observations about the role of TRIP6 during cell migration. The prediction is that cellular responses to exogenous changes in TRIP6 are dependent on endogenous TRIP6 levels, relative to the levels of SV and other interactors in the proposed FA dis/assembly pathway, and on cell type–specific regulation. We suggest that the SV–TRIP6 interaction demarcates a FA subdomain, perhaps a signaling scaffold, which controls FA integrity and/or turnover. One possibility is that the SV–TRIP6 interaction brings other SV-binding proteins into proximity with other TRIP6 partners. In addition to TRIP6, the SV N terminus (SV1-830) binds to the S2 subdomain of nonmuscle myosin II, to F-actin, and to the plasma membrane (; ). SV binding to TRIP6 may recruit myosin II in stress fibers into the vicinity of TRIP6-binding proteins that destabilize FAs. Candidate TRIP6 interactors include CAS/CasL (), c-Src (), Crk (), the LPA receptor (), the tyrosine phosphatase PTPN13/PTP-BL/FAP-1, and the adaptor protein RIL/PDLIM4 (), which is known to increase stress fiber dynamics (). In agreement with this hypothesis, both SV and the LPA–TRIP6–CAS pathway positively regulate ERK signaling (; ), a process implicated in FA turnover (). Alternatively, SV binding to TRIP6 may displace a positive regulator of FA stability through steric hindrance or binding to a shared site on the TRIP6 C terminus. For instance, SV may disrupt the interaction of TRIP6 with endoglin/CD105, which is a transmembrane component of the TGF-β complex that promotes stress fiber formation and the localization of TRIP6 and zyxin to FAs (; ). Finally, we cannot exclude the possibility that SV indirectly influences FA structure through potentiation of TRIP6 effects in the nucleus. Despite its lack of a canonical NLS, TRIP6 can accumulate in the nucleus and modulate transcription (; ). The TRIP6 LIM domains are sufficient for both SV binding () and nuclear transport (). SV contains functional NLS () and, like TRIP6, is implicated in steroid hormone signaling (). Thus, some of the effects reported here might be caused by changes in transcriptional activity induced by an SV–TRIP6 complex. In summary, we show that the myosin II– and actin-binding protein SV regulates FA function through binding to TRIP6, a LIM domain–containing protein associated with signaling scaffolds that control cell motility. This is the first evidence for a myosin II–binding protein at FAs and for an explicit role for TRIP6 in adhesion. The direct binding of SV and TRIP6 suggest several specific, testable hypotheses by which TRIP6-associated scaffolds may control FA function. The SV–TRIP6 interaction may provide a “missing link” for actin-independent attachment of myosin II to the membrane at FAs and insight into molecular mechanisms for FA disassembly and/or recycling. Chemicals were obtained from Sigma-Aldrich, Calbiochem-Novabiochem, Fisher Scientific, or VWR International, Inc., unless otherwise noted. EGFP-tagged SV constructs, purified GST-tagged SV proteins, and Flag-tagged murine TRIP6 were previously described (; ; ). Vectors encoding His-tagged TRIP6 LIM domains or Tctex-1 were generated by excising the insert from the pYESTrp prey vector using the vector KpnI and XhoI sites and ligating in-frame into the corresponding sites of the pET30a bacterial expression vector. The TRIP6 insert encoded amino acids 265–476, and the Tctex-1 insert was full length. Proteins tagged with 6×His were expressed and purified as previously described (). EGFP-SV lacking the TRIP6/Tctex-1–binding site (EGFP-SVΔ343-561) was created by deleting the coding sequence for aa 343–561, converting Ser 343 to a tyrosine. PCR was used to introduce an AgeI site upstream of the codon for Gly 562 in double-stranded DNA (dsDNA) that included the unique endogenous EcoRV restriction site after the codon for Asp 830 (). The PCR product was digested with AgeI and EcoRV and ligated into similarly cut, full-length SV in pBluescript II SK (–) (), replacing the codons for 343–830 with those for 562–830. SVΔ343-561 was then transferred into the mammalian expression vector EGFP-C1 (BD Biosciences and CLONTECH Laboratories, Inc.) with KpnI and XbaI. COS7-2 cells (), an SV40-transformed derivative of monkey kidney epithelial CV-1 cells, and rat A7r5 aorta cells (American Type Culture Collection [ATCC]) were grown in DME with 10% FCS. CV-1 cells (ATCC) were maintained in MEM Alpha (Invitrogen) with 10% FCS. A549 human lung carcinoma cells (ATCC) were grown in Ham's F12K medium, 2 mM L-glutamine, and 10% FCS. Cells were transfected using Effectene Transfection Reagent (QIAGEN). Populations of 100% transfected cells were obtained by FACS 48 h after transfection with plasmids encoding EGFP or EGFP-SV. Methods for indirect immunofluorescence microscopy have been previously described (). In brief, cells transfected for 24 h were fixed with 4% paraformaldehyde in PBS in the presence of 1 mM MgCl and 1 mM EGTA for 10 min and permeabilized with 0.1% Triton X-100 in PBS for 5 (COS7-2) or 10 min (A7r5) before immunostaining. Cells were stained for vinculin (mouse clone hVIN1; 1:200; Sigma-Aldrich), Flag (rabbit polyclonal; 1:100; Sigma-Aldrich), Myc (mouse clone 9E10; 1:1,000; ATCC), Myc (rabbit monoclonal clone 71D10; Cell Signaling Technology, Inc.), TRIP6 (rabbit polyclonal B65; 1:100; ; or mouse anti-TRIP6, clone 16; 1:100; BD Biosciences), zyxin (rabbit polyclonal B71; 1:100; ), and/or SV (affinity-purified rabbit polyclonal H340; 1:100; ; ). Cross-adsorbed secondary antibodies, conjugated with Alexa Fluor 350, 488, 568, 594, or 633 were obtained from Invitrogen. F-actin was visualized with phalloidin conjugated to Texas red or Alexa Fluor 350 (Invitrogen). Slides were analyzed at room temperature with a 100× Plan-NeoFluar oil immersion objective (NA 1.3) on a fluorescence microscope (Axioskop; both Carl Zeiss MicroImaging, Inc.) with a charge-coupled device camera (RETIGA 1300; QImaging Corp.) and OpenLab software (Improvision), or with a 100× Plan Apo oil objective, NA 1.35, on a confocal SP2 microscope (both from Leica) running Leica acquisition software. Each channel of the epifluorescence micrographs was scaled automatically using the Auto Level submenu in Photoshop 7.0 (Adobe). In , the images were treated identically to show antibody specificity. Because of the high background noise in Alexa Fluor 350 images (; , c′, g, and k; and ), background noise had to be subtracted after Auto Level manipulation. This operation may also remove weak Alexa Fluor 350 signals. The number of total and large FAs per cell were counted manually in Photoshop from images obtained at room temperature with a 25× Plan-Neofluar air objective lens (NA 0.8; Carl Zeiss MicroImaging, Inc.) on an Axioskop fluorescence microscope. FA sizes were determined by quantifying pixels in cells stained with antibody against vinculin. Because peripheral focal complexes were too numerous and/or too small to quantify, vinculin foci within 20 pixels (∼1.6 μm) of the cell border were excluded. FAs were defined as vinculin foci ≥ 30 pixels (3.75 μm); large FAs were those with ≥80 pixels (10 μm), and with pixels × mean luminosities ≥ 6,000 arbitrary units. Pixel sizes were calibrated with a micrometer. A ∼3.5-kb sequence encoding the 5′ end of monkey kidney SV cDNA (GenBank/EMBL/DDBJ accession no. ) was cloned by nested PCR using Herculase polymerase blend (Stratagene) and primers corresponding to SV sequences conserved in human, mouse, ferret, and bovine SV. The initial template was oligo-dT–primed, and first-strand cDNA was prepared from COS7 cell RNA using commercial kits (RNAqueous 4-PCR and Message Sensor RT-PCR; Ambion, Inc.). The first round of PCR used the primers MSV-F1 and CRATY-R () and a touch-down protocol, as follows: 92°C, 2 min, 1×; 92°C, 30 s; 55°C, 45 s; 72°C, 8 min; 5×; 92°C, 30 s; 50°C, 45 s; 72°C, 8 min; 5×; 92°C, 30 s; 45°C, 45 s; 72°C, 8 min; 30×; and 72°C, 10 min, 1×. The reaction mixture was diluted 1:100, and 5 μl was reamplified using MSV-F1 and R-KDFW (5′-TGGCYRCCCARRAGCTTCCAGAAGTCTTT-3′) in a second touch-down reaction: 95°C, 2 min, 1×; 92°C, 30 s; 58°C, 45 s; 72°C, 8 min; 5×; 92°C, 30 s; 55°C, 45 s; 72°C, 8 min; 5×; 92°C, 30 s; 50°C, 45 s; 72°C, 8 min; 30×; and 72°C, 10 min, 1×. Two clones encoding 3,524 bp of COS7 SV were sequenced and used to identify five candidate RNAi sequences (; ). Cell–substrate adhesion was assayed by a modification of the method of . In brief, FACS-sorted cells expressing increased levels of EGFP-SV or EGFP, or dsRNA-treated cells (1.0 × 10 cells/ml, 3.7 ml serum-free media), were transferred into the center four wells of 24-well plates. Each well was covered with an 18-mm round coverslip (18 Circle, #1.5; VWR Scientific) coated with 10 μg/ml fibronectin. The plate was inverted and incubated for 60 min at 37°C to permit cell spreading. The plate was inverted again and centrifuged at 300 for 5 min to remove unbound cells. Cells remaining on the coverslips were fixed with 4% paraformaldehyde and counted. Cell contractility was assayed using substrate deformation assays. Flexible substrates were prepared by coating 22 × 22-mm coverslips with 0.1% gelatin (Type A; Sigma-Aldrich) on top of 5 μl silicon (30,000 CS; Dow Corning; William F. Nye Inc.). Sparsely seeded CV1 cells (<100 cells/coverslip) were allowed to attach for 30 min, transfected with plasmids encoding EGFP or EGFP-SV fragments, and grown for an additional 20 h before counting the number of fluorescent cells that were or were not associated with wrinkles, indicating the production of contractile forces (). The living cells were imaged at 37°C in a temperature-controlled Plexiglass chamber on an inverted microscope (DMIRE2; Leica) with a 20× air objective lens (20 DL, NA 0.4; Nikon) and a cooled charge-coupled device camera (RETIGA EXi; QImaging Corp.). Other methods were as described in the Immunofluorescence Microscopy section. A bait plasmid encoding bovine SV residues 171–830 was constructed in pHybLex/Zeo, transformed into the EGY48/pSH18-34 strain of , and used to screen a HeLa cell library in the prey vector pYESTrp (Hybrid Hunter Premade cDNA Library and Two-Hybrid System; Invitrogen) as previously described (). In brief, a library of ∼1.03 × 10 primary transformants was screened approximately four times on selection medium (-ura, -trp, +Z200). Large colonies grown on induction medium (-ura, -trp, -leu, Raff/GAL,+Z200) were picked after 24 or 48 h and tested for β-galactosidase activity on modified induction medium (-ura, -trp, Raff/GAL,+Z200). Out of 506 initial colonies, 172 passed both the leucine autotrophy and β-galactosidase expression tests. Interacting prey vectors were segregated by growth for three generations on –Trp medium, recovered, electroporated into XL1 Blue cells (Stratagene), and sequenced using the pYESTrp forward primer. Sequences with an open reading frame were verified by retransformation into yeast containing the pHybLex/BSV171-830 bait vector. Nonspecific interactions were eliminated by control transformations into yeast with pHybLex/Zeo (empty bait). To localize potential binding sites, confirmed clones were transformed into yeast strains containing either pHybLexA+BSV171-342 or pHybLexA+BSV570-830, and assayed for leucine autotrophy and β-galactosidase activity. Yeast cells (300 μl packed cell vol) expressing V5-tagged bait proteins after a 6-h induction were washed with ice-cold SLB (25 mM Tris-Cl, pH 7.5, 1 mM DTT, 2 mM EDTA, 150 mM NaCl, 30% glycerol), and protease inhibitors (1 μM aprotinin, 2 μM ALL-M, 1 mM benzamidine, 10 μM E64, 1 μM leupeptin, 1 μM pepstatin A, 1 mM PMSF). Yeast were lysed with 400 μl 0.2% Triton X-100 and SLB by vortexing three times at maximum speed for 30 s with 0.5 mm glass beads (Biospec Products Inc., Bartlesville, OK). For GST pull-down assays, COS7-2 cells (10 cells) were lysed with 1 ml lysis buffer (1% NP-40, 25 mM TrisCl, pH 7.5, 50 mM NaCl, 25 mM NaF, 1 mM sodium pervanadate, 50 mM sodium pyrophosphate, and the same protease inhibitors). Lysates were centrifuged at 8,000 for 7 min, and the resulting supernatants were incubated for 3 h at 0–4°C with 25 μl glutathione–Sepharose (GE Healthcare). The beads were washed once with lysis buffer and three times with washing buffer (50 mM sodium phosphate, pH 7.5, 300 mM NaCl, and 10% glycerol and 2-mercaptoethanol). Bound proteins were eluted with 20 mM glutathione in washing buffer, resolved by SDS-PAGE, and transferred to nitrocellulose membranes (Protran BA85; Schleicher & Schuell BioScience, Inc.). Endogenous proteins in mammalian cell lysates were analyzed after extraction and precipitation with 10% TCA (). Blot strips were stained with antibodies against SV (rabbit polyclonal H340; 1:10,000), β-actin (mouse clone AC-74; 1:3,000; Sigma- Aldrich), zyxin (rabbit polyclonal B71; 1:5,000), TRIP6/ZRP-1 (rabbit polyclonal B65; 1:2,000; and mouse monoclonal C.16; 1:2,000), or Tctex-1 (rabbit polyclonal R5205;, 1:25; ). R5205 was provided by S.M. King (University of Connecticut Health Center, Farmington, CT). Purified (100 μg) 6×His-tagged TRIP6 LIM domains or 6×His-tagged, full-length Tctex-1 protein were incubated with ∼0.6 nmol GST-SV343-571 (30 μg) or GST (15 μg) on glutathione–Sepharose (25 μl). Bound proteins were sedimented, washed, eluted, and analyzed by SDS-PAGE and staining with Coomassie brilliant blue or anti-6×His antibody. In competition assays, 6×His-TRIP6 LIM domains (100 μg; 5 nmol) were incubated with GST-SV343-571 (30 μg; 0.6 nmol) on glutathione–Sepharose (25 μl) in the presence of increasing amounts of premixed 6×His-Tctex-1: 0, 10 (0.8 nmol), 20 (1.5 nmol), and 100 μg (7.7 nmol). Fig. S1 shows EGFP-SV overlap with vinculin-stained FAs at the basal surface of a COS-7 cell. Fig. S2 shows that the RY/AA mutant of EGFP-SV342-571 exhibits a less deleterious phenotype than EGFP-SV342-571 on stress fibers and large FAs. Fig. S3 shows the specificities of the antibodies used in this study on COS7 and A7r5 cells. Fig. S4 shows that TRIP6 remains at FAs after SV knockdown. Online supplemental material is available at .

Endocytosis is essential for a variety of cellular functions, including the internalization of nutrients and communication among cells, or between cells and their environment. Internalized molecules must be precisely sorted to their final cellular destinations to fulfill their specific function. Distinct endocytic pathways have been described to date, including clathrin-dependent endocytosis and caveolae-mediated uptake, which remain the two best-characterized mechanisms of internalization (). Neurons have adapted their endocytic pathways to better adjust to their specific requirements. Thus, synaptic vesicle (SV) recycling is the predominant form of neuronal endocytosis at the presynaptic terminal, whereby the fast fusion of neurotransmitter-containing vesicles is coordinated with an efficient mechanism of membrane recovery, which involves clathrin (for review see ). In neurons, clathrin-independent routes have also been documented, although the physiological relevance of endocytosis via caveolae has been questioned in these cells because several of the caveolin isoforms found in other tissues are not detectable. A special feature of motor neurons (MNs) is that their presynaptic terminal, which forms the neuromuscular junction (NMJ), is located in the periphery, whereas the soma is located in the central nervous system. Therefore, any material that enters the MNs at the NMJ and is transported toward the cell body, such as neurotrophins, crosses the blood–brain barrier. To gain more insights into the endocytic events at the NMJ, we followed the endocytosis of tetanus neurotoxin (TeNT). TeNT is a neurospecific toxin that binds to MNs at the NMJ, where it is internalized and undergoes axonal retrograde transport to the cell body. It is then secreted and taken up by adjacent inhibitory interneurons, where it blocks neurotransmitter release by cleaving VAMP/synaptobrevin, which is a synaptic SNARE (). The TeNT receptor complex has been shown to comprise lipids and proteins (). The polysialogangliosides GD1b and GT1b (; ), as well as one or more glycosylphosphatidylinositol (GPI)-anchored proteins (; ) are required for toxin binding to the neuronal surface. TeNT is associated with detergent-resistant membranes (DRMs), which are enriched in cholesterol and GPI-anchored proteins (), and its uptake is sensitive to cholesterol depletion (). Furthermore, pretreatment of neurons with phosphatidylinositol-specific phospholipase C to cleave GPI-anchored proteins from their lipid anchor prevents TeNT intoxication (). Altogether, these findings suggest that TeNT follows a polysialoganglioside- and DRM-dependent route for its internalization in neuronal cells. However, in previous EM studies on spinal cord neurons, gold-labeled TeNT was detected in surface pits resembling clathrin-coated invaginations, as well as in coated and uncoated vesicles (; ). Because clathrin-mediated internalization and the endocytosis of proteins associated with DRMs have been largely viewed as mutually exclusive (), the association of TeNT with clathrin coats was unexpected. To resolve this apparent paradox, we have studied the internalization machinery responsible for the uptake of TeNT into MNs using a C-terminal binding fragment of TeNT (). In this study, we show that TeNT H endocytosis in MNs is independent of SV recycling, the major route of internalization at the presynaptic terminal, and demonstrate that although TeNT H binds to DRMs on the MN surface, it uses a clathrin-mediated pathway for its entry. This specialized clathrin- and AP-2–dependent uptake mechanism does not require the endocytic adaptor protein epsin1, further indicating that specific adaptors play important functions in initial sorting events during endocytosis. Although previous studies implied that TeNT does not enter the NMJ via SV endocytosis (), some studies suggested that the toxin can take this route in brain-derived neurons, such as hippocampal neurons () and that it may enter SV-like vesicles in spinal cord neurons in culture (). In light of these findings, we assessed whether SV exo/endocytosis is the physiological route of TeNT entry in MNs. Several lines of evidence indicate that this is not the case. First, we tested the colocalization of internalized TeNT H and the SV protein VAMP-2. MNs were incubated with Alexa Fluor 555–TeNT H at 37°C, fixed, and stained for VAMP-2. Under resting conditions, colocalization in the cell body, neurites, or synaptic contacts was very limited (). Moreover, stimulation of SV exo/endocytosis by depolarization did not increase the extent of colocalization (). To further investigate the endocytic pathway of TeNT, we used a biotinylated, thiol-cleavable form of TeNT H (b-TeNT H). By exposing intact neurons to cell-impermeable reducing reagents, such as 2-mercaptoethane sulfonic acid (MESNA; ), biotin can be cleaved off surface-bound TeNT H, while the internalized b-TeNT H is protected. Staining for the remaining biotin allows the internalized TeNT H to be detected selectively over the surface-bound TeNT H even in thin structures such as axons. Biotinylation does not change the binding and internalization properties of TeNT H because preincubation with a 100-fold excess of unlabeled toxin completely abolished the binding of b-TeNT H to MNs (Fig. S1 A, available at ). Furthermore, under internalization conditions, b-TeNT H colocalized with Alexa Fluor 555–TeNT H (Fig. S1 B). Importantly, biotin could be cleaved off the surface-bound b-TeNT H by treatment with MESNA on ice. In contrast, in cells incubated at 37°C the label remained cell-associated, showing that b-TeNT H had been taken up in the soma and neurites (Fig. S1 C). For an independent assessment of a role for SV recycling in TeNT H uptake, we then preincubated MNs with botulinum neurotoxin (BoNT)/A and D for 22 h to block SV exocytosis, and, thus, recycling (). The samples were then incubated with b-TeNT H for 30 min at 37°C, treated with MESNA on ice, fixed, and stained for VAMP-2 and SNAP-25, as well as for biotin, to visualize internalized TeNT H. The complete cleavage of SNAP-25 and VAMP-2 by BoNT/A and D was confirmed by Western blotting () and by indirect immunofluorescence (), indicating that SV exo/endocytosis was blocked under these conditions. However, TeNT H internalization was not affected in intoxicated MNs () compared with untreated cells (). We next used the b-TeNT H to test the requirement for dynamin in this process. Dynamin is a GTPase essential for clathrin- and caveolin-mediated endocytosis, as well as for several other endocytic and vesicle-trafficking events. Incubation of MNs with the cell-permeable peptide P4, which inhibits dynamin function (), but not with the scrambled peptide P4S, significantly reduced uptake of TeNT H, whereas its overall binding to the neuronal surface was not affected (). These results were confirmed by overexpressing the well-characterized dynamin mutant K44A, which is defective in GTP binding and hydrolysis and restrains invaginated pits from pinching off (; ). We used microinjection to introduce foreign DNA into MNs because lipid-based transfection reagents abolished axonal transport in MNs. In contrast, microinjection of plasmids driving the overexpression of control proteins had no effect on cell viability and retrograde transport (). Expression of dynamin2 significantly reduced TeNT H endocytosis at the level of both the soma and neurites (), without affecting its binding to the MN surface (). These results indicate that dynamin is a central player in TeNT H internalization and rule out differences in the mechanism of uptake of TeNT H between axons and the soma. This is important because, topologically, only the axon is physiologically relevant for TeNT H uptake and retrograde transport. A total block of TeNT H endocytosis by the expression of dynamin2 was seen in >95% of the cells (). However, dextran internalization still took place under these conditions () or upon P4 treatment (), confirming that MNs were viable and still capable of endocytosis via dynamin-independent pathways. We chose dextran as a control marker for internalization because we found that cholera toxin subunit B (CTB), which is another widely used marker for clathrin-independent endocytosis, is internalized in a strictly dynamin-dependent fashion in MNs (unpublished data). This was surprising because CTB has been shown to use a dynamin-independent entry pathway in other cell types, such as HeLa and mouse embryonic fibroblasts (; ; ; ). In previous studies, TeNT was found in coated pits, as well as in coated and uncoated vesicles in spinal cord neurons (; ). Furthermore, TeNT is taken up by a clathrin-independent route in nonneuronal cells (). Therefore, we decided to investigate whether TeNT internalization is strictly clathrin-mediated in cultured MNs. First, we assessed the colocalization of clathrin and TeNT H at the light microscopy level in MNs microinjected with GFP–clathrin light chain (CLC). Upon incubation with TeNT H at 37°C, we could see only a partial overlap between GFP-CLC and TeNT H, more readily in the cell body than in the axon (). To verify this colocalization at a fine structural level, we used TeNT H coupled with HRP. Just like b-TeNT H, binding of this fusion protein to MNs was inhibited by preincubation with an excess of unlabeled TeNT H and, upon internalization, Alexa Fluor 488–TeNT H and HRP–TeNT H showed extensive colocalization (Fig. S2, A and B, available at ). The relationship between internalized TeNT H and clathrin-coated structures was then investigated at early stages of the internalization process. To this end, we synchronized the uptake of HRP–TeNT H by preincubating the MNs at 4°C and, subsequently, warming to 12°C. This low-temperature treatment allows the plasma membrane to invaginate but inhibits vesicle–pinching off (unpublished data). The effectiveness of this protocol was confirmed by incubating MNs with either b-TeNT H or HRP–TeNT H at 12°C and, subsequently, treating cells with MESNA or performing the DAB reaction in the presence of ascorbic acid, which is a membrane-impermeable inhibitor of the HRP staining (). Upon MESNA treatment, no biotin was detectable by immunofluorescence (Fig. S2 C). Similarly, when the DAB reaction was performed in the presence of ascorbic acid on cells incubated at 12°C, we observed immunolabeling of clathrin, but no DAB staining (Fig. S2 D). These findings indicated that the coated pits remained open to the external medium at 12°C and confirmed the suitability of this temperature block for the study of the initial stages of endocytosis. After 12°C incubations, the electron-dense DAB reaction product generated by HRP–TeNT H was readily observed in coated pits on the plasma membrane of soma, dendrites, and axons (). The nature of these coated domains was confirmed by immunogold staining with clathrin heavy chain (CHC) antibodies, which labeled pits containing HRP–TeNT H (). The DAB reaction product found in both shallow pits and in deeper invaginations was closely associated with clathrin lattice components (). After the 12°C block, we allowed MNs to internalize TeNT H at 18°C to monitor its intracellular axonal transport. Fine structural analysis suggests that progression along the endocytic pathway was slowed down at this temperature and lead to an increase of early endosomal carriers. At 18°C, HRP–TeNT H was found in coated vesicles () and other vesicular and tubular structures within the axon. Gold immunolabeling of coated pits in axons was not easily discerned in thin EM sections, as permeabilization of these structures and access to the antigen appeared to be impaired because of the highly packed cytoskeleton in these areas. Therefore, we decided to prepare whole-cell mounts by extracting neurons with Triton X-100 before fixation, thereby improving the antigen availability and providing a better overview over the total population of TeNT H–positive membranes. To stabilize HRP–TeNT H–containing pits and protect them from solubilization, DAB cross-linking was performed before detergent extraction. At 12°C, the vast majority of DAB-positive structures located along the axons () and on the cell body () were clathrin positive. Polysialogangliosides of the b-series, including GD1b and its analogues GT1b and GQ1b, have previously been described as essential components of the TeNT receptor complex (). However, these lipids, like other residents of sphingolipid-rich microdomains, are thought not to enter clathrin-coated pits (CCPs; ). Therefore, we asked where GD1b localizes on the neuronal surface in relation to TeNT H and clathrin-coated invaginations. By light microscopy, we were able to confirm colocalization of TeNT H and GD1b on the neuronal surface by using a specific anti- GD1b antibody (MOG-1; ). Furthermore, preincubation of MNs with MOG-1 inhibited the binding of TeNT H in a dose-dependent manner (Fig. S3 A, available at ), confirming that GD1b is an essential component of the TeNT receptor complex. To obtain a higher resolution view of the association between TeNT H and polysialogangliosides, we incubated MNs with HRP–TeNT H in the presence of noncompeting concentrations of MOG-1 at 12°C (Fig. S3 B) and analyzed the samples by EM. As previously described for other components of lipid microdomains, we found clusters of gold-labeled GD1b on the cell surface, often in close proximity to the DAB precipitate generated by HRP–TeNT H (). In addition, the DAB precipitate was frequently associated with coated structures (, arrows and arrowheads). However, we were unable to detect GD1b in CCPs containing the DAB cross-linking product. Instead, gold-labeled GD1b was frequently found at the edge of HRP–TeNT H–positive pits (). Furthermore, internalized gold particles were very rarely detected upon incubation at 37°C, suggesting that GD1b remains surface-bound (). Under the same conditions, TeNT H was identifiable in many vesicles and tubules, all of which were free of GD1b gold label (). To confirm that the DAB precipitate did not conceal any gold particles in internalized structures, we incubated MNs with HRP–TeNT H together with gold-conjugated TeNT H. Under these conditions, gold label was clearly visible in all HRP-positive structures (Fig. S2 E). We next examined whether CCP formation is required for TeNT H internalization by affecting specific components of the clathrin-dependent endocytic machinery. The blockade of transferrin uptake, which strictly relies on a clathrin-mediated pathway (), was taken as a positive control, whereas the CTB entry, which occurs via clathrin-independent routes in several cell types (; ; ), allowed us to monitor the specificity of the inhibition. In control MNs, both CTB and b-TeNT H readily entered neurons in soma and neurites (). Expression of the phosphorylation-deficient mutant of the AP-2 subunit μ2 (μ2), which is incorporated into AP-2 complexes but cannot be phosphorylated at Thr156, thus, impairing AP-2–dependent clathrin-mediated endocytosis (), blocked the uptake of TeNT H (), as well as transferrin internalization (not depicted). In contrast, CTB entry in μ2-expressing MNs is barely altered (), suggesting that, in contrast to that observed in hippocampal neurons (), its mechanism of uptake in MNs is mainly clathrin independent. As expected, binding of TeNT H to the cell surface was not affected by expression of μ2 (not depicted). Similar results were obtained by expressing a truncation mutant of the accessory protein AP180 (AP180-C). This mutant inhibits uptake of EGF and transferrin in COS-7 cells (). Expression of AP180-C inhibited both TeNT H () and transferrin endocytosis (not depicted), whereas CTB internalization was not visibly affected (). In contrast to AP-2 and AP180 dominant-negative constructs, expression of a mutant version of the adaptor protein epsin1 had no significant effect on TeNT H internalization (). This epsin1 mutant is unable to bind phosphatidylinositol-4,5-bisphosphate (PtdInsP) and blocks transferrin uptake in COS-7 cells (). As expected, transferrin endocytosis was completely inhibited in epsin1-expressing MNs (). These findings were confirmed by an independent EM analysis, where expression of this epsin1 mutant abolished the uptake of transferrin-HRP (not depicted), leaving the internalization of HRP–TeNT H unaffected (, arrows). In addition, epsin1-expressing cells showed no obvious morphological alterations and displayed occasional CCPs and clathrin-coated vesicles (CCVs; , arrowheads and insets). These findings, together with previous works reporting the existence of AP-2–independent routes (; ; ; ; ), suggest that different subsets of adaptors proteins functionally define distinct clathrin-dependent pathways. To investigate the spatial relationships of TeNT H, transferrin, and epsin1 during the early stages of endocytosis, we analyzed the distribution of epsin1 by immuno-EM in MNs incubated with gold-conjugated TeNT H and transferrin-HRP at 12 or 20°C. Under these conditions, we could detect TeNT H in clathrin-coated structures either containing or devoid of transferrin, as well as transferrin-containing pits and vesicles devoid of TeNT H (Fig. S4 A, available at ). Quantitative analysis of all TeNT H–containing structures revealed that 54% of these were devoid of transferrin. Interestingly, less than half of these TeNT H single-positive structures labeled for epsin1, whereas two-thirds of the TeNT H and transferrin dual-positive vesicles and pits were also positive for epsin1 (Fig. S4 B). Therefore, epsin1 accumulates preferentially, but not exclusively, on transferrin-containing structures. It should be considered, however, that permeabilization with digitonin is likely to affect the stability of HRP-negative membranes (TeNT H ± epsin1). Therefore, these structures are likely to be underestimated, as observed by independent experiments in which the occurrence of transferrin-HRP and HRP– TeNT H containing CCP and CCV has been quantified (unpublished data). The effects of the disruption of different components of the clathrin-dependent machinery on TeNT H uptake are evident in the quantitative analysis provided in . Although TeNT H endocytosis into MNs can be blocked by disruption of the clathrin adaptors AP-2 and AP180, it does not require the adaptor protein epsin1. Given that epsin1 is targeting ubiquitinated receptors to the late endosomal/lysosomal pathway (), the independence of TeNT H internalization from epsin1 function is in agreement with the finding that TeNT H escapes targeting to acidic compartments and degradation in MNs (; ). An open question in the field of membrane trafficking is how distinct extracellular ligands following internalization via the same endocytic pathway (i.e., CCPs, caveolae), are sorted in early endosomes to their different intracellular destinations. In neurons, this process is crucial for the targeting of growth factors and their receptor complexes to short- and long-range trafficking routes, ultimately leading to diverse and often opposite biological functions. This is exemplified by the action of nerve growth factor, which has been shown to alter growth cone dynamics by local signaling, while it acts as a survival factor following axonal retrograde transport and transcriptional activation in the nucleus (). The fine balance between these two processes is fundamental for our understanding of differentiation, synaptogenesis, and plasticity in the nervous system. TeNT H was chosen as a tool to monitor endocytosis in MNs based on its high binding affinity to neuronal membranes () and its ability to enter the same axonal transport compartment used by nerve growth factor () and the neurotrophin receptor p75 (unpublished data). Previous work highlighted the association of TeNT with both coated and uncoated structures, whereas at later time points it was found in many endocytic organelles, including coated and uncoated vesicles, tubules, and SV-like profiles (; ; ; ; ; ). However, a functional analysis assessing the role of the various internalization routes has yet to be made. We show that clathrin-dependent internalization plays a nonredundant role in the uptake of TeNT in MNs. At 37°C, CCPs and CCVs that are positive for TeNT are rarely found, especially along axons. To explore if the localization to coated structures is representative for the entire TeNT H pool, we lowered the temperature to inhibit fission, thus, trapping forming pits on the cell surface. A striking colabeling of TeNT H and clathrin was seen under these conditions. Importantly, whole-mount EM analysis revealed that the HRP–TeNT H–containing areas along the axon were positive for clathrin, indicating that TeNT H does enter MNs via CCPs. In this light, the uncoated structures containing TeNT H observed previously (; ; ) may represent vesicles from which the clathrin lattice was rapidly removed (; ). Several virulence factors, such as cholera and Shiga toxins, are taken up by clathrin-dependent and -independent routes (; ; ), which may display different extents of cross talk and redundancy in various cell types (). To ensure that clathrin-dependent endocytosis is the main, nonredundant route of TeNT H internalization and identify the endocytic machinery responsible for its uptake, we used dominant-negative constructs interfering with distinct steps of coat recruitment and/or pinching off from the plasma membrane. Impairing dynamin function led to a block in TeNT H uptake, showing that this GTPase is necessary for TeNT H endocytosis. In addition, disruption of the clathrin machinery by mutants of AP180 nearly completely abolished TeNT H internalization. Furthermore, the uptake of TeNT H, as well as of transferrin in MNs, are strictly AP-2–dependent, confirming previous findings obtained with transferrin in other cell types (; ; ). Thus, a functional clathrin machinery is strictly required for TeNT H endocytosis (). In contrast, expression of a dominant-negative epsin1 mutant did not interfere with TeNT H internalization. Epsin1 is an endocytic adaptor for clathrin-dependent and -independent internalization that binds to ubiquitinated receptors via its ubiquitin interacting motifs (; ; ). It has a modular structure comprising binding sites for PtdInsP, CHC, AP-2, and other accessory factors for clathrin-mediated endocytosis. The mutant used in our study is deficient in PtdInsP binding () and was reported to block endocytosis by sequestering AP-2. However, we observed opposite effects with the AP-2 μ2 and the epsin1 mutants on TeNT H uptake in MNs, suggesting that their inhibitory activity may not completely overlap. In contrast with that reported in COS-7 cells, we did not detect aggregation of AP-2 in MNs or glial cells expressing epsin1 (Fig. S5, available at ). In these conditions, uptake of transferrin into MNs was blocked by expression of this epsin1 mutant, indicating that epsin1 displays a dominant-negative effect in these cells that is likely to be independent of AP-2 sequestration. Epsin1 function overlaps with that of the homologues Eps15 or Eps15R, as shown by the limited effect of single and double knockdown on transferrin and EGF internalization in HeLa cells (). In contrast, transferrin uptake was completely blocked by expression of epsin1, demonstrating that epsin1 has a nonredundant function in a subset of clathrin-mediated, AP-2–dependent endocytic events in MNs (). Moreover, the block of transferrin internalization by epsin1 suggests that epsin1 acts at an early step of the uptake process, before pit closure, implying that sorting in the endocytic pathway initiates at the plasma membrane. In this regard, we found epsin1 to be associated preferentially, but not exclusively, with transferrin-containing CCPs and CCVs. In agreement with this view, the sorting of endocytic cargoes internalized via clathrin-mediated uptake, such as low- density lipoprotein and influenza virus, to distinct population of endosomes has been shown to begin at the level of CCPs (). In spite of its entry into MNs via a clathrin-dependent mechanism, TeNT H binds to DRMs, and its uptake can be blocked by cholesterol sequestration and cleavage of GPI-anchored proteins (; ). Therefore, TeNT H may use endocytic mechanisms that have, until recently, been viewed as mutually exclusive. Some components of DRMs, such as GM1, are excluded from CCPs (); others do not require a functional clathrin machinery or dynamin for their internalization (; ; ). However, evidence suggesting an overlap between these two endocytic routes has been recently reported in the case of anthrax toxin (), chemokine receptor 5 (; ), and prion protein (). In light of these findings, it is clearly of interest to determine if TeNT H, on recruitment to DRMs, remains within its lipid environment during internalization or is transferred to a modified receptor complex before sorting into CCP. To address this issue, we examined the spatial relationship between TeNT H, coated pits, and GD1b. Although we readily observed TeNT H and GD1b clustered together at the neuronal surface, we were unable to detect GD1b within CCP. Interestingly, GD1b-associated immunogold was frequently found at the edge of TeNT H–positive pits. These observations suggest that even though GD1b and other b-series gangliosides are essential for TeNT binding to the neuronal surface and toxicity (), TeNT H is no longer in complex with the bulk of GD1b during internalization. This hypothesis is strengthened by the lack of internalization of the anti-GD1b antibody over the time intervals used in our experiments (unpublished data) and the slow kinetics of retrograde transport of gangliosides in vivo (). In this model, TeNT H is initially captured by GD1b microdomains before being targeted to CCP (). This lateral sorting, which could require the integrity of lipid rafts (), might be mediated by glycosylated proteins binding the carbohydrate-binding pockets of TeNT H that were previously occupied by GD1b or other b-series gangliosides (). CTB instead binds to GM1-enriched lipid rafts on the plasma membrane, leading to its internalization via a clathrin-independent, dynamin-dependent pathway in MNs and its late appearance in axonal carriers distinct from those containing TeNT H (). The strength and specificity of the binding to gangliosides are therefore primary determinants of the kinetics of internalization and endocytic sorting of TeNT H and CTB. In conclusion, we have shown that TeNT H internalization occurs via a specialized clathrin-dependent pathway, which is distinct from SV endocytosis and is preceded by a lateral sorting from its lipid raft–associated ligand GD1b. As for transferrin, TeNT H uptake relies on a nonredundant function of AP-2. However, transferrin endocytosis is dependent on epsin1, whereas TeNT H uptake is not, and may result in targeting of TeNT to neutral long-range transport compartments (; ). These findings indicate that clathrin adaptors are assembled in a cargo-selective manner to drive the internalization of plasma membrane proteins and their ligands (; ). This process has, in turn, the power to generate different populations of early endosomes, which have different targeting determinants and fates within the cell. Chemicals were obtained from Sigma-Aldrich, BDH Chemicals Ltd., or Invitrogen, unless otherwise stated. Sulfo-NHS-SS-biotin and EZ-link–activated maleimide-HRP were purchased from Pierce Chemical Co. Antibodies 9E10, X22, and 12CA5 were obtained from the Cancer Research UK antibody facility, antibody 69.1 was purchased from Synaptic Systems, and the antibody against the C-terminus of SNAP-25 was a gift from O. Rossetto (University of Padova, Padova, Italy). The epsin1 antibody was a gift from L. Traub (University of Pittsburgh, Pittsburgh, PA). The IgG3 mouse monoclonal antibody MOG1 reacts with 8 M), and GD2, but not with GT1b, GQ1b, or GD3 (Boffey et–GD1b (Kd = 10 al., 2005). Plasmids encoding dynamin, epsin1, and AP180 C-terminal mutants were a gift from H. McMahon (Laboratory of Molecular Biology, Cambridge, UK), AP-2 μ2 was a gift from E. Smythe, and GFP-CLC was a gift from L. Greene (National Institutes of Health, Bethesda, MD). TeNT HC was labeled with Alexa Fluor–maleimides () or biotin, according to the manufacturers' instructions, followed by dialysis against PBS. To couple TeNT H to HRP, 10 nmol of cysteine-tagged TeNT H were incubated with 5 mM EDTA and 6.5 mg EZ-link–activated maleimide-HRP in PBS overnight at 4°C. The conjugate was purified first on ConA–Sepharose (GE Healthcare) and eluted with 0.25 M α-methylmannoside in 10 mM sodium phosphate, pH 7.2. HRP–TeNT H was bound to NiNTA-agarose (QIAGEN) and eluted in 20 mM Hepes-NaOH, pH 7.4, 150 mM NaCl, and 500 mM imidazole. Samples containing HRP–TeNT H were pooled and dialyzed against PBS. To double label TeNT H with an Alexa Fluor dye and HRP, fluorophore labeling was performed first, according to the manufacturer's instructions, using half of the recommended amount of dye and without the addition of glutathione to stop the reaction. Alexa Fluor–labeled TeNT H was dialyzed against PBS to remove the excess dye before HRP-conjugation. MN cultures were prepared and maintained in culture as previously described (). Cells were injected with 0.05 mg/ml of plasmid between 4 and 7 d in vitro. In cases of microinjection of multiple plasmids (e.g., the pTRE-μ2 T156A plasmid that requires a transactivator ptTA for expression; CLONTECH Laboratories, Inc.), 0.04 mg/ml of each construct were mixed before injection. MNs were incubated with 15–20 nM TeNT H and then either biotinylated or Alexa Fluor–labeled for 30 min at 37°C. In selected experiments, 20 μg/ml Alexa Fluor 594–transferrin, 10 ng/ml Alexa Fluor 555–CTB, or 0.2 mg/ml tetramethylrhodamine dextran (mol wt 3,000) were mixed with TeNT H before addition to the cells. 60 mM KCl was added to the medium just before addition of the ligands to test the effects of depolarization. In experiments where MNs were pretreated with P4 or P4-scrambled peptide (), 50 μM of peptide was added to the medium at 37°C for 2 h before incubation with 20 nM Alexa Fluor 488–TeNT H and 0.2 mg/ml tetramethylrhodamine dextran. To test the effect of SV exo/endocytosis on TeNT H uptake, MNs were seeded on 13-mm coverslips. At 6 d in vitro, MNs in two wells were incubated with 15 nM BoNT/A and 2 nM BoNT/D for 22 h at 37°C, while control wells were left untreated. Coverslips from treated and untreated wells were then transferred into a new dish and incubated with 20 nM b-TeNT H for a further 30 min at 37°C before treatment with MESNA on ice, fixing, and processing as described in the following paragraph. The remaining cells from each well were washed in PBS, scraped, centrifuged, and then resuspended in SDS sample buffer. Proteins were analyzed by Western blotting using standard procedures. Antibodies were used as follows: anti–VAMP-2 (69.1), 1:500; anti–SNAP-25, 1:1,000; anti-actin (AC-40), 1:1,000; and HRP-conjugated secondary antibodies (GE Healthcare), 1:1,000. Cells were fixed in 4% PFA and 20% sucrose in PBS for 15 min at room temperature, permeabilized with 0.1% Triton X-100 in PBS for 5 min, blocked in 2% BSA, 10% normal goat serum, and 0.25% fish skin gelatin in PBS for 30 min, and then incubated with the relevant antibodies (anti–VAMP-2 [69.1], 1:300; anti–SNAP-25, 1:300 []; anti-HA [12CA5], 1:1,000; anti-Myc [9E10], 1:250; secondary antibodies, 1:500; or streptavidin 1:500) for 30 min in blocking solution. Cells were mounted in Mowiol-488 and imaged using a LSM 510 laser scanning confocal microscope equipped with a 63×, 1.4 NA, Plan Apochromat oil-immersion objective (both Carl Zeiss MicroImaging, Inc.). Images were processed using LSM 510 software. Images showing GFP-CLC and TeNT H colocalization were taken on living cells at 37°C using a laser scanning confocal microscope (IX70; Olympus) equipped with a 60×, 1.2 W, UPlan Apochromat oil-immersion objective and an environmental chamber. Images were captured using the Ultraview Imaging Suite Version 5.5 software (Perkin Elmer) and processed using AQM Advance 6 Kinetic Acquisition Manager software (Kinetic Imaging). MNs grown on glass coverslips were incubated with 80 nM HRP–TeNT H and/or with 10 μg/ml MOG-1 antibody in serum-free neurobasal medium for 45 min at 4°C. Cells were then washed and chased in prewarmed medium at different temperatures for the indicated time. When appropriate, cells were incubated with a 10-nm gold-conjugated anti–mouse antibody (British Biocell International) on ice for additional 30 min and washed before chase in medium alone. Cells were then fixed with 2% PFA and 1.5% glutaraldehyde in 100 mM sodium cacodylate, pH 7.5, for 15 min and treated with DAB (0.75 mg/ml in 50 mM Tris-HCl, pH 7.4) and 0.02% HO to cross-link HRP. Samples were postfixed and embedded in Epon as previously described (). MNs were then sectioned en face, and 60-nm sections stained with lead citrate were viewed in an electron microscope (CM12; Philips). For whole mounts, MNs were treated as in the pervious paragraph, but instead of being permeabilized with digitonin, they were extracted with 1% Triton X-100 in PBS containing 1 mM MgCl and 0.1 mM CaCl for 10 min at 5°C. After gold labeling, samples were fixed in 4% glutaraldehyde and 1% osmium, dehydrated, and critical-point dried before being prepared for carbon replicas (). For EM of microinjected cells, MNs were seeded on CELLocate glass-gridded coverslips (Eppendorf). A plasmid encoding ssHRP-KDEL () was used as an injection marker. 26 h later, cells were treated with HRP–TeNT H and then with DAB, as described in the previous paragraphs. Alternatively, coverslips were incubated with 20 μg/ml human transferrin-HRP () after a 15-min preincubation at 37°C for in serum-free neurobasal medium. After fixation and embedding in Epon, ultrathin sections were cut from the grid area containing the microinjected cells and imaged by EM. For immunolabeling, samples were incubated with DAB in 50 mM Tris-HCl, 110 mM NaCl, pH 7.4, or with ascorbic acid buffer (20 mM Hepes-NaOH, 70 mM NaCl, and 50 mM ascorbic acid, pH 7.0) at 5°C for 30 min after treatment with HRP–TeNT H and chased in medium. Cells were then permeabilized with 40 ng/ml digitonin in permeabilization buffer (25 mM Hepes-KOH, 38 mM aspartate, 38 mM glutamate, 38 mM gluconate, 2.5 mM MgCl, and 2 mM EGTA, pH 7.2), fixed in 2% PFA, quenched with 50 mM glycine, and blocked with 1% BSA before treatment with primary antibody in PBS containing 1% BSA for 60 min at room temperature. To enhance the signal, intermediate species-specific antibodies were used. MNs were washed and incubated with an appropriate 10-nm gold-labeled secondary antibody (British Biocell International) in 2% BSA and 2% FCS in PBS for 45 min at room temperature. After washing, cells were fixed and processed for conventional EM. In double- and triple-label experiments, MNs were incubated with 80 nM HRP–TeNT H or with 20 μg/ml transferrin-HRP together with TeNT H directly conjugated to 10-nm gold particles (as described by ) in serum-free neurobasal medium for 45 min at 4°C. Cells were washed and shifted to 12 or 20°C before incubation with DAB/HO. Fig. S1 shows biotinylated TeNT H as a probe to study membrane trafficking in MNs. Fig. S2 shows characterization of HRP–TeNT H. Fig. S3 shows that binding of TeNT H to MNs can be competed by preincubation with a specific anti-GD1b antibody. Fig. S4 shows distribution of TeNT H, epsin1, and transferrin in the endocytic pathway of MNs. Fig. S5 shows that overexpression of epsin1 does not lead to AP-2 aggregation in spinal cord cells. Online supplemental material is available at .

The tendency of large complexes to congregate drives the organization of the genome and other cellular structures, according to Peter Cook (University of Oxford, UK). In a crowded, enclosed space (such as the cell), the aggregation of two large spheres increases the system's entropy by giving lots of little molecules more room to move around. In physics, this effect is known as depletion attraction. Using mathematical modeling, Cook and colleagues found that this attraction explains the looping of DNA found naturally in cells. The model is based on transcription and replication complexes that are spaced along DNA like beads on a string. Measurements of the attraction between the “beads” suggest that the entropy gained by their clustering is enough to cover the energy costs of looping the DNA string between them. In vivo, the effects are seen as the clustering of active genes and the looping of highly transcribed genomes. As predicted, in slow-growing cells, which have little replication and transcription, polymerases did not cluster. According to Cook, other large cellular structures are always trying to congregate. The cell must therefore fight the depletion attraction. “Otherwise, everything would become one big blob,” he says. This entropic effect might explain why the size of stable complexes is limited to that of a ribosome. It also explains why larger structures, such as the cytoskeleton, are made by loosely associated subunits that constantly turn over. “The cell is spending energy,” says Cook, “to maintain the structure.” Energy is similarly expended to keep pools of vesicles from fusing into one by turning over their surfaces. Cook explains, “if the vesicles are continually being broken up into smaller pieces that quickly diffuse away, the attraction becomes smaller.” Reference: Transcriptional control elements are like zip codes for genes, based on new findings presented by Barbara Sollner-Webb (Johns Hopkins University, Baltimore, MD). Thousands of copies of transfected plasmids find a common area to inhabit if they share the same zip code, sorted away from plasmids with different zip codes. “That endogenous genes have specific locations is not so surprising,” said Sollner-Webb. “But why are ribosomal genes [for instance] inside the nucleolus? What gets them there?” Her group's findings, although using exogenous DNA, may offer the best explanation so far. During transient transfection experiments, the group noticed that tens of thousands of copies of a plasmid went to the same location in the mammalian nucleus. The locale was promoter specific. Plasmids containing RNA polymerase (RNAP) I promoter sequences were found only in nucleoli, whereas those with RNAP III promoter sequences formed perinucleolar foci. RNAP II promoters took the plasmid DNA to nucleoplasmic foci, but different promoter sequences resulted in different foci. The zip codes seem to be read by transcription factors. As even untranscribed plasmids localized in this manner, polymerase activity is not necessary. The perinucleolar localization required just an 18- or 26-base-pair region that binds to TFIIIA or TFIIIC, respectively. These RNAP III factors are found throughout the nucleus, but Sollner-Webb hypothesizes that the subset in these perinuclear foci is special, either because it is bound to DNA or because it is modified (by phosphorylation, for example) to favor DNA binding. Most RNAP I transcription factors are already concentrated in the nucleoli. RNAP I transcription factors are probably abundant enough, based on published estimates, for each plasmid to have its own copy. Life might be more complicated for less abundant RNAP II factors. The modularity of RNAP II promoters suggests that multiple sequence elements and transcription factors might be necessary. The RNA interference (RNAi) machinery holds together copies of silencing elements from multiple genes, as described by Giacomo Cavalli (CNRS, Montpellier, France). This group hug reinforces developmental gene silencing. The silenced state of a developmentally regulated homeotic gene is maintained by chromatin-modifying Polycomb group (PcG) proteins. PcG response elements (PREs), the DNA sequences that recruit PcG proteins, are self locating: PRE-containing sequences from multiple genes travel long distances to cluster with each other, and this association enhances the transcriptional silencing. The silencing that accompanies heterochromatin formation in yeast requires RNAi components. Cavalli's group has found that PcG-mediated euchromatin silencing also depends on RNAi proteins. Silencing of multiple PRE-containing genes was relieved in fly mutants lacking RNAi proteins such as AGO1 and Dicer-2. PcG proteins were still recruited to their targets, but clustering was lost. The glue for the PRE clusters seems to be small RNAs. Several species of small RNAs were found that matched the PRE region, and their production depended on the RNAi machinery. Clustering correlates with the presence of small RNAs, but Cavalli has not yet ruled out protein–protein interactions as the cause. The physical properties of a nanocompartment such as a PRE cluster might help to reinforce silencing. Chromatin rearrangements might be more difficult, or PcG proteins less mobile. Cavalli plans to test the latter theory using GFP-tagged PcG proteins. Reference: RNAs transcribed from between ribosomal genes alter the chromatin structure and thus silence their rRNA-coding neighbors, as discussed by Ingrid Grummt (German Cancer Research Center, Heidelberg, Germany). Ribosomal genes are in vast excess in the mammalian genome. Of 400 or so copies of rRNA genes, about half are permanently silenced after embryogenesis. Their heterochromatic state is induced by a complex called NoRC, which recruits histone-modifying enzymes to rDNA. When Grummt and colleagues noticed that RNase treatments dispersed NoRC from the nucleoli, they began searching for an RNA component. The group has since found that the RNAs responsible for targeting NoRC to chromatin originate from spacer regions that separate individual rRNA genes. Like the rRNA genes, the spacer RNAs are transcribed by RNA polymerase I. The RNAs bind to a subunit of NoRC, and this association is necessary for NoRC to grab onto the rRNA chromatin. The intergenic RNAs are a few hundred base pairs long and are complementary to the rRNA gene promoter. They have the potential to base-pair with the promoter as well as bind to NoRC, with the latter association depending more on RNA secondary structure than on specific sequence. Reference: Oxygen-starved cells conserve their limited energy by shutting down ribosome production. In his talk, Stephen Lee (University of Ottawa, Ottawa, Canada) suggested that cells silence ribosomal genes during hypoxia by locking a ubiquitin ligase in the nucleolus. In abundant oxygen, this ubiquitin ligase, called VHL, is a free-moving protein that keeps hypoxia-induced factor α (HIFα) levels low. But when cells are using anaerobic metabolism pathways, the resulting decrease in pH somehow causes VHL to stick in the nucleolus, where it cannot curb HIFα. Until this immobile form of VHL was identified, the only other known static protein was histone H2B, which silences chromatin. All other tested proteins exchange dynamically. Based on a supposed need for protein exchange during chromatin remodeling, Tom Misteli (NIH, Bethesda, MD) proposed in 2001 that static proteins in general might induce transcriptional silencing. Now with VHL in hand as a second immobile protein, Lee's group has put Misteli's hypothesis to the test. They found that VHL is indeed required to silence rRNA genes and thereby protect cells from hypoxia-induced death. The return of oxygen is expected to free VHL from the chromatin and restore rRNA synthesis. Lee was still unclear, however, how a ubiquitin ligase—or degradation of its substrate—is able to remodel chromatin to bring about silencing. References:

On the development of new probes, Atsushi Miyawaki (RIKEN) described new generations of fluorescent proteins that allow the use of light to switch them reversibly between bright and dim fluorescent states (“protein highlighting;” ; ), whereas Gerard Marriott (University of Wisconsin) prepared organic probes that undergo rapid and reversible optical switching between states of widely different chemical and photophysical properties (,). When conjugated to target proteins, these probes could be used to measure the dynamics or manipulate the chemical state of the protein repeatedly. For example, fluorescent protein highlighting was used to demonstrate the stimulation of bidirectional nuclear transport of MAP kinase upon growth factor activation (), and optical switching has allowed researchers to reversibly and rapidly control protein dipolar interactions inside cells. These techniques also can be used in combination with innovative strategies of FRAP, FRET, and super-resolution imaging (described later). Equally significant are new inorganic probes that are small, bright, and resistant to photodegradation (Robert Dickson, Georgia Tech; ). Once these probes are conjugated to specific targets, they should allow the detection of single molecules in living cells. Other biosensors can report the activation states (e.g., conformation and phosphorylation) of endogenous proteins with minimal perturbation (Klaus Hahn, University of North Carolina; ; ). These biosensors consist of an affinity element that binds only to the activated state of the target protein and a bright dye whose fluorescence emission responds to the binding. The development of the affinity element was further facilitated by using a phage display library combined with high-throughput screening of sequences capable of reporting a wide range of protein activities. A long-standing challenge with biosensors has been delivering them into living cells and targeting them to specific proteins or organelles of interest. Newly designed peptide sequences have now allowed the import of probes into living cells via nonendocytic pathways, thereby alleviating the problem of generating a constellation of brightly fluorescent vesicles (Klaus Hahn). In addition, Alice Ting (Massachusetts Institute of Technology) described how to exploit natural enzymes for protein-specific labeling (; ). For example, biotin ligase is known to ligate biotin to a lysine side chain within a specific 15–amino acid substrate sequence. By genetically fusing the 15–amino acid acceptor peptide to proteins of interest, coexpressing the biotin ligase, and providing a ketone analogue of biotin, her group showed that proteins in living cells could be specifically labeled with a variety of small-molecule probes, including fluorophores and photoaffinity labels. #text The integration of multiple disciplines is evident in the recent advancement of several imaging techniques. For example, fluorescence correlation spectroscopy (FCS) is being used with laser scanning microscopes, high-speed electron multiplying charge coupled device (EMCCD) cameras (Enrico Gratton, University of California, Irvine; Petra Schwille, Technische Universität, Dresden), total internal reflection fluorescence (TIRF) optics, and increasingly sophisticated mathematical analysis tools (Nancy Thompson, University of North Carolina, Chapel Hill; Enrico Gratton) to measure molecular concentrations, diffusion, binding, and aggregation densities both on the surface and inside living cells with high spatial and temporal resolution (; ; ,; ; ). Cross-correlation methods further allow direct comparison of the mobility of multiple components labeled with different color fluorophores and reveal their interactions within complexes (Petra Schwille and Enrico Gratton). These studies will benefit from new fluorescence dyes that can be excited at a common wavelength while emitting at easily separable wavelengths (Atsushi Miyawaki). Another elegant example is the application of switchable fluorescent probes, as described above, to build images by switching on (and bleaching off) a few diffraction-limited spots (Airy disks) at a time. Referred to as photoactivated localization microscopy (PALM), this approach provides superb spatial resolution (although at the expense of temporal resolution, since it takes time to cover the entire image with hundreds of thousands of Airy disks), as centroids of these individual Airy disks are well separated and may be determined at nanometer precision (Eric Betzig, Howard Hughes Medical Institute; ). A number of clever optical strategies have increased both the resolution and versatility of microscopy. For example, Tony Wilson (Oxford University) showed that the quality of images with conventional optics can approach that of confocal images, by illuminating thick samples with a laser through a one-dimensional grid, followed by removal of the pattern with computer processing (). A second application of patterned illumination, combining stripe illumination at multiple orientations with computer image processing, was developed by the University of California, San Francisco, cooperative (John Sedat, David Agard, and Mats Gustafsson) to overcome the diffraction limit and obtain unprecedented details from images of chromosomes (). Here, periodic illumination is used to extend the optical response of the microscope to high-frequency signals along a particular direction, by generating a moiré interference effect with greater (and therefore detectable) periodicity (; ). A technically more demanding technique of patterned illumination was described by Betzig, who used complex optical interference phenomena to generate a 3D lattice of diffraction-limited illumination spots for high-speed, high-resolution imaging of thick samples (). These spots illuminate a large number of regions simultaneously, with a photon efficiency far exceeding that provided by a confocal microscope. By combining optical manipulations with computation or innovative probes, many new powerful approaches may be invented with surprisingly low cost and great simplicity. In addition to patterned illumination and PALM described above, this includes combining oblique focus scanning (where optical sections are collected while the sample is shifted horizontally at the same time) and deconvolution to generate stereo pairs (Wilson and Sedat), and using flash illumination to limit how far single molecules diffuse during the detection phase as in strobe photography of high speed objects (Sunney Xie, Harvard University; ). Future developments may further incorporate new mechanisms to generate image contrast and to extend the usable range of electromagnetic waves, in order to take advantage of special chemical and physical characteristics of the materials. For example, Xie developed 3D coherent anti-stoke Raman scattering (CARS) microscopy (), which combines two laser beams at different wavelengths to generate a beating frequency that matches the vibration frequency of target molecules. These beams produce much stronger signals of Raman light scattering than conventional Raman scattering techniques. The technique has been used to image the dynamics of lipid droplets and to distinguish different lipids in unstained living cells, based on their different vibration frequencies. Finally, Carolyn Larabell's group (Lawrence Berkeley Lab) is developing soft X-ray microscopy as a new tool for whole-cell imaging (). The approach uses zone plate (diffractive) optics to focus X-rays onto the sample, and then onto a detector. At the wavelengths used, X-rays are absorbed by carbon and nitrogen, but not by water. The technique should be capable of producing high-contrast, 3D images of unstained cells at a resolution as high as 10 nm. Half of the meeting was dedicated to the applications of cutting edge imaging techniques to various important biological problems. A significant trend is the emphasis on single-molecule or single-organelle techniques, which remove the ensemble average in conventional fluorescence imaging to obtain behavior distributions, local dynamics, and kinetics without the need to synchronize individual molecules or cells. Using single-molecule FRET or single-particle tracking, investigators can observe the kinetics of molecular interactions, the appearance of conformational intermediates, and the molecular responses to force applied by an optical trap (Xiaowei Zhuang, Harvard University; Taekjip Ha, University of Illinois, Urbana-Champaign; ). Single-molecule observations are also being used in cells to determine distances at nanometer accuracy, by accurately measuring the center of the diffraction-limited spot emanating from single molecules. This approach has been used to reveal the mechanisms by which molecular motors such as dynein, kinesin, and myosin V move (Paul Selvin, University of Illinois, Urbana-Champaign; ; ; ). In addition, single-molecule tracking in cells is revealing distributions of diffusion coefficients for membrane proteins and lipids, as well as linear, directed motion arising from treadmilling of bacterial actin before and after cell division (W.E. Moerner, Stanford University). Single-organelle techniques are being used to understand the mechanisms that underlie endocytosis, exocytosis, and viral entry into cells (Zhuang; Rebecca Heald, University of California, Berkeley; Sandy Simon, Rockefeller University; Jennifer Lippincott-Schwartz, National Institutes of Health), while correlation spectroscopy, speckle microscopy, and molecule counting regimes are being used to understand dynamics of the cytoskeleton and substrate adhesions (Clare Waterman-Storer, Scripps Research Institute; Edward Salmon, University of North Carolina, Chapel Hill; Enrico Gratton). Although imaging cells in tissues is not as well developed as studying cells in culture, it is a critical frontier, and progress is promising. If excitation light enters the specimen orthogonal to the objective lens (using single plane illumination), fluorescence can be measured from all molecules that are located in the very narrow volume defined by the plane of excitation light. This greatly reduces photodamage and photobleaching and enhances the axial resolution (Ernst Stelzer, EMBL, Heidelberg; ). Although such new instrumentation (as well as lattice illumination described earlier) is likely to make a strong impact, significant advances have already been made using commercially available confocal optics. Peter Friedl (University of Würzburg) and John Condeelis (Albert Einstein College of Medicine) showed impressive imaging of cellular movements in skin and breast tissue, respectively, in live animals. Tumor cells in skin reside in a variety of states including single cells and collectives, and they migrate by either mesenchymal or amoeboid type motility (). Migration through collagen requires strategic clipping of collagen at constriction points as the cell squeezes through the matrix network, leaving a localized trail of oriented collagen fibers. In addition, breast tumor cells undergo a chemotactic migration along collagen fibers, guided by epidermal growth factor secreted by perivascular macrophages (). Equally impressive is the mapping of neural responses in the visual cortex by applying calcium indicators and two-photon optics to the brain of live animals (Clay Reid, Harvard Medical School). The responses of neurons, monitored within cubes that have sides ∼300 μm in length, show sharp changes as the animal receives the visual stimulation of a bar moving in different directions (). Finally, Scott Fraser and his colleagues (California Institute of Technology) have developed comprehensive imaging and informatic systems for high-throughput, parallel analysis of the cellular trajectories of hundreds or thousands of cells in live zebrafish embryos, with the ultimate goal of understanding the exact pathways of cellular movements during embryonic development (). The synergy and enthusiasm generated during the meeting reflect the importance of live cell imaging, the incisive applications to date, and the prospects for significant developments in the near future. The localized nature of biological phenomena serves as the driving force for the development of technologies to observe, assay, and parse their mechanisms. The meeting pointed to a very bright future for live cell imaging, by showing that higher resolution, better reagents, and more powerful methods of revealing interactions and dynamics are well within reach. In addition, as suggested by Roger Tsien (University of California, San Diego), one of the keynote speakers, basic research in this area may soon lead to revolutionary advances in clinical diagnosis and treatments, both by direct applications of “optical physiology”—the probing and manipulation of physiological events with optical techniques—and by spinning off techniques involved in probe development and delivery.

Many of the current questions facing cell biologists are related to the mechanisms that allow cells to move and divide. Cell migration and division are profoundly important in shaping the embryo and wiring the developing nervous system. In the adult, these two processes are involved in tissue maintenance, regeneration, and the immune response, as well as in the pathology of numerous diseases. The morphological changes required for cell migration and during cytokinesis are similar in many respects. Both processes involve elaborate control of the cytoskeleton to produce and dynamically modulate cell polarity. A moving cell typically maintains a single front and back, whereas a dividing cell is bipolar, with two poles flanking the future site of cleavage. In many migratory cells, there is a seamless transition between these morphologies (). As a migrating cell prepares for cytokinesis, it rounds up and appears to “erase” its polarity. It then elongates and begins to constrict at the midline. Membranes from opposite sides of the cell invaginate, defining the fission furrow, and then fuse, pinching off the newly formed daughter cells, which typically move apart. The backs of the new cells correspond to the point of cleavage, and the fronts derive from the former poles of the dividing cell. Some cells lack a significant migratory phase, yet still cycle through the unipolar, bipolar, and nonpolar stages. The morphological symmetry seen between migration and cytokinesis carries through on a molecular level. A series of proteins that localize to the fronts of migrating cells, including RacE, dynacortin, coronin, and scar, are also found at the poles of dividing cells, whereas others, such as myosin II, Rho GTPases, and cortexillin, which associate with the back of moving cells, are concentrated at the furrow during cytokinesis (; ; ; ; ; ; ). These localizations ensure that newly polymerized actin creates leading edge projections, or polar ruffles, whereas actomyosin creates contraction at the rear and within the ring that pinches apart the daughter cells (; ). The polarization of the cytoskeleton requires signaling from the plasma membrane, and this communication involves the phosphoinositides (PIPs), chiefly phosphatidylinositol 4,5-bisphosphate (PIP) and phosphatidylinositol 3,4,5-trisphosphate (PIP). PIP, which is relatively more abundant in the plasma membrane, interacts permissively with a variety of cytoskeletal proteins in different spatial localizations. PIP influences or activates several regulators of actin polymerization, including WASP, profilin, cofilin, and capping protein (). In vitro PIP also binds to septins and ezrin-radixin-moesin (ERM) family members like α-actinin (; ; ; ; ). Studies on migrating cells show that PIP accumulates in the membrane at the cell's leading edge and aligns the cell along the chemoattractant gradient (; ). The locally elevated PIP levels instruct new actin polymerization and pseudopodia extension at the front. The lower levels at the rear inhibit projections in this region and help specify the back (). Positive feedback between the actin cytoskeleton and plasma membrane PIP reinforces the asymmetric response and can spontaneously bring about polarity, even in the absence of a gradient (; ; ). The participation of PIPs, which typically serve as secondary messengers, seems natural for migration, but might seem unusual for cytokinesis, where there is no known receptor-mediated signaling. Polarity during cytokinesis depends instead on internal cues. It is thought that astral microtubules interact with and alter the cell cortex to set polarity, and that feedback from the polarizing cortex controls the progress and orientation of the spindle (; ; ; ; ; ). New evidence now indicates that lipid-signaling pathways appear to coordinate membrane and cortical events during cytokinesis. Several investigators have recently found that regulation of PIP is necessary for proper furrow ingression during cytokinesis. Similarly, we have found that a “polarity circuit,” involving the temporal and spatial regulation of PIP metabolism, plays a key role in establishing polarity for cell division. In one of the first experiments to directly suggest a role for PIP in cytokinesis, injection of anti-PIP antibodies into embryos was shown to lengthen the cell cycle and inhibit cleavage furrow progression (). Earlier experiments on sea urchin zygotes had shown that Li blocks cytokinesis (). Li inhibits inositol monophosphatase and inositol polyphosphate-1-phosphatase, which can lead to depletion of PIPs (; ). This inhibitory effect was reversed by the precursor myoinositol (; ). Genetic manipulations and drugs have assessed the role of PIP metabolism in cytokinesis. Mutations in the gene (), which encodes PI4Kβ, resulted in a cytokinesis defect during male meiosis, suggesting a vital role for PIP synthesis during this process (). In both PIP 5-kinase () and its product PIP were concentrated in the medial ring during cytokinesis. Also in , the homologue of both and were found to be required for proper cell division (; ). PIP was also found to be important in mammalian cells, as its depletion or sequestration with neomycin blocked cytokinesis in cultured cells (). Furthermore, in crane fly spermatocytes, Li, the PI-kinase inhibitors wortmannin and LY294002, and the PLC inhibitor U73122 all stopped or slowed furrowing after initiation (). These initial studies have raised further questions. In general, are PIP and the enzymes that regulate it localized to the furrow? What are the targets of PIP and/or its products? reported a series of experiments indicating that hydrolysis of PIP to IP3 and DAG could be a key step for completion of cytokinesis. They depleted PIP by various methods, such as dephosphorylation by the phosphoinositide phosphatase SigD (; ) and sequestration by PLCδ-PH-GFP and by the cell-permeable PIP-binding peptide PBP10 (). All three treatments caused cytokinesis defects (Wo). With SigD treatment, the spermatids contained multiple nuclei; the majority of spermatids contained three or four nuclei, suggesting failures at both meiosis I and II. Expression of PLCδ-PH-GFP caused a statistically significant, but minor, defect in <2% of transformed cells. PBP10, which was administered to cells after furrow ingression had initiated, caused regression followed by cytokinesis failure in the majority of cells treated. When cells were treated with the PLC inhibitor U73122, furrow regression occurred in a rapid and dose-dependent manner (). When cells were treated with the IP3R antagonist 2-APB, cleavage furrows regressed, suggesting that Ca is required for completion of cytokinesis (Wo). Further evidence for Ca was obtained when cells maintained in Ca-free buffer were treated with the membrane permeable Cachelator, BAPTA-AM. Cytokinesis failure rates as high as 40% (10 μM) and 90% (300 μM) were recorded. Based on these observations, the authors suggest that the hydrolysis of PIP and the release of Ca may be involved in contractility by activating myosin or by stimulating vesicle fusion events by activating SNAREs (; ). Interestingly, PLCδ-PH-GFP, which specifically binds PIP, was found localized to the plasma membranes, but was not highly enriched in the furrow. Although PIP was uniformly distributed in spermatocytes, two other groups found that PIP was localized in the furrow in mammalian tissue cells during ingression (; ). They also reported that interference with PIP impaired cytokinesis. expressed a series of PH domains specific for PIP in HeLa, CHO, RAW, and 3T3 cells. PLCδ-PH-GFP and GFP-TubC were associated with the plasma membrane and enriched in the cleavage furrow of dividing cells, suggesting that PIP was elevated in these regions (). Low expression levels of these constructs did not interfere with cytokinesis. However, a 10-fold overexpression resulted in a separation of cytoskeletal actin from the plasma membrane. Similarly, there was detachment of actin from the plasma membrane at the furrow in cells expressing the PIP phosphatase synaptojanin. They also found that high levels of PLCδ-PH-GFP and GFP-TubC both independently doubled the duration of cytokinesis and increased the frequency of multinucleate cells. This is consistent with the PLCδ-PH-GFP data from Wo. Furthermore, expression of synaptojanin, as well as a dominant-negative, kinase-inactive PIP-5 kinase, both reduced the plasma membrane PIP and increased the amount of multinucleate cells. Based on these observations, the authors suggest that plasma membrane PIP levels in the constricting furrow must be sufficient to sustain adhesion between the plasma membrane and the underlying actin cytoskeleton for proper maintenance of the furrow. Emoto et al. similarly found that PIP localized to the furrow and were able to cause cytokinesis defects when they interfered with PIP accumulation (). In both normal cells and in cells specifically arrested during late cytokinesis, PIP-5 kinase, RhoA, and PIP were highly localized in the furrow of CHO cells (). Both the overexpression of catalytically inactive PIP-5 kinase and the microinjection of specific anti-PIP antibodies impaired cytokinesis. The antibody treatment increased the amount of multinucleate cells from ∼6 to ∼32%. These authors suggest that localized production of PIP is needed for proper cytokinesis and that there is likely a unique lipid domain in the cleavage furrow. This is also consistent with a recent paper showing a requirement for PLCγ in sea urchin cytokinesis (). Recent studies of have revealed a regulatory circuit involving PIP that controls polarity in cytokinesis, as well as in migration (). The two enzymes controlling PIP, PI3K, and PTEN, were reciprocally regulated during cell division in a strikingly similar manner, as has been described previously for migrating cells (; ). As cells rounded up at the onset of cytokinesis, PTEN-GFP moved uniformly to the membrane, closely resembling the localization of myosin II. At the same time, membrane PI3K-GFP localization was inhibited. Then, as the cell elongated, PI3K-GFP and PTEN-GFP associated with the cell membrane at the poles and furrow, respectively (). This spatial modulation of PI3K and PTEN controls the local distribution of PIP, and perturbation of the polarity circuit including these enzymes adversely affects migration. This new study demonstrated that inhibiting these enzymes had deleterious effects on the process of cell division as well. PTEN was disrupted in a cell line already lacking PI3K1 and PI3K2. These two PI3Ks account for ∼90% of the PI3K activity in wild-type cells; there are an additional six PI3K enzymes in cells (; ; ). These mutant cells, which contained basal PIP levels higher than normal, but lacked temporal or spatial regulation by external or internal cues, failed to undergo cytokinesis in shaking suspension, resulting in large multinucleate cells. Mutant cell lines missing PI3K1 and PI3K2 or PTEN alone also had modest cytokinesis defects. Furthermore, we found that cells grown on surfaces and treated with the PI3K inhibitors wortmannin or LY294002 were dramatically delayed or failed to undergo cytokinesis nearly 25% of the time. Based on studies of migrating cells, we have speculated on the role of PIP during cytokinesis. Because local increases in PIP promote actin polymerization and pseudopod extension, cells must carefully regulate PIP accumulation during cell division. At the onset of cytokinesis PIP signaling is globally suppressed, which leads to a loss of actin polymerization and membrane ruffling. This allows the cell to round up and reset its polarity. Then, as the cell elongates, the poles gradually become active zones of PIP-labeled, actin-filled projections, whereas the midzone and forming furrow are devoid of PIP and remain quiescent. This polarization of the membrane and cytoskeleton is critical in positioning the initial furrow and in driving the progression of cytokinesis. When we imaged the cells lacking PIP-modulation, actin-based projections were apparent all around the cell perimeter. This unregulated activity of the cytoskeleton likely contributes to failure at multiple stages of cytokinesis. First, it would be more difficult for these cells to reset polarity at the onset of cytokinesis. Without appropriate PIP regulation, the elongating spindle may be unable to gain control of the membrane and cytoskeleton. Second, the hyperactivity would destabilize localized activity of the cytoskeleton during furrow formation. The regulation of PIP is likely involved in many types of cell shape changes in numerous systems. A similar association of PTEN with the septa of during cell division is likely related to the mechanism described here (; ). A recent report found PTEN localized to cell–cell junctions in photoreceptors, where the enzyme regulates PIP levels and is essential for epithelial cell morphogenesis and apical basal polarity (). Evidence that has emerged in the last few years and been brought together here suggests that PIPs play a key role in regulating polarity in cytokinesis, as well as in migration, although the precise mechanisms remain to be defined. The involvement of these intracellular signaling molecules in cytokinesis is consistent with the close parallel of the front and back of migrating cells with the poles and furrow of dividing cells. Similar cytoskeletal components localize at each corresponding zone. There appears to be a seamless transition in cell morphology as cells shift from cell migration to cell division. As cells round up before cytokinesis, proteins normally located at the back, such as myosin, cortexillin, and PTEN, are recruited along the entire cell periphery. Then as the spindle elongates, they localize to the furrow. Likewise, stitching together data from various cell types, there seems to be a general consensus that PIP is found at the front or at poles, whereas PIP is localized to the back or the furrow. PIP may play a permissive role in regulating cytoskeletal proteins, several of which contain PIP-binding motifs, whereas in other instances it is clearly the products of PIP hydrolysis that are important. Interestingly, PTEN contains a PIP-binding motif (; ). Is it possible that PIP helps direct PTEN to the furrow? Considering that the phosphatase activity of PTEN regulates the conversion of PIP to PIP, it is easy to imagine that local levels of PIP might rise, and PTEN associations would form a powerful feedback loop for polarity in cytokinesis, as well as migration. Although is the only organism in which the localization of PTEN has been observed at the midline of a dividing cell, observations of PIP at the furrow have been made in other organisms. Also, although PIP has been imaged at the front of many migrating cells, it has only been imaged during cytokinesis in . Nevertheless, the studies discussed in this paper clearly illustrate a key role for phosphoinositide signaling in cytokinesis and provide a missing link between the plasma membrane and cytoskeleton. Future studies will address the generality of these findings and the precise role of the signaling in this fascinating process.

xref #text The conserved membrane-proximal NPXY motif in the β1 tail regulates integrin activation (; ). To test whether this motif is required for cell proliferation, we generated CHO cell lines stably expressing either a wild-type (WT) β1 tail or a mutant β1 tail with an alanine substitution at tyrosine 783 within the NPIY motif (Y783A cells) in the context of the αIIb-5β3-1 heterodimeric chimeric integrin. These chimeras contain the extracellular and transmembrane domain of the αIIbβ3 fibrinogen (Fg) receptor connected to the tails of the α5β1 fibronectin (Fn) receptor (), allowing CHO cell adhesion to Fg (). We isolated the function of the recombinant chimeras by adhering cells to Fg in the serum-free growth medium CCM1 that does not support CHO cell proliferation in the absence of a preexisting matrix (unpublished data). WT cells showed robust proliferation on Fg in CCM1, whereas CHO K1 and Y783A cells proliferated poorly (). CCM1 similarly promoted proliferation of Y783A and CHO K1 cells on Fn (). Furthermore, infection of Y783A cells with an adenovirus that directed the expression of the β3-1 chimeric subunit containing the WT β1 tail restored cell proliferation of Y783A cells (unpublished data). Although Y783A cells show slow adhesion kinetics on Fg (Fig. S1 A, available at ), most cells adhere and spread by 3 h (Fig. S1 B). Thus, the defect in proliferation is not simply due to a lack of adhesion. Because anchorage-dependent cells require integrin signaling for entry into S phase (), we compared cyclin D1 induction and DNA synthesis in WT and Y783A cells. As expected, both were low in serum-starved cells. Surprisingly, cyclin D1 accumulation in Y783A cells at 7 h was equivalent to that in WT cells (). Furthermore, WT and Y783A cells incorporated similar levels of BrdU (). Thus, the proliferation defect is unlikely to be at the G1–S transition. The poor proliferation of the Y783A cells was accompanied by accumulation of bi- and multinucleated cells (). Analysis of the binucleation kinetics revealed a uniform increase in the nuclei per cell in Y783A cells with no significant change in WT cells (). Because the most likely explanation was a defect in cytokinesis, we examined this process by time-lapse microscopy. After rounding at mitosis, WT cells completed cleavage furrow ingression within 5–10 min and cytokinesis within 20–30 min (unpublished data). In contrast, most Y783A cells attempting cytokinesis showed cleavage furrow regression and cytokinesis failure. This phenotype was suppressed when Y783A cells were adhered to Fn (). Quantification of cytokinesis attempts and successes during the first cell cycle on Fg in CCM1 indicated that ∼90% of the WT cells attempted and successfully completed cytokinesis within 16–20 h (). A significant percentage of Y783A cells attempted cytokinesis, but most failed to divide (). Many Y783A cells showed a partially constricted cleavage furrow changing in diameter through an extended period of time and finally regressing to produce binucleated cells. In those few cases where cleavage furrow ingression was completed, midbody formation and/or daughter separation was significantly delayed or inhibited (unpublished data). As expected, Y783A cells successfully completed cytokinesis under the same conditions that promoted their proliferation (). Together, our data indicate that the Y783A mutation inhibits the successful completion of cytokinesis. To gain mechanistic insight, we compared the actin and MT cytoskeletons in mitotic WT and Y783A cells that had proliferated on Fg in CCM1 for 15–18 h (a time of peak in cytokinesis attempts; ). Cells at prometaphase/metaphase were identified by their round morphology and the presence of condensed chromosomes. At this stage, the majority of WT cells (85%) formed functional bipolar spindles, as judged by α- and γ-tubulin distribution and chromosome congression at the equatorial plane (). In contrast, most Y783A cells showed random distributions of chromosomes and multipolar spindles or no evidence of spindle assembly (). As expected, Y783A cells formed functional bipolar spindles on Fn (). At anaphase, WT cells showed normal contractile rings and chromosome segregation (); in contrast, most Y783A cells showed evidence of multiple contractile rings and a lack of chromosome segregation consistent with the presence of aberrant spindles (, middle). The few Y783A cells that were able to segregate chromosomes (, right) exhibited abnormalities at telophase (Fig. S2, available at ). Collectively, our data strongly suggest that the Y783A mutation inhibits cytokinesis by preventing the formation of a normal bipolar spindle. To determine whether the defects were restricted to mitosis, we compared the MT cytoskeletons of Y783A and WT cells at interphase. WT cells had a complex array of numerous, long, distinct polymers originating from the centrosome and radiating to the cell cortex as expected (). In contrast, Y783A cells had fewer and more randomly organized MTs, which did not appear to emanate from the centrosome (). The Y783A phenotype was not due to decreases in expression of tubulin or surface chimeric integrins (Fig. S1, B and C). However, when we compared MT regrowth after nocodazole washout (), we found that the regrowth of a radial MT array from the centrosome was clearly inhibited in Y783A cells (). These phenotypes were rescued when Y783A cells were adhered to Fn ( and not depicted). The Y783A mutation in the β1 tail is known to inhibit the expression of the high-affinity conformation of αIIb-5β3-1 in CHO cells (). Therefore, we tested whether LIBS6, a specific activating antibody for αIIbβ3 (), could prevent the mutant phenotypes in Y783A cells. In control experiments, LIBS6 activated the Y783A chimeric integrin (unpublished data) and promoted rapid adhesion and spreading of Y783A cells on Fg (Fig. S1 D). Importantly, LIBS6 rescued the assembly of a radial MT network (), MT regrowth from centrosomes at interphase () and spindle poles at mitosis (), the assembly of a bipolar spindle (), and cytokinesis (; and Videos 1–3, available at ). In agreement with these results, LIBS6 also prevented binucleation (unpublished data). To determine whether the effects are specific for the Y783A mutation and the LIBS6 antibody, we tested additional mutations known to regulate integrin conformation (, ). Coexpression of the αIIb-L deletion (αIIb-LΔ) with the β3-1(Y783A) chimeric subunit resulted in a constitutively active integrin (Fig. S2, D–F) and rescued MT regrowth (), as observed with LIBS6. In addition, coexpression of a β3-1 subunit containing an N780A mutation in the NPIY motif (β3-1[N780A]) with the αIIb-5 subunit prevented soluble Fg binding (Fig. S2, D–F) and inhibited MT regrowth (), mimicking the effects of the Y783A mutation. Thus, the inhibition and rescue of MT regrowth are not specific to the Y783A mutation and the LIBS6 activating antibody. To demonstrate that the effects of the Y783A mutation were not specific to CHO cells or due to its expression in the context of the αIIb-5β3-1, we generated GD25 cell lines (β1-null; ) expressing full-length human β1 containing either the WT (GD25 h-β1WT) or the Y783A mutant tail (GD25 h-β1Y783A; ). In contrast, to the GD25 h-β1WT cells, MT regrowth from interphase centrosomes was inhibited in GD25 h-β1Y783A cells adhered to laminin-1 (Lm). Moreover, this phenotype was suppressed by TS2/16 (), a specific β1-activating antibody (). In addition, when these cells were mitotically arrested and replated on Lm, WT but not Y783A cells formed a normal bipolar spindle (). Thus, the effects of the Y783A mutation are not cell-type or integrin specific. Our results provide the first evidence that integrins can regulate the assembly of the MT cytoskeleton during interphase and the bipolar spindle during mitosis and indicate that integrin activation is important for both. Previous studies demonstrated that the NPXY motif in β tails regulates activation by binding to talin (). Activating integrin antibodies circumvent the requirement for talin–β tail interactions in integrin activation (). The ability of LIBS6 and TS2/16 to rescue the MT regrowth, bipolar spindle formation, and cytokinesis suggests that protein interactions with the NPIY motif, including talin binding, are not required downstream of integrin activation to regulate these processes. These antibodies may promote the association of integrins with other receptors and/or the interaction of the tail with cytoskeletal or signaling proteins to regulate the MT cytoskeleton. MT dynamics are under tight and complex control throughout the cell cycle. At interphase, the organization of the MT network requires the regulation of MT nucleation, growth, and anchorage at the centrosome and the association of MTs with the cell cortex. The assembly of the bipolar spindle at mitosis also requires the regulation of kinetochore-associated MTs, as well as centrosome duplication and cohesion (; ; ; ). The Y783A mutation may inhibit the activity of one or more of the proteins that regulate these events. The goal of future studies will be to identify the aspects and targets of integrin function required for spindle assembly and cytokinesis. CHO K1, WT, and Y783A CHO cell lines were cultured in F12 medium ± 10% FBS or CCM1 (Hyclone). WT and Y783A-GD25 cells were cultured in DME + 10% FBS or CCM1 as indicated. The generation of stable CHO and GD25 cell lines is described in the supplemental text (available at ). Transient transfection of CHO K1 cells was performed using the Mojo transfection reagent (Mirus). Where indicated, cells were serum starved in F12 or DME and replated on the indicated matrices. For binucleation and cytoskeletal analysis, ∼5 × 10 cells were replated in CCM1 in 24-well dishes on 15 μg/ml Fg/Fn (CHO cells) or 30 μg/ml Lm (GD25 cells) and processed as described in the figure legends. Serum-starved cells were replated onto either Fg- or Fn-coated (15 μg/ml) 12-well dishes (3 × 10 cells/well, in triplicate), fixed at the indicated times in 3.7% paraformaldehyde, and stained with crystal violet (0.5% in 20% methanol) for 1 h at room temperature. Incorporated dye was extracted with 1% SDS and quantified by measuring A in a spectrophotometer. To assay DNA synthesis, serum-starved cells and serum-starved cells replated on Fg in CCM1 for 5 h were supplemented with BrdU for 18 h. Cells positive for BrdU incorporation were quantified by microscopy. Serum-starved cells (∼5 × 10) were replated on 15 μg/ml of either Fg or Fn (CHO cells) or on 30 μg/ml Lm (GD25 cells) in CCM1. 3–14 h after plating, cells were treated with 10 μg/ml nocodazole (Calbiochem) for 2 h at 4°C, washed with cold PBS to remove the drug, allowed to nucleate MTs for 5–15 min in warmed CCM1 ± LIBS6 (CHO cells) or TS2/16 (GD25 cells), and processed for immunofluorescence microscopy. Serum-starved cells were replated (24-well dishes) on Fg or Fn in CCM1 supplemented with 10 mM Hepes, pH 7.4. 3 h after seeding, when most cells were fully spread, dishes were transferred to a microscope equipped with a heated (37°C) chamber, and phase-contrast images were recorded (12 frames/h) during the first cell cycle (16–20 h). Multiple fields (three per sample) were analyzed using a rotary stage and 20× objective. Where indicated, cells were arrested at metaphase by nocodazole treatment, isolated, replated on Fn or Fg in CCM1 ± LIBS6, and imaged as before but at 2 frames/min to analyze cytokinesis in greater detail. Cells were permeabilized for 30 s in 80 mM Pipes, pH 6.8, 5 mM EGTA, 1 mM MgCl, and 0.5% Triton X-100; fixed for 10 min in the same buffer containing 5% glutaraldehyde; and incubated for 7 min in 1% sodium borohydrate in PBS. To coanalyze MTs and F-actin, cells were fixed in 3.7% paraformaldehyde and 1% sucrose in PBS for 5 min, permeabilized in PBS and 0.5% Triton X-100 for 10 min, and processed for immunostaining as glutaraldehyde-fixed cells (excluding the glutaraldehyde-quenching step). The following antibodies were used: α-tubulin (mouse mAb DM1A; Sigma-Aldrich), γ-tubulin (rabbit AK-15; Sigma-Aldrich), Alexa Fluor 488–conjugated goat anti-mouse (Invitrogen), and Alexa Fluor 594–conjugated goat anti-rabbit (Invitrogen). F-actin was analyzed with Alexa Fluor 594–conjugated Phalloidin (1:300 dilution), and DNA was stained with 1 μM of either Hoechst 33342 or DAPI (Invitrogen). Samples were analyzed using an inverted microscope (TE2000-E; Nikon) equipped with phase contrast and epifluorescence, a digital camera (CoolSNAP HQ; Roper Scientific), a Ludl rotary encoded stage, 37°C incubator, and MetaVue (Molecular Devices) and AutoQuant deconvolution software (AutoQuant Imaging, Inc.). Protein expression was analyzed by Western blotting using the following commercially available antibodies: α-tubulin (mouse mAb DM1A), cyclin D1 (mouse mAb DCS-6; BD Biosciences), Pan ERK (mouse mAb 16; BD Biosciences), and cyclin B1 (mouse mAb GNS1; Santa Cruz Biotechnology, Inc.). Blots were stripped and reprobed with the indicated antibodies for loading controls. The supplemental text provides information relating to the generation of chimeric integrins and stable cell lines. Fig. S1 shows the adhesion and spreading of WT and Y783A cells on Fg in CCM1, expression levels of α-tubulin and chimeric integrins under these conditions, and the promotion of rapid adhesion and spreading of Y783A cells on Fg by LIBS6. Fig. S2 shows the characteristic telophase phenotypes in these cells under similar conditions (S2A-C), as well as the characterization of the ability of CHO cells transiently expressing WT and mutant chimeric integrins to bind soluble and immobilized Fg (S2D-G). Videos 1 and 2 show the completion and failure of the cytokinesis in WT and Y783A cells, respectively, on Fg in CCM1. Videos 3 and 4 show the rescue of cytokinesis in Y783A cells on Fn or on Fg when treated with LIBS6, respectively. Online supplemental material is available at .

During the mitotic cell cycle, the regulated segregation of sister chromatids is achieved by the mitotic cohesin complex, and upon entering meiosis, additional meiosis-specific cohesin variants appear to choreograph meiosis-specific chromosomal events (; ). The meiotic cohesin complex not only mediates sister chromatid cohesion but also plays a critical role in assembling a proteinaceous chromosome axis as its major component (). This chromosome axis later forms the axial/lateral element of the synaptonemal complex (SC; ; ; ; ). In many organisms, longitudinal compaction of the chromosome is observed along with chromosome axis formation during meiotic prophase (; ; ; ). However, the molecular mechanisms that underlie meiotic prophase chromosome compaction remain unclear. In addition to the core cohesin complex, a conserved protein known as Pds5, Spo76, or BimD is associated with cohesin and is implicated in sister chromatid cohesion and chromosome structure maintenance. This protein has been studied in (called Spo76), (called BimD), , and (, ; ; ; ). Two Pds5s, Pds5A and Pds5B, have been identified in vertebrate cells (; ). A role for Mcd1/Scc1 (the Rad21 homologue) and Pds5 in mitotic chromosome condensation has been suggested in : FISH analysis demonstrated that these molecules are required for ribosomal DNA (rDNA) condensation at metaphase of mitosis (; ). It has also been shown that Pds5/Spo76 has a role in meiosis because its loss results in defects in spore formation and SC integrity (; ). In the fission yeast , the mitotic cohesin complex is composed of two structure maintenance of chromosome (SMC) subunits, Psm1 and -3, and two non-SMC subunits, Rad21 and Psc3 (). In meiosis, Rad21 is largely replaced by a meiosis-specific cohesin, Rec8, and this exchange is essential for reductional segregation of chromosomes in the first meiotic division (meiosis I; ; ; ; ). Rec11 is the meiotic Psc3 counterpart and acts at the chromosome arms (). mutants (). Unlike many other organisms, shows no obvious chromosome condensation in meiotic prophase. In addition, this organism does not assemble canonical SC structures but forms so-called linear elements (LEs), which are evolutionally related to the axial/lateral elements of the SC (; ; ). and mutants (, ). On the other hand, Pds5 is nonessential for mitotic growth, but its loss reduces viability in G2-arrested cells (; ), suggesting a role for Pds5 in maintenance of sister chromatid cohesion. Pds5 is also required for spore formation in (), suggesting that it plays a role in meiosis. In , meiotic prophase is characterized by an elongated nucleus, which is generally called a “horsetail” nucleus. The horsetail nucleus moves back and forth between the cell ends during meiotic prophase, and telomeres remain clustered at the leading edge of the moving nucleus (; ). Observation of homologous pairing in living meiotic cells has demonstrated that telomere clustering and oscillatory chromosome movements spatially align homologous chromosomes in the early stages of meiotic prophase to promote their contact, which is stabilized later by homologous recombination (). The telomere-clustered movement aligns chromosomes along the direction of the movements, providing a unique opportunity to examine chromatin structures within a defined orientation of the chromosome. In this study, we have identified a role for Rec8 and Pds5 in chromosome compaction by directly measuring chromosome compaction in living cells: Rec8 modulates chromosome compaction during meiotic prophase, and Pds5 is required for stable binding of Rec8 to the chromosome. Our results demonstrate that meiotic cohesins are essential for compaction of chromosomes in meiotic prophase. In a screen of meiotic mutants, we have observed nuclear movement during meiotic prophase in living cells of (). mutant: although the telomeres repeatedly traversed the cell, the bulk of the chromosomes did not follow ( and Video 1, available at ; ). This suggests that chromatin architecture is altered in this mutant. mutant, we found that cells had a shorter nucleus. double mutant showed an elongated morphology similar to that observed in the single mutant (), indicating that the shorter nucleus in cells is dependent on Rec8. These results suggest that Rec8 and Pds5 are required for proper chromosome structure in meiotic prophase and that the function of Pds5 requires the presence of Rec8. To further examine chromosome structures, we took advantage of the polarized orientation of chromosomes in the telomere-led nuclear movement to directly measure chromosome compaction in living cells. We measured the distance between the telomere and the locus, using a GFP-tagged telomere protein (Taz1) and a /lacI-GFP tag at the locus (). cells (), indicating that the chromosomes were more extended or more flexible in the absence of Rec8. cells than in wild-type cells (). as well as wild-type cells (). Thus, chromatin structure is altered in the absence of Rec8, being more flexible to allow for the pulling forces of nuclear movements. To determine whether meiotic prophase chromosome structure depends on the presence of other cohesin subunits, we examined chromosome compaction in the absence of another meiotic cohesin component, Rec11. mutant ( and see ). mutant, although localization of Rec8 and Psm3 was similar to that of wild-type cells (Fig. S1 A, available at ), the mitotic counterpart of Rec11, Psc3, which concentrates at the centromere in wild-type cells (), was relocalized along the entire length of the chromosomes (Fig. S1 B). Thus, Psc3 could partially supplement the usual roles of Rec11, thereby resulting in a milder phenotypic alteration in chromosome compaction. and mutants, we found that in cells, the distance from the telomere to or from to loci was about one half of that seen in wild-type cells (). cells than in wild-type cells throughout meiotic prophase (). Thus, the chromosome was hypercompacted in the absence of Pds5. We then calculated the apparent longitudinal DNA compaction of meiotic prophase chromosomes using the measured distance data. In wild-type cells, the DNA compaction ratio was 80–110, approximately two times more compact than the 30-nm chromatin fiber (; ). However, in the absence of Rec8, the apparent DNA compaction ratio decreased to 32 at the moving edge, similar to the 30-nm chromatin level. mutant, the apparent compaction ratio significantly increased to 140–290 (). , , and wild-type cells (). , , and wild-type cells. To address whether Rec8 and Pds5 are also involved in chromosome architecture in both mitosis and meiosis I, we evaluated chromosome compaction at early anaphase by labeling two loci on the same arm of chromosome I: the locus and locus (). and mutants (). These results suggest that Rec8 and Pds5 do not significantly contribute to the chromosome architecture, at least not at anaphase. In , Rec8 is required for LE formation and homologous recombination (; ; ). and mutants. mutant, no LE structures form (; ). Rec12, a homologue of Spo11, generates double-strand breaks and is required for homologous recombination (Cervantes et al., 2000). , , and wild-type cells (). double mutant, the hypercompaction caused by Pds5 loss was also observed (unpublished data). Thus, the meiotic cohesins and Pds5 support chromosome compaction during meiotic prophase in a manner independent of LE formation or recombination. To evaluate the contribution of cohesins to meiotic chromosome structures, we examined the interdependency of cohesins and Pds5 in their localization. In wild-type meiotic prophase nuclei, Rec8-, Rec11-, and Psm3-GFP formed thin filaments (, Fig. S2, and Fig. S3, available at ), which probably represent the axis of the chromosome. Pds5-GFP also formed thin filaments, with additional foci at the centromeres (), as confirmed by its colocalization with the centromere protein Mis6 (). In cells, however, Rec11- and Psm3-GFP did not concentrate at the chromosome axis; only a residual punctate Psm3-GFP signal was found at the leading edge and putative centromere regions of the nucleus in such cells ( and Fig. S3). Thus, localization of Rec11 and Psm3 on the chromosome axis is dependent on Rec8. On the other hand, localization of Pds5 was not greatly affected by the loss of Rec8 or Rad21 (). In mitotic cells, Pds5 binds chromosomes in a Rad21-dependent manner (; ; ; ). Whereas Rad21 binds along with Rec8 on the chromosome axis in meiotic cells of many other eukaryotes (; ), Rad21-GFP is confined to the rDNA region in , which is located at the leading edge of the horsetail nucleus (because rDNA locates next to the telomeres of chromosome III; ; ). In cells, some Rad21 extends to the chromosome arm (; ). Localization of Pds5 on the chromosome axis decreased when Rad21 was inactivated in the temperature-sensitive mutant in the -deletion background (), indicating that Pds5 localization depends on Rec8 and, in the absence of Rec8, on Rad21. These localization results are consistent with the results of immunoprecipitation experiments. Pds5 is known to coprecipitate with Rad21 (), and we found that mitotically expressed Rec8-HA could also coprecipitate with Pds5-Myc, but only when Rad21 was absent (). These results suggest that Pds5 associates with either cohesin, predominantly Rad21 in mitosis, and could associate with Rec8 in the absence of Rad21. cells had loose chromosomes despite the presence of Rad21 on the chromosome; thus, the relocated Rad21 is not sufficient for replacing the function of Rec8 in compaction of chromosome arms. Collectively, the results show that Rec8 recruits cohesins to the chromosome axis and plays a key role in forming meiotic prophase chromosome structure. In the absence of Pds5, Rec8-GFP showed distinct staining along the compacted chromosomal axis ( and Fig. S2 A), and Rec11-GFP and Psm3-GFP showed a similar pattern to Rec8-GFP (, Fig. S2, and Fig. S3). mutant. mutant. Because Rec8 is a key molecule in meiotic prophase chromosomes, the binding of Rec8 to chromosomes must be precisely controlled. We therefore examined the amount of Rec8-GFP in meiotic cells. Because it is difficult to follow temporal changes of the amount of Rec8 binding on chromosomes by detergent extraction or chromatin spreads, we determined the amount of Rec8 by measuring GFP fluorescence intensity in the nucleus of individual living cells from karyogamy to meiosis I. mutants before karyogamy, with its intensity reaching a peak during the horsetail stage (∼60 min in and Fig. S2). mutants, as determined by Western blot analyses (unpublished data). mutants when compared with wild-type cells (). The total intensity of the GFP signal in wild-type cells was, on average, ∼1.5 times higher than in mutants (, left), even though the cohesin axis was more clearly observed in mutants. mutant cells (, right). As chromatin unbound cohesin subunits increased in the cytoplasm (Fig. S3), their retention within the nucleus may be mediated through binding to chromatin. Thus, this decrease in nuclear Rec8 fluorescence may reflect a reduction in the amount of chromatin bound Rec8 in the absence of Pds5, raising the possibility that Rec8 binding to the chromosome is reduced in the absence of Pds5. To further explore this possibility, we performed chromatin immunoprecipitation (ChIP) of Rec8 in synchronized meiosis induced by inactivation of (), followed by hybridization to a high-density oligonucleotide array (ChIP-chip analysis; ; ). The association of Rec8 with chromosomes II and III was examined. In wild-type cells, Rec8 bound to ∼300 distinct sites, each spanning 2–5 kb, along the arms of the 5.5 Mb of chromosomes II and III (complete DNA array data in Gene Expression Omnibus under accession no. GSE5284). We repeated the experiment twice. Of the Rec8 binding sites, ∼200 sites were detected in both of the two independent experiments. In the absence of Pds5, ∼100 of the Rec8 binding sites were lost in each set of experiments (see for a portion of the array data). Thus, these Rec8 binding sites depend on Pds5 for stable binding to chromatin. In wild-type cells, 83% of the neighboring Rec8 binding sites were within a distance of 2–30 kb (), whereas in the absence of Pds5, this population decreased to 58%, and those at a distance of >30 kb increased (). mutants. These data suggest that loss of Pds5 reduced Rec8 binding to the meiotic prophase chromosome. mutants (). As it is known that Rec8 plays a central role in sister chromatid cohesion (; ), we examined the role of Pds5 in this Rec8-mediated meiotic event. and mutants (, respectively), but the frequency of such events was significantly higher in the double mutant than in either single mutant (). double mutant (), we propose that in the absence of Rec8, Rad21 cooperates with Pds5 to assist sister chromatid cohesion, although Rec8 normally promotes cohesion independently of Pds5. #text The fission yeast strains used in this study are listed in Table S1 (available at ). Strains bearing Rec8- and Psc3-GFP were gifts from Y. Watanabe (University of Tokyo, Tokyo, Japan). Disruption of the gene and construction of the Pds5-GFP fusion gene were performed as previously described (). Psm3-GFP, Rec11-GFP, and Rad21-GFP fusions were constructed using the PCR-based gene targeting method (Bahler et al., 1998), where the ORF of GFP was integrated at the C-terminal end of the endogenous gene locus in the genome. All the GFP fusions showed vegetative growth rates indistinguishable from that of wild-type cells at normal (33°C), lower (20°C), and higher (36°C) temperatures. Spore formation and viability were also comparable with that of wild-type cells. A computer-controlled fluorescence microscope system (DeltaVision [Applied Precision]; ) was used for imaging of live cells. This microscope system is based on an inverted fluorescence microscope (IX70; Olympus) equipped with a charge-coupled device (CoolSNAP HQ; Photometrics). The objective lens used was an oil-immersion lens (Plan Apo 60×; NA = 1.4; Olympus). For time-lapse observation, living fission yeast cells were mounted in a 35-mm glass-bottomed culture dish (MatTek) coated with concanavalin A and observed in EMM2 medium at 26°C. A set of images at 10 focal planes at 0.3-μm intervals was taken at each time point. Image deconvolution was performed using an imaging workstation (SoftWoRx; Applied Precision). For quantitative analysis of GFP signals, a set of images at 10 focal planes at 0.4-μm intervals was taken at each time point. Quantitative projections were generated using an additive image projecting method. On the projected images, 2D polygons () were drawn with an automatically set threshold value using the 2D Polygon Finder in the software, and the sums of the fluorescence intensities in the polygons were obtained. To minimize any error resulting from the progressive decline of the mercury-arc output, we collected datasets from wild-type and mutant cells using the same lamp and within 8 h on the same day. Photoshop 6.0 (Adobe) was used to adjust the linear image intensity (brightness and contrast) for figure production. Chromosomal loci were visualized by the use of a lac repressor (lacI)/lac operator () recognition system; i.e., repeats of the sequence were integrated at a chromosome locus and detected by the GFP-lacI fusion protein (; ). Original constructions for each chromosomal loci used in this work are cited in the strain list (Table S1). For observation of meiosis, haploid cells of the opposite mating type were conjugated on a plate to form a diploid zygote. GFP-labeled chromosomal loci or proteins were observed in living zygotes at 26°C, as described previously (). gene was cloned and inserted at the locus under the control of an inducible promoter, and the cells were cultured in medium without thiamine. Cells carrying the mutation were precultured at 26 or 30°C overnight and then cultured at 36°C for 5 h. The 36°C step was used to inactivate the temperature-sensitive rad21-K1 protein. Protein extracts were prepared as previously described (). To liberate chromosome bound proteins, cell extracts were treated with DNase I at 25°C for 5 min. Immunoprecipitation was performed with anti-Myc mouse antibody (9E10; Santa Cruz Biotechnology, Inc.). Proteins from whole cell extracts and from precipitates were electrophoresed in SDS-polyacrylamide gels and immunoblotted with anti-HA (3F10; Roche) and anti-Myc antibodies. strains carrying the gene were synchronized to meiosis by overnight nitrogen starvation at 26°C, and the culture temperature was shifted to 34°C. After 3 h at 34°C, the cells were fixed with 1% formaldehyde for 30 min at room temperature. ChIP and DNA chip analyses were performed as previously described (; ). In brief, 5 × 10 cells were disrupted using a multibeads shocker. Whole cell extracts were sonicated (250D; Branson) to obtain 400–600-bp genomic DNA fragments. Anti-HA mouse monoclonal antibody (16B12; Babco) coupled to protein A Dynabeads (Dynal) were used for ChIP. The immunoprecipitates were eluted and incubated overnight at 65°C to reverse the cross-linking. The genomic DNA was precipitated, purified, and amplified by PCR using random primers. For ChIP-chip analyses of Rec8-3HA, chromosome II and III tiling array (part 520106; Affymetrix, Inc.) was used. Chip data presented in this paper can be obtained from Gene Expression Omnibus (; accession no. ). Table S1 shows the strain list used in this study and references about the source of the strains. Fig. mutant cells. Fig. or mutant cells. Fig. , and mutant cells. Fig. S4 shows the synchronization of -induced meiosis in the ChIP-chip analysis. mutant cells. Online supplemental material is available at .

The saprophytic fungus is a serious health hazard in hospitals (, ). is responsible for >90% of invasive aspergilloses (IA), with mostly fatal outcomes in immunocompromised patients suffering from AIDS, tuberculosis, cancer, or bone marrow/organ transplants (; ). The true incidence of infections is underestimated because of the inherent difficulty of positive diagnosis (). Second-generation anti-fungals, such as amphotericin B lipid forms, echinocandins, and azoles, have only shown modest improvements in efficacy and at lower toxicity. Thus, anti– therapy remains inadequate and, as a consequence, high morbidity and mortality from IA prevails (; ). One reason for this is our poor understanding of the pathobiology of . A multitude of putative virulence factors, such as extracellular metalloprotease, serine protease, aspartic protease, catalase, phospholipases, haemolysin, and the cytotoxin ASPF1 have been implicated in IA, but none has yet been shown to be involved in the pathogenesis of in experimentally induced infections (). Gliotoxin (GT), an abundant mycotoxin produced by and other fungi, such as , belongs to the epipolythiodioxopiperazine class of secondary metabolites () and is characterized by a reactive disulfide bridge across the piperazine ring (). GT has been proposed to constitute a virulence factor in IA because of its immunosuppressive properties (; ). GT has been shown in vitro to inhibit multiple processes associated with activation, differentiation, and/or effector functions of immune cells (; ; ). This includes activation of NF-κB (), neutrophil and macrophage oxidative killing (; ; ), polymorphonuclear neutrophils chemotaxis (), polymorphonuclear neutrophils and macrophage phagocytosis (; ; ; ), activation of cytolytic T cells (; ), and IFNγ production by CD4 lymphocytes (). Most important, GT was found to induce mammalian cell apoptosis (; ) accompanied by the production of reactive oxygen species (ROS) and mitochondrial membrane disruption (; ; ; ). The proposition that GT is a virulence factor is supported by more recent findings demonstrating that GT is expressed in vivo during experimental and human aspergillosis () and that the decreased levels of pulmonary GT observed with an mutant defective in LaeA, a global regulator of secondary metabolism, is associated with impaired virulence (; unpublished data). However, it should be noted that LaeA regulates expression of a cassette of genes, including GT and other secondary metabolites. Thus, the virulence of the wild-type (wt) strain of is not necessarily linked to GT alone. Definitive evidence on the molecular basis of GT-mediated apoptosis and its relevance in the parasitized vertebrate host is still lacking. Many diverse stimuli, including irradiation, toxic drugs, and pathogens, transduce apoptotic signals to mammalian cells, resulting in the disruption of mitochondrial membranes (; ; ) and the subsequent release of apoptotic factors, such as cytochrome and apoptosis-inducing factor (AIF; ; ; ). Complex formation of cytochrome with Apaf-1 and caspase-9 leads to further induction of effector caspases, i.e., caspase-3 and -7 (). Permeabilizing the mitochondrial membranes by diverse stimuli strictly depends on the proapoptotic Bcl-2 family members Bak and Bax (; ). This is indicated by the fact that deficiency in Bax and Bak renders cells resistant to numerous apoptotic stimuli (, ). Although how Bax and Bak are regulated is still debated, it has become clear that they are somehow activated by BH3-only proteins, which trigger their conformational change, oligomerization, and pore forming activity (; ). Among BH3-only molecules, Bid and Bim may directly activate Bax and Bak by a hit-and-run mechanism (), whereas other members of this family seem to act by antagonizing the survival activity of antiapoptotic proteins, i.e., Bcl-2, Bcl-x, and Mcl-1 (; ; ). However, it has not yet been studied by which mediators/effectors GT induces mitochondrial membrane disruption and apoptosis. We show now for the first time that Bak, which constitutively resides on mitochondria (), is the primary intracellular target in GT-mediated and ROS-facilitated apoptosis in vitro and that in a corticosteroid-based IA model, a knockout mouse strain lacking Bak is more resistant to than wt mice. As GT may induce apoptosis and necrosis in mammalian cells, depending on the concentration of GT and the target cell used (; ; ; ), we first established optimal conditions that allowed us to readily monitor proapoptotic processes in MEFs. As shown in , μM GT induced nuclear fragmentation in the majority of wt MEFs. Furthermore, ∼50% of these cells were apoptotic, i.e., had phosphatidylserine (PS) exposed (annexin V staining) without plasma membrane disruption, whereas the rest of the cells already showed secondary necrosis (loss of membrane integrity as shown by propidium iodide [PI] staining; ; ). Thus, 1 μM GT was used in all subsequent experiments. We next compared GT-induced PS exposure and PI staining between wt and knockout (−/−) MEFs. In addition, we studied the impact of GT on the Δψ and the production of ROS in these cells. As shown in , GT significantly increased the number of annexin V/PI–positive cells in wt, Bax, and Bid MEFs as compared with mock-treated cells, whereas Bak and Bak × Bax MEFs did not. Similarly, GT treatment led to a significant reduction in the mitochondrial membrane potential (Δψ) and a parallel increase in ROS production in wt, Bax, and Bid, but not in Bak and Bak × Bax MEFs (). It is our experience that there is considerable variation in the numbers of cells induced to express the respective proapoptotic markers from experiment to experiment. However, the differentials between experimental and control groups in individual experiments were always highly significant. This suggests that in MEFs, Bak, but not Bax, is critical for GT-induced loss of plasma membrane integrity and the Δψ. Moreover, Bid, which is known to activate Bak and Bax during apoptosis (; ; ), seems to be dispensable for these processes. In support of a key role of Bak in GT-mediated cell death, we found that trypan blue exclusion was significantly reduced only in Bak and Bak × Bax, but not in wt or Bax MEFs (). In contrast to GT-induced apoptosis, the apoptosis-inducing drug staurosporine induced reduction in the Δψ and a parallel increase in ROS, which was only prevented in the absence of both Bak and Bax, as already described (; ). Upon their activation, Bak and Bax undergo conformational changes, leading to the exposure of their N-terminal domains (; ; ). To test whether this process also occurs during GT treatment, wt and Bak × Bax MEFs were incubated with GT and subsequently analyzed by FACS with conformation-specific antibodies against the N termini of Bak or Bax. As shown in , GT was able to readily induce N-terminal epitope exposure in Bak but not in Bax. As expected, no (Bax) or only marginal staining (Bak) was seen with these antibodies in GT-activated Bak × Bax MEFs. There was no inherent failure of Bax to undergo an N-terminal conformational change in MEFs, as their treatment with staurosporine led to the expected N-terminal opening of Bax (; ; ). To determine the order of events during Bak activation, ROS production, and caspase activation, we incubated GT-treated MEFs with the antioxidants N-acetylcysteine (NAC) or the manganese porphyrin Mn(III) tetrakis(4-benzoic acid) porphyrin chloride (MnTBAP; ; ) or the pan-caspase inhibitor ZVAD-fmk (). NAC and MnTBAP were both effective as antioxidants, as they significantly reduced GT-induced ROS production in MEFs (). Moreover, ZVAD-fmk blocked apoptosis induced by the anti(α)-Fas mAb Jo-2 in MBL-2–Fas cells (). GT-induced N-terminal opening of Bak was unaffected by any of the three inhibitors (), indicating that ROS production and caspase activation occur downstream of Bak activation. Furthermore, as similar activation of Bak was seen in GT-treated wt and Bid MEFs, GT-mediated conformational change of Bak is also independent of Bid (). Although the association of ROS generation with GT- or CTL-mediated apoptosis is well documented (; ), the contribution of ROS to cell death is still controversial (; ). We therefore tested the effect of NAC and MnTBAP on GT-induced plasma and mitochondrial membrane integrity and cell death. As shown in , the addition of NAC to wt MEFs at a concentration known to inhibit ROS generation (; ) abrogated GT-induced PS exposure and plasma membrane permeability, as well as the reduction of the mitochondrial Δψ. Moreover, at 1 μM GT, NAC totally inhibited cell death (according to the absence of trypan blue staining; ). Similar results were obtained with MnTBAP, indicating that the effect of NAC was due to its antioxidant properties (unpublished data). These data reveal that ROS generation is crucial for GT-induced changes of plasma membrane permeability, the mitochondrial Δψ, and cell death. To determine the role of caspases in the GT-induced reduction of the mitochondrial Δψ and apoptosis, we tested whether GT could induce cell death and changes in mitochondrial Δψ in the presence of the broad spectrum caspase inhibitor ZVAD-fmk. 100 μM ZVAD-fmk did not prevent loss of the mitochondrial Δψ in GT-treated cells, indicating that caspase activation was not needed for this event. However, ZVAD-fmk partially reduced GT-induced annexin V/PI staining () but did not affect cell death (). These data also exclude a possible role of death receptor–mediated cell death, which is completely blocked by caspase inhibitors (; ). To test whether the major downstream effector caspase, caspase-3, was involved in these processes, and whether activation of this caspase was dependent on Bak, we performed a FACS analysis of wt and −/− MEFs using an α–caspase-3 mAb specific for the processed active form of caspase-3. As shown in , caspase-3 was significantly activated in GT-treated wt and Bax but not in Bak and Bak × Bax MEFs. In contrast, staurosporine induced caspase-3 activation in Bak and Bax but not in Bak × Bax MEFs (), again showing that both Bak and Bax can undergo conformational activation upon appropriate stimulus (). The activation of Bak and the subsequent production of ROS and the reduction of mitochondrial Δψ in response to GT indicate increased mitochondrial membrane permeability, leading to the release of apoptogenic factors such as cytochrome and AIF. Although cytochrome activates caspase-3 via the apoptosome (), AIF translocates to the nucleus and contributes to DNA fragmentation in a caspase-independent manner (; ). The release of both cytochrome and AIF from mitochondria is absolutely dependent on activation of Bax, Bak, or both (; ). To test whether GT-induced cytochrome release selectively required Bak, wt, Bak, Bax, or Bak × Bax MEFs were incubated with GT and mitochondrial cytochrome was quantitatively measured by FACS analysis (). In addition, AIF release was monitored by α-AIF immunofluorescence (). As shown in , mitochondrial cytochrome was reduced in GT-treated wt and Bax MEFs but retained in Bak or Bak × Bax MEFs. In addition, a high portion of GT-treated wt and Bax MEFs displayed cytosolic localization and nuclear translocation of AIF, whereas Bak and Bak × Bax MEFs retained most of the AIF in the mitochondria (). These data confirm that GT induced mitochondrial membrane permeability; i.e., cytochrome and AIF release occurs by a process selectively involving Bak. To test whether GT could directly act on mitochondrial Bak without the requirement of any cytosolic factors, such as, for example, a particular BH3-only protein, we compared GT-induced cytochrome release on isolated mitochondria from wt MEFs with those of knockout MEFs and factor-dependent myeloids (FDMs). As shown in , 10–50 μM GT caused cytochrome release from isolated wt mitochondria in a dose-dependent manner. The requirement for much higher concentrations of GT to induce cytochrome release in isolated mitochondria as compared with intact cells is in line with a previous study in which GT-mediated calcium release was analyzed (). In fact, it is known that levels of GT determined intracellularly do exceed those originally applied in solution by up to 1,500-fold (; ). The release of cytochrome was as efficient as that induced by recombinant tBid, a known inducer of mitochondrial membrane permeability via Bak/Bax (, ) and significantly greater than the background release of cytochrome observed in mock-treated mitochondrial preparations. Strikingly, although mitochondria from Bax showed similar GT-induced cytochrome release as wt mitochondria, mitochondria from Bak or Bak × Bax did not, irrespective of whether they were derived from MEFs or FDMs. This suggests that GT does not need cytosolic factors such as tBid or caspases to elicit cytochrome release but may act directly on mitochondrial Bak or some as-yet-unknown mitochondrial protein that activates Bak with a potency similar to tBid. We excluded the possibility that the resistance of GT-facilitated cytochrome released could be due to the absence of Bax on mitochondria, as appreciable amounts of this protein were detected on isolated, washed mitochondria from both wt and Bak FDMs ( ii). Moreover, tBid was capable of inducing cytochrome release from mitochondria isolated from both Bak FDMs and MEFs. This would not have been possible if they had no Bax. Consistent with a direct action of GT on mitochondrial membrane permeability, GT-induced cytochrome and AIF release was not blocked by ZVAD in wt MEFs ( and ). Most important, however, both cytochrome and AIF release from mitochondria was greatly diminished by NAC ( and ). This suggests that GT-induced ROS production is crucial for effective release of apoptogenic factors from mitochondria but does not exclude the possibility that NAC has some other mitochondrial membrane stabilizing activity. To determine whether the selective activation of Bak during GT-mediated cell death as seen in our in vitro analysis is of biological relevance and pathophysiological significance, we monitored the mortality of immunosuppressed (hydrocortisone-treated) wt (C57BL/6) and Bak ko (Bak) mice subsequently infected with a GT-producing strain. The presence of GT synthesis by this fungal strain was verified by HPLC analysis of culture SN (). shows that 5 out of 6 wt mice died within 2 wk after intranasal infection, and only 1 out of 6 Bak mice succumbed over the same time period (P < 0.015). These in vivo data correlate with our in vitro findings and show for the first time that Bak is a host susceptibility factor for virulence in mice, probably because of its direct activation by GT. Although GT has long been proposed to constitute a virulence factor in IA (; ), most probably by suppressing immune responses via induction of mammalian cell apoptosis (; ), the molecular mechanisms underlying the putative in vitro and in vivo processes have not been elucidated. Here, we present evidence that Bak, but not Bax, is a key host factor in GT-mediated cell death in vitro. We used MEFs and FDMs deficient in the Bcl-2 family members Bak and/or Bax or their activator Bid () and isolated mitochondria from these cells to show that GT-mediated activation of Bak occurs independently of Bid or other cytosolic factors. Once activated, Bak triggers the generation of ROS, which is crucial for effective mitochondrial membrane pore formation, including the release of cytochrome and AIF, and ultimate cell death. The additional finding that the virulence of GT-producing was significantly decreased in Bak over wt mice strongly implicates that GT is an important modulator in mammalian host defense and that Bak is a prominent host susceptibility factor. The interrelation of GT and Bak in pathology is also supported by the recent finding that an mutant lacking GT expresses a drastically reduced virulence (unpublished data). At present, it is unclear how GT activates Bak. One mechanism by which GT may activate Bak is by breaking up inhibitory complexes between Bak and Bcl-2–like prosurvival factors (; ; ; ) or VDAC2 () on the mitochondrial membrane. This could be by forming transient disulphide bonds between the reactive disulphide bond in GT and individual cysteine residues in Bak or its binding partners, leading to the release of active Bak. Three-dimensional structures of Bak–Bcl-2 or Bak–VDAC2 complexes have not yet been determined, so we do not know if any cysteine residues are involved in these interactions. Moreover, we do not have any experimental evidence that GT directly binds Bak, Bcl-2–like proteins, or VDAC2. However, our data obtained with isolated mitochondria favor an interpretation that GT-facilitated activation of Bak occurs by direct interaction with antiapoptotic Bcl-2 family members or other mitochondrial membrane–associated constituents. The former possibility is supported by the observation that protection against GT-mediated monocyte apoptosis by agonists of nerve growth factor receptors is associated with the up-regulation of Bcl-2 and Bcl-x (). The finding that GT specifically acts through Bak and not Bax is intriguing. Although both proteins are supposed to exert the same pore-forming activity on mitochondria (), they are activated differently. There is increasing evidence for selective Bax- or Bak-specific apoptosis, depending on the cell type and the apoptotic stimuli (; ). Thus, the activation mechanism of Bax and Bak may be distinct, although both ultimately oligomerize and form pores in the outer mitochondrial membrane. In this respect, GT may be unable to interact with Bax and/or to release any of its inhibitory components. Moreover, an interaction of GT with Bcl-2 or Bcl-x would not affect Bax because it is not sequestered by these proteins in healthy cells. This would explain why GT induces conformational activation of Bak but not of Bax. The findings that the mitochondrial protein VDAC2 associates with and inhibits Bak in healthy mitochondria () and that in monocytes, Bak but not Bax is part of the VDAC channel (unpublished data) suggest VDAC2's involvement in GT-mediated cell death. VDAC2 is one of three mammalian isoforms of VDAC proteins (VDAC1, -2, and -3), which constitute the major pathway for metabolic exchange across the outer mitochondrial membrane (; ; ). Together with cyclophilin D and adenine nucleotide transporter (ANT), VDAC forms the mitochondrial permeability transition pore (MPTP), involved in cell apoptosis and/or necrosis (; ). How the function of MPTP is regulated by members of the BH3 family is still highly controversial (; ; ). One could postulate that GT somehow modulates the VDAC complex, leading to the liberation of Bak, a subsequent increase of mitochondrial membrane permeability and hence a Bak-dependent cytochrome release and cell death. The contribution of the MPTP in the latter process is further supported by the findings that GT-induced apoptosis of activated hepatic stellate cells is associated with a specific thiol redox-dependent interaction with MPTP component ANT () and that cyclosporin A, an inhibitor of cyclophilin D and mitochondrial pore opening (), affected mitochondrial depolarization and ROS production (; unpublished data). Most notable, the data suggested that oxidative cross-linking of two matrix-facing cysteine residues on the ANT (Cys and Cys) plays a key role in regulating the MPTP (). However, more detailed studies, including MPTP inhibitors such as bongkrekic acid or cyclosporins A, are required to dissect the role of the VDAC–ANT complex in GT-mediated and Bak-dependent cell death. Our data further show that GT-induced production of ROS is mandatory for cell death. The sequence of events leading to ROS production by GT was revealed by analyzing the various proapoptotic processes in the presence of inhibitors for ROS and for caspases, including NAC, MnTBAP, and ZVAD-fmk, respectively (). Accordingly, activation of Bak precedes the generation of ROS, which then facilitate the release of cytochrome and AIF from mitochondria, leading to caspase activation as well as mitochondria- and caspase-independent events to mediate cell death. As to the source of ROS, it is possible that they are generated from a perturbance of mitochondrial respiration that is due to Bak-mediated pore formation and/or activation of MPTP. Why ROS are, at least partially, required for cytochrome release is unclear, although it has been shown that ROS generation is crucial for cytochrome release under different apoptotic stimuli (; ; ). The putative relevance of GT-mediated apoptosis for the parasitized vertebrate host was analyzed by comparing the course of infection in wt and Bak mice. The significantly decreased virulence of the pathogen observed in Bak as compared with wt mice, as revealed by the differential kinetics of mortality, supports the following assumptions: GT is released during infection, as suggested before (; ), and induces apoptosis in multiple target cells, most probably via Bak activation. This process subsequently leads to an accelerated colonization of target organs by breaching physical barriers, such as lung and renal epithelial cells, and establishes an immunosuppressed state of the host. Although the present data do not formally proof a cause–effect relationship between GT, Bak activation, and pathogenicity (virulence), the previous (; ) and present assumption that GT is a virulence factor of is supported by a recent report () and our own unpublished data. These results have shown that low levels of pulmonary GT observed with an A. fumigatus mutant defective in LaeA, a global regulator of secondary metabolism, is associated with impaired virulence of the pathogen. Furthermore, by using a recently generated gene knockout mutant of lacking GT, we found that this mutant is much less virulent in mice than the wt strain and that cell culture supernatants were unable to induce cell death (unpublished data). Based on the sequence of intracellular events occurring during GT-induced apoptosis (), we conclude that GT is a critical virulence factor in . This is supported by the fact that GT is one of the most abundant secondary metabolites produced by the fungus () and that Bak mice are more resistant to infection by . The distinct potential of GT to activate Bak, but not Bax, may be of relevance for the development of anti-IA drugs that selectively block cell death pathways via Bak and, at the same time, spare the residual proapoptotic proteins relevant for the control of the pathogen by the host's immune system. SV40 transformed MEFs () and MBL-2–Fas cells were cultured in MEM supplemented with 10% FCS and 2-mercaptoethanol (10 M) at 37°C and 7% CO. The IL-3–dependent (FDM) cell lines were generated by coculturing embryonic day 14.4 fetal liver single-cell suspensions with fibroblasts expressing a HoxB8 retrovirus in the presence of high IL-3 concentrations, as previously described (). Bak mice were obtained from C. Thompson (Harvard Medical School, Boston, MA), backcrossed for nine generations to ensure a “pure” C57BL/6 genetic background, and intercrossed with Bax C57BL/6 mice to obtain Bax × Bak mice (provided by D. Huang, The Walter and Eliza Hall Institute, Melbourne, Australia) as described previously (). The cell lines were cultured in MEM with Earle's salts and -glutamine. For Western blot analysis of isolated mitochondria, antibodies to H-ATPase (Invitrogen) were used as controls. GT was purified from as described previously (). The purity of this preparation was analyzed by TLC and HPLC showing the same quality as commercial GT. For apoptosis induction, 2 × 10 cells/ml MEFs were incubated with the indicated concentration of GT or staurosporine (Sigma-Aldrich) for 4 h, and apoptosis assays were performed as described in the following paragraphs. In some cases, the general caspase inhibitor Ac-ZVAD-fmk (Bachem) or the ROS scavengers NAC (Sigma-Aldrich) or MnTBAP (Calbiochem) were added as described previously (). To test the inhibitory potency of Ac-ZVAD-fmk, MBL-2–Fas cells were incubated with 1 μg/ml α-Fas mAb Jo-2 for 24 h in the presence or absence of 100 μM of the caspase inhibitor, and cell death was analyzed by trypan blue exclusion. Nuclei were stained with 10 μg/ml Hoechst 33342 (Invitrogen). PS exposure and PI uptake was analyzed by FACS or fluorescence microscopy as described previously () using the annexin V–FITC kit from BD Biosciences. Δψ was measured with the fluorescent probe 3,3′-dihexyloxacarbocyanine iodide (DiOC; Invitrogen) and ROS generation with 2-hydroxiethidine (2-HE; Invitrogen) as described previously (). Nuclear morphology was analyzed by fluorescence microscopy with Hoechst 33342. For that, cells were fixed with 1% PFA and mounted on a drop of Fluoromount-G (Southern Biotechnology Associates, Inc.) containing 10 μg/ml Hoechst 33342, and images were taken at room temperature using a microscope (Axioskop 10; Carl Zeiss MicroImaging, Inc.), an analysis camera (Axiocam; Carl Zeiss MicroImaging, Inc.), and Vision 3.1.0.0 software (Carl Zeiss MicroImaging, Inc.). The objective used was a PlanNeofluor (Carl Zeiss MicroImaging, Inc.), with a magnification of 40 and a NA of 0.75. Photoshop CS2 (Adobe) was used for minor adjustments to contrast and image overlay. In some cases, membrane perturbation was also analyzed by fluorescence microscopy by staining the cells with the annexin V–FITC kit from BD Biosciences. Images were taken in the same conditions as described above, but cells were mounted on annexin binding buffer (BD Biosciences). Cells were fixed with 2.5% PFA, incubated with a mAb FITC-labeled against the active form of caspase-3 (clone ; BD Biosciences), and analyzed by FACS as described previously (). Cytochrome release was quantified by FACS analysis as recently described (). In brief, 10 MEFs were mildly permeabilized with 25 μg/ml digitonin plus 100 mM KCl on ice for 5 min. This led to the cellular loss of cytosolic cytochrome . Cells were washed once with cold PBS, fixed in 4% PFA, permeabilized with 0.05% saponin and 3% BSA, and incubated with the α–cytochrome mAb 6H2.B4 (BD Biosciences) or mouse IgG isotype control (Jackson ImmunoResearch Laboratories) followed by α-mouse-FITC secondary antibody (Jackson ImmunoResearch Laboratories). The cells were resuspended in 100 μl PFA in PBS and analyzed by FACS with a FACScan (BD Biosciences) and CellQuest software (BD Biosciences). For the analysis of the nuclear translocation of AIF, cells were fixed, mounted on poly--lysine cover slides, stained with a rabbit polyclonal α-AIF antibody (Sigma-Aldrich) followed by the secondary goat α-rabbit antibody labeled with Alexa 488 as described previously (), and mounted on a drop of Fluoromount-G. Afterward, the cells were analyzed by confocal microscopy. Fluorescence images were taken at room temperature on a confocal microscope (TCS SP2; Leica) using a 40× objective (HCX PL APO CS; Leica), NA 1.25, immersion oil, and confocal software (version 2.61; Leica). Photoshop CS2 was used for minor adjustments to contrast. 8 × 10 MEFs or FDMs were centrifuged and washed once in PBS. The cell pellets were resuspended in 500 μl MSH buffer (210 mM mannitol, 70 mM sucrose, 20 mM Hepes, 1 mM EDTA, pH 7.5, 100 μM PMSF, 400 ng/ml pepstatin, 10 μg/ml leupeptin, 10 μg/ml aprotinin, and 5 μg/ml cytochalasin B). The resuspended cell pellet was incubated on ice for 15 min before the cells were broken by passaging 25 times through a 23-gauge needle. The lysate was centrifuged at 500 for 5 min to remove cell debris and nuclei. A crude mitochondrial pellet was then obtained by centrifugation at 10,000 for 15 min and resuspended in MSH buffer. The isolated mitochondria were incubated with different concentrations of GT (10, 20, and 50 μM) or 40 nM of recombinant tBid (provided by J.C. Martinou, University of Geneva, Geneva, Switzerland) as a positive control at 37°C for 4 h. After incubation, the mitochondria were pelleted, and both pellet and supernatant were tested for cytochrome release by SDS-PAGE. MEFs were fixed in 4% PFA, permeabilized with 0.1% saponin in PBS/5% FCS, and incubated with 2 μg/ml rabbit polyclonal α-Bak (NT; Upstate Biotechnology), 5 μg/ml rabbit polyclonal α-Bax (NT; Upstate Biotechnology), or 5 μg/ml rabbit purified IgG (control). After two washes with 0.1% saponin in PBS, the cells were incubated with α-rabbit-FITC antibody in 0.1% saponin/PBS/5% FCS, washed twice in 0.1% saponin/PBS, resuspended in 1% PFA/PBS, and analyzed by FACS with a FACScan and CellQuest software. The amount of Bax and Bak on isolated mitochondria from wt, Bak, Bax, and Bak × Bax FDMs was determined by washing the centrifuged mitochondria twice in large amounts MSH buffer (to eliminate cytosolic contamination) followed by lysing the mitochondria in SDS sample buffer and analysis by α-Bax (Bax-NT) and α-Bak (Bak-NT) Western blotting on the same gel. As mitochondrial marker and loading control, an antibody against the F0F1 ATPase was used. Female mice (C57BL/6, Bak [Jackson ImmunoResearch Laboratories; C57BL/6.129, six times backcrossed in C57BL/6], or 129/Sv) were immunosuppressed by subcutaneous injection of 3 mg (112 mg/kg) of hydrocortisone (Sigma-Aldrich) diluted in 200 μl of PBS/0.1% Tween 20 on days −4, −2, 0, 2, and 4, as described previously (). On day 0, mice (6 per group) were infected intranasally with 5 × 10 B5233 conidia in 20 μl of PBS or with PBS alone. Disease development was analyzed by morbidity/mortality of the mice after infection. There was no difference in the sensitivity of C57BL/6 or 129/Sv mice to infection, and all infected recipients of both mouse strains died during the first week after infection. GT presence on B5233 culture supernatants was analyzed after 48 h by HPLC as described previously ().

Ca is a versatile and ubiquitous messenger in all living cells. Increases in the intracellular Ca concentration occur as long-lasting global oscillations, such as those that occur during clonal selection of white blood cells. In contrast, Ca increases also occur as transients with a very short lifetime and high spatial confinement, so-called elementary Ca signals (; ). All Ca signals require proteins to relay the Ca concentration downstream to different signaling networks. These specialized sensors contain Ca-binding domains and include such ubiquitous proteins as calmodulin, the structure of which comprises four highly conserved binding domains, the so-called EF hands (). Another very important Ca-binding domain is the C2 domain; e.g., during synaptic signaling, where it functions as the Ca-sensing domain of Munc-13 (). A subfamily of the PKCs, the conventional PKCs (cPKCs), also contain C2 domains and, hence, have the principle ability to decode cellular Ca transients (). This family belongs to the AGC kinases () and comprises 12 isoforms, which are distinguished depending on their sensitivity toward signaling molecules. The intrinsic function of PKC isozymes is regulated by multiple mechanisms (), and they decode a variety of stimulation-induced intracellular messengers (). PKCs are believed to translate messengers into the phosphorylation of many target proteins, such as ion transporters (e.g., Na/Ca and Na/H exchangers; ; ) and proteins involved in the control of cell growth and proliferation (e.g., MAPK, Ras, and Raf; ; ). Phosphorylation assays with target proteins provide one possible way to detect PKC activity, but these approaches might be complicated by the fact that many signaling events occur only in confined subcellular spaces, the so-called signaling microdomains, and/or display a brief lifetime. Another approach is to follow translocation events in cells. Most of these experiments were performed by using long-lasting stimuli such as fatty acids () or phorbolesters (e.g., phorbol 12-myristate 13-acetate [PMA]; ). These methods were typically combined with an in vitro readout at predefined time points during the stimulation process; thus, interpretation at the single-cell level was complicated by cell–cell variability and limited temporal resolution. In contrast, real-time live-cell imaging overcomes many of these drawbacks. For this, fluorescent proteins of various colors are coupled to the C termini of PKCs, and translocation is followed by video or confocal imaging (). With such an approach, presented evidence for diffusion-limited fast translocation kinetics of PKCα in histamine receptor–overexpressing human embryonic kidney (HEK) 293 cells (). To follow these events with appropriate temporal resolution, these authors used confocal linescanning by sacrificing one spatial dimension for a gain in acquisition speed. In contrast, used total internal reflection microscopy imaging and found fast and robust responses of the neuronal cPKCγ. In addition, they could provide some initial evidence for spatially more complex translocation events, but these studies were restricted by their limited temporal resolution of several seconds per image and their confinement to PKC measurements at the plasma membrane level. We have used 2D real-time confocal microscopy of fluorescent PKCα fusion constructs in two cell types to comprehensively characterize cPKC translocation dynamics. The α-isoform was chosen because it is the most ubiquitously expressed PKC isoform and, in contrast to other ubiquitous cPKCs (e.g., PKCβ), there is only a single variant (). We particularly concentrated on the question of whether and to what degree cPKCα can decode the spatially and temporally very complex Ca transients that occur during physiological Ca signaling in living cells () into translocation signals and on PKCα membrane interactions via their C2 and C1 domains. Physiologically, PKCα is stimulated by diacylglycerol (DAG) through binding to its C1a domain (), which is produced, e.g., as a product of PLCβ-mediated hydrolysis of phosphatidyl-4,5-bisphosphate (PIP), and by Ca through its C2 domains. To activate these signaling networks, we used ATP stimulation of HEK293 and COS1 cells because they endogenously express P2Y receptors that couple to G proteins. In the vast majority of cells stimulated with robust agonist concentrations, we found global oscillations of both Ca and PKCα-EYFP (Fig. S1, available at ), and variations of the agonist concentration were readily translated into graded Ca signals and PKCα-EGFP translocations (Fig. S1 and Video 1). Stimulation of cells with lower concentrations of Ca-mobilizing agonists often leads to Ca transients with complex spatiotemporal properties, such as waves or spatially restricted Ca responses (for review see ). Thus, we questioned to what degree PKCα is able to decode the complexity of Ca signals into translocation events. depicts an analysis of the translocation process evoked by a Ca wave using rapid 2D confocal imaging. Regularly, linescans or slow 2D image acquisition can produce data that are difficult to interpret (). The Ca wave underlying the response shown in brought about a wave-like translocation of PKCα-EGFP that can be seen in the pseudocolored self-ratio images (, left column). To selectively depict the spatial properties of such translocation waves on the whole-cell level, we have plotted the relative fluorescence changes as increases or decreases in the height of 3D surface for the entire 2D plane (, middle and right columns; and Video 2, available at ). From such data, we concluded that PKCα is able to reliably decode even fast Ca waves propagating through cells into translocation waves. Low agonist concentrations provoke spatially complex Ca signals (). The investigation of subcellular Ca signals requires Ca indicators with a good signal-to-noise ratio, especially for small and spatially confined Ca signals (). In PKCα-DsRed2–expressing HEK293 cells, we analyzed responses to the application of different lower ATP concentrations to provoke such Ca signals. depicts the result of such a representative experiment. Four regions of interest (ROIs; ) have been analyzed, and the time courses of the normalized fluo-4 (green), plasma membrane (blue), and cytosolic (red) fluorescence of the DsRed2 constructs have been replotted in . Application of 5 μM ATP caused a Ca signal that was largely restricted to the lower extension of the cell (, compare traces 1 and 4). The subcellular distribution of this Ca transient can be seen in the self-ratio image displayed in (second column). Despite the fact that this Ca signal was brief and spatially restricted, the PKCα translocation reflected that spatiotemporal property (compare traces in with the corresponding images in C). A detailed analysis of the fluo-4 fluorescence and the plasma membrane PKCα-DsRed2 fluorescence revealed a small difference between the time courses in that the fluo-4 fluorescence increase preceded the increase in the plasma membrane PKCα-DsRed2 signal by ∼800 ms (i.e., four image pairs at a 5-Hz acquisition rate; in all nine experiments with the same stimulation regime the differences were always between three and four images at 5 Hz). In addition, we observed multiple small local Ca fluctuations after the initial spike (, traces 1 and 2, arrows) that were not reflected in translocation events. The application of 25 μM ATP caused a Ca wave covering the entire cell (, B and C, third and fourth columns). From these data we conclude that PKCα translocation is able to follow Ca signals with small amplitudes, brief lifetimes, and spatial restrictions. During experimental series in which we stimulated PKCα-EYFP–expressing COS1 cells with various lower ATP concentrations (5–25 μM), we found that ∼15% of all COS1 cells gave spatially and temporally very complex responses, such as those exemplified in . shows relative fluorescence traces from six ROI pairs outlined in . These tracings illustrate very short-lived and highly confined increases in the PKCα-EYFP fluorescence (, traces 1–5). The individual “hot spots” were spatially independent of one another, but their activity increased, with a variable delay, after ATP application and ceased after ATP washout (unpublished data). As illustrated in the inset of (b), the confocal slice was positioned close to the bottom plasma membrane level of this COS1 cell; thus, we recorded PKCα-EYFP fluorescence close to or at the lower plasma membrane. Therefore, the transient increases in the PKCα-EYFP fluorescence were caused by spatially restricted translocation signals to the plasma membrane (local translocation events [LTEs]). LTEs were not restricted to the bottom plasma membrane; similar signals could also be observed at the upper plasma membrane (unpublished data). The detection of LTEs failed either when we increased the ATP concentration or when we moved the confocal section more toward the central region of the cell (unpublished data). Such a loss was not caused by the temporal resolution of our confocal microscope because with charge-coupled device camera–based fluorescence acquisition we do not miss individual events; instead, they might only be blurred over time. Once a cell like the one displayed in was found repetitive, stimulations with the same ATP concentration after 10–15-min resting periods evoked similar LTE patterns, although not all LTE locations could be restimulated and new ones were additionally triggered during the second ATP stimulation (). In the following, we analyzed a larger population of LTE sites and signals ( = 160 from 60 locations; 18 cells). The results of this analysis are shown in . The distribution of the normalized occurrences of spatial spread (), lifetime (), and amplitude () have been plotted. From these data, it becomes evident that for all of the parameters analyzed the LTEs can be subdivided into two populations. To investigate the relationship between these parameters, we plotted the amplitude of the LTEs against their individual lifetime (). From this graph the following major findings were identified: (a) long and small events were absent, (b) the amplitudes of LTEs did not exceed 150% above resting values, and (c) the two LTE populations found in the individual distributions were also found in . To visualize the spatiotemporal properties of these two populations of translocation signals, we plotted typical LTEs with long lifetimes (>6 s; and Video 3, available at ) and shorter durations (). exemplifies two LTEs in close spatial proximity (). Despite continuous ATP stimulation, LTEs can display rather complex time courses (). For four selected time points, the PKCα-EYFP fluorescence distributions were plotted as 3D surface plots in (b). The 3D surface representation of a pseudolinescan () revealed that the spatial spread appears to be functionally limited, as depicted for the long-lasting LTE (black arrow). After an initial 3-s growth period, the spatial spread came to a standstill, despite ongoing translocation activity. This finding was typical for all long-lasting LTEs we investigated. From these findings, and from those in (i.e., two populations of LTEs), we conclude that long-lasting LTEs are characterized by limited amplitudes and spatial spreads. The C2 domain of cPKC is supposed to mediate the Ca-dependent portion of the cPKC membrane interaction (; ), whereas the C1a and C1b domains are thought to be responsible for DAG and PMA binding, respectively (; ). To investigate the contributions of the C2 or the C1a domains to the translocation of cPKCs and their membrane interactions, we have constructed two PKCα-EGFP fusion proteins lacking a functional C2 and a C1a domain. depicts results obtained with the wild-type (WT) PKCα-mRFP and the C2-mutated (C2-mut) PKCα-EGFP fusion protein. Although the distribution of the two constructs in resting cells was indistinguishable (), the ATP stimulation revealed protein translocation only for the WT construct (; PMA provoked translocation of both proteins). Thus, the mutation we have introduced into the C2 domain specifically suppressed the ATP-induced, but not the PMA-induced, translocation. Because ATP application leads to the generation of two second messengers that interact with the cPKC, i.e., Ca (C2 domain) and DAG (C1a domain), we designed experiments that allowed us to distinguish between these two remaining possible membrane interactions. For , HEK293 cells coexpressing the two fusion protein constructs were loaded with NP-EGTA–AM, which is a membrane-permeable caged Ca. Upon a bright UV flash, NP-EGTA rapidly releases Ca () and, thus, specifically generates Ca signals while avoiding the production of DAG. Two UV flashes in succession triggered the translocation of the WT protein to the plasma membrane (, red traces; compare top images in ), whereas the C2-mutated cPKC did not display any change in its subcellular distribution (, green traces; compare also the bottom images in ). The ATP application at the end confirmed the translocation properties already described for . For the aforementioned LTEs with lifetimes of <1 s, the translocation speed of cPKC fluorescence proteins has to be very fast. In , we addressed the question of how fast the translocation of the PKCα fusion proteins is in response to a stepwise increase in the intracellular Ca concentration. Cells were coloaded with NP-EGTA and fluo-4, and a series of four consecutive UV flashes was delivered. The first two flashes induced only small step changes in the Ca concentration (), causing no apparent or only a minute (second flash) translocation of the PKCα construct. Flashes 3 and 4 triggered substantial increases in the intracellular Ca concentration that were accompanied by rapid translocations toward the plasma membrane (). To characterize the speed of translocation, we analyzed consecutive images acquired with a frequency of 4 frame pairs/s. depicts individual images of the fluo-4 ratio (left) and the relative differential changes in the PKCα-mRFP fluorescence (right) before () and directly after the fourth UV flash (; note that both the fluo-4 and mRPF images were normalized to pre–fourth UV flash fluorescence intensity). Within <250 ms, the cPKC fusion proteins accumulated at the plasma membrane level (see the green outline of the cell in the , right). The Ca concentration had increased in a stepwise manner and remained constant from (left column). From this, we concluded that (a) cPKC translocation required a certain threshold Ca concentration and (b) that cPKC translocation to the plasma membrane can occur within <250 ms. Thus, cPKC translocation to the plasma membranes appeared fast enough for the LTEs recorded, but is the dissociation of the cPKC–membrane interaction also fast enough to cause the fast relaxation seen for the LTEs? What contribution do the C2 and C1a domains make to the membrane residence time or dissociation kinetics? To address these questions, we expressed WT PKCα-EGFP and coloaded the cells with Fura red and diazo-2, which is a caged Ca buffer (). For these experiments, we restricted the UV illumination during the UV flash to only part of the field of view to simultaneously record cellular responses with and without the UV flash. For the cell depicted in , we applied a UV flash ∼10 s into the ATP stimulation. This caused a rapid release of a high- affinity Ca buffer ( = 73 nM) and, consequently, the Ca concentration quickly dropped to resting values (, red trace and ). After this, the cytosolic PKCα-EGFP fluorescence increased (, black trace), whereas the fluorescence at the plasma membrane level decreased (, green trace). In (b), the black and orange arrows point to a fast and slow component of this decline, respectively. The transition between these two phases was present in all cells analyzed with the same experimental regime (15 cells in five independent experiments), but the fractional contribution of each phase to the total PKCα dissociation process varied between 30 and 60%. From these experiments and from the two populations of LETs we concluded that, indeed, these two time constants of membrane dissociation represent two PKCα pools at the plasma membrane: (a) Ca-C2–bound PKCα causing the fast component (, black arrow) and (b) binding of PKCα via its Ca-C2 and DAG-C1 domains causing the second, slower component (, orange arrow). C1 domain–mediated membrane interactions are much more stable and have a longer lifetime (according to in vitro data; ). To test this hypothesis, we characterized the effect of a functional C1a mutation on the dissociation kinetics of PKCα from the plasma membrane (). For this, HEK293 cells expressing C1a-mutated PKCα-EGFP were loaded with Fura red and diazo-2. The basic experimental design was the same as described for . Stimulation of the cells with 100 μM ATP resulted in a robust and reversible translocation to the plasma membrane without apparent differences to the WT construct (compare ). Illumination of the cells with a UV flash caused a rapid drop in the intracellular Ca concentration (, red trace) that induced an increase in the cytosolic PKCα(C1a-mut)-EGFP fluorescence back to prestimulation levels (, black trace). When we analyzed the time course of membrane fluorescence, we found that the fluorescence decay after the UV illumination was only monophasic (, black arrow), the kinetics of which was similar to the kinetics of the fast component of the WT PKCα (compare green traces in ). This lack of a slower decay phase constant was a consistent finding in all cells expressing the mutated cPKC ( = 7 cells in four independent experiments). From this we conclude that the slower dissociation time constant seen in the cells expressing the WT PKCα-EGFP was most likely caused by C1a-mediated membrane interactions, whereas the faster decay was dependent on a functional C2 domain in the protein (). Signaling cascades using the stimulation of PLCβ result in the production of the InsP and DAG that activate InsP receptors and PKCs, respectively, with Ca release from the ER triggered by InsP. A tight coupling between intracellular Ca signals and PKCα translocation was proposed recently (). By using the physiological agonist ATP and flash photolysis of caged compounds, we were not only able to provide strong supporting evidence for such an idea but also revealed various pools of PKCα at the plasma membrane. Our findings suggest fast and tight mechanisms for decoding incoming stimuli into translocation signals. Nevertheless, the molecular mass of conventional PKCs (PKCα, ∼82 kD) gives rise to principle doubts as to whether such a protein will be able to follow very fast and brief Ca signals (lifetime, <1 s) observed during cellular Ca signaling. Although the fluorescence tags of the PKC constructs added another 27 kD to the molecule's molecular mass, the Stokes radius of the fluorescent PKCα only increased slightly (from 42 to 46 Å; ). We showed that fluorescent PKCα fusion proteins display graded translocations, which are similar to those reported for glucose stimulation and PKCβ (). Further establishment of such a notion appears important in the context of PKC-mediated signal transduction because interpretation of amplitudes adds a new dimension to the parameter space that cPKCs can exploit and cells can use to decode incoming stimuli. A similar ability in decoding Ca signals has been suggested for calmodulin (). Cells do not only use the frequency and amplitude domains to decode incoming stimuli into cellular responses, but spatiotemporally complex Ca signals have also been described, ranging from global Ca waves to spatially restricted Ca waves and elementary Ca signals, such as Ca sparks and Ca puffs (for review see ). The importance of such complex signaling processes for cellular signal transduction has been suggested only in an indirect way, by linking local Ca signaling to cellular responses or by electrophysiological measurements that allow direct readouts of, e.g., Ca-dependent membrane ion conductance in smooth muscle cells (; ). Nevertheless, it appears important to present direct evidence and visualization of spatiotemporally very complex processes linking local Ca signaling to other signal transduction mechanisms. We provide such evidence in that the Ca signaling toolbox, from waves to spatially restricted Ca signals and elementary Ca release events, can indeed be decoded into PKCα translocations to the plasma membrane. Assuming similar properties for all cPKCs, it appears feasible to extend such behavior to the entire cPKC subfamily. An important requirement for such mechanisms is a rapid translocation, despite the substantial molecular mass of our constructs (∼119 kD). With real-time imaging of PKCα translocations after UV flash–induced Ca increases, we demonstrated sufficiently fast translocations kinetics (). It should be noted that not every elementary Ca signal will necessarily result in local cPKC translocation because the k and k rates of Ca binding are very fast, and translocation also requires a certain Ca threshold concentration. Ca release deeper in the cell will lead to cPKC Ca binding, but unbinding of the ion will occur before the cPKC can reach the membrane and encounter PSs and DAG. Thus, only peripheral Ca signals or, in other words, only the peripheral component of Ca signals, will lead to a significant cPKC translocation. Thus, this is another example of the versatility of Ca signals solely depending on their particular location inside cells. Although peripheral transients can induce, e.g., phosphorylation of membrane proteins by the PKC-signaling networks, perinuclear Ca release events have been suggested to result in specific nuclear Ca signals and trigger accompanying Ca-dependent transcription events (). Although translocation of the cPKC molecule to its target, the plasma membrane, is an important step in the activation mechanism, the question still remains whether translocation itself can, in fact, indicate activation of cPKCs. This might be rather likely for long-lasting Ca transients, given the very high DAG affinity of Ca-bound cPKC. In this study, we could provide strong direct evidence for a long-lasting membrane interaction of C1a and DAG in living cells; functionally mutating the C1a domain of PKCα rendered the dissociation process very fast (, d); thus, C1a-mediated membrane binding of PKCα represents a significant pool of long-lasting PKCα membrane interactions out of which PKCα could have sufficient time to encounter and interact with target proteins and perform phosphorylation. In contrast, this is not necessarily obvious for short-lived LTEs. If binding of the C1 and C2 domains to the plasma membrane occurs independently (), and the initial membrane affinity is mainly determined by its membrane C2 interactions, ionic interactions between the Ca-occupied C2 domain and PS in the plasma membrane are the next step (). The C2 membrane interaction provides the first part of the energy necessary for the allosteric release of the pseudosubstrate from the kinase domain (). These processes have brought the C1 domain into the proximity of the plasma membrane for C1–membrane interactions that are believed to predominantly determine the residence time of the cPKC at the membrane (). After C1a–DAG interactions, the pseudosubstrate is released from the kinase domain that is now ready for substrate interactions and phosphorylation of the target proteins (). Despite the fact that this sequence of events is, in principle, agreed on quite widely, some details are still open for discussion and might have to be reinterpreted in light of our findings of very short-lived LTEs and the two pools of PKCα membrane interactions. If DAG binding of Ca-cPKC is, indeed, the principle event linking the cPKC to the membrane, then how does this notion account for the two populations of lifetimes observed in this study? We propose that these distinct lifetimes reflect different states of the cPKC membrane interactions. As we have previously described, rises in the Ca concentration usually “precede” cPKC translocation, and very low-amplitude signals are not necessarily translated into translocation events ( and ), which is indicative of a certain Ca threshold concentration necessary for cPKC translocation. ). of 35 μM (). Therefore, we suggest that the LTEs with the shorter lifetimes reflect pure Ca-dependent binding of PKCα to the plasma membrane. With an apparent diffusion coefficient of 4.67 × 10 cm/s, a cytosolic viscosity of around 2.0–3.2 cP (; ), and an average lifetime of the Ca–PKCα complex of ∼12–15-ms molecules can only interact with the plasma membrane when Ca binding occurs ≤1 μm away from the plasma membrane (for an illustration see Fig. S2). Furthermore, even for long-lasting Ca increases, slower and longer lasting membrane interactions solely driven by Ca increases have been described at the global cellular level (; ). When binding of the Ca-C2 domain to the PS of the plasma membranes occurs, the lifetime of the Ca-binding complex increases fivefold and, thus, plasma membrane interaction times are significantly prolonged (calculated for PKCα-C2 domain according to in vitro data; ; ; ). With a putative phosphorylation turnover rate of 380–500 min measured in vitro (; ), such brief membrane interactions might, in fact, be sufficient to phosphorylate target proteins in case of an encounter. Elementary Ca signals, such as Ca puffs, have average lifetimes of ∼1–2 s (; Bootman et al., 1997b; ; ); thus, the shortest lifetimes of the LTEs observed here are compatible with a purely Ca-driven process. This notion is further substantiated in that, upon a stepwise decrease, the Ca concentration PKCα(C1a-mut) displayed a rapid membrane dissociation, very closely following the Ca decline (, d, red transparent trace). Longer lifetimes of the cPKC membrane complex might indicate additional, and possibly more specific, interactions. Therefore, we propose that the membrane translocations with longer lifetimes represent interactions between the Ca-PKCα and DAG. This is particularly interesting because investigation of these lifetimes could yield additional information of that important interaction. Such extended membrane localizations could also indicate additional interactions of the activated cPKCs with receptors of activated kinases (). Moreover, simultaneous cPKC interactions with several partners will increase the membrane-residence periods overproportionally because redistribution into the cytosol can only occur when the cPKCs unbind from both partners almost simultaneously. Thus, the observed long-lasting translocation events might represent a temporal “walking” of the molecule between dynamic binding/unbinding cycles to the various partners in the plasma membrane. Two additional possible schemes might give rise to such long-lasting translocation events, including highly repetitive or long-lasting Ca puffs, but these can most likely be excluded. Their frequency ought to be well above 2–3 Hz because we have visualized LTEs with lifetimes of around 300–500 ms. Such high-frequency Ca puffs have not been imaged yet and are, in fact, rather unlikely taking into account adaptive processes and refractory time periods of the InsPR (). Our experiments () do not seem to favor the notion of receptor of activated kinase binding because mutation of the C1a domain alone completely suppressed that PKCα pool with the longer membrane-residence time. On the other hand, long-lasting Ca puffs with extended lifetimes (>3–5 s) are also unlikely as the responsible mechanism. Such long-lasting Ca puffs have not been described yet, and local depletion mechanisms in the ER might actually terminate local Ca release much earlier, as has been proposed for Ca sparks (). Nevertheless, this cannot easily account for the limited spatial spread of such long-lasting LTEs (). For this to occur, at least one of the binding partners has to exhibit a restricted lateral diffusion in the plane of the plasma membrane (∼1 μm/s; unpublished data). The likely candidates for such a mechanism are phosphorylation target proteins that have been anchored in the plasma membrane or membrane lipid domains that were rich in DAG after receptor stimulation. In these situations, residence times of the cPKCs are primarily determined by C1a-DAG binding or target protein interactions in the plasma membrane, and the extent of membrane location will be increased largely. In conclusion, we propose that the two LTE lifetimes described here represent visual correlates of distinct states of cPKCs during their activation/deactivation cycle. K rates seem to be predominantly regulating the duration of cPKC membrane interactions with purely C2 domain–dependent membrane interactions being the most dynamic process. C1a domain–mediated membrane or cPKC target protein interactions are more stable, and, consequently, LTEs display a longer lifetime. These notions are supported by our mutation and UV flash experiments, in which we illustrated that C2-mediated membrane interactions were very fast in both directions and that association and dissociation and functional C1a domains largely increased membrane residence times, i.e., decay times were significantly extended. In addition, we provided first evidence that two PKCα populations coexist in the membrane (C2- and C1a-bound proteins), which most likely undergo rapid interconversion. Our results, thus, foster the understanding of cellular signaling via PKCα (possibly also for the other cPKCs). At resting Ca concentrations, the C2 domain repels the PKCα from the membrane because of electrostatic interactions. As soon as the Ca concentration is increasing, subplasmamembrane PKCα molecules with a Ca-C2 domain bind to the PS in the plasma membrane and the PS–Ca–C2 complex is stabilized. Failing to interact with PS causes rapid dissociation from the membrane. This is an essential step that can be imaged as the association of fluorescently tagged PKCα molecules to the membrane. We thus propose that during cellular Ca signals Ca-C2-PKCα in fact “tiptoes” on the membrane until it “sticks” to a PS. This stabilization step gives the PKCα sufficient time to “probe” for DAG in the membrane. Binding to the DAG (via its C1a domain) further stabilizes the membrane interactions and further increases membrane-residence times. Such a stabilized complex is now given enough time to phosphorylate target proteins. From this we conclude that, at least for cPKCs, spatial and temporal targeting of cPKC-mediated signaling is mostly, if not solely, driven by the spatiotemporal properties of the underlying Ca-signaling machinery. We believe that these findings put cPKCs in at least the same position as the classic Ca signal decoder calmodulin. We have visualized that a downstream signaling molecule, PKCα, is able to decode cellular and subcellular Ca signals in all their complexity into translocation and, thus, is an essential linker between Ca transients and downstream signal transduction networks (e.g., RhoA, Rac1 or ERK/MAPK; ), but also directly phosphorylates membrane proteins such as ion channels and transporters. To our knowledge we illustrated for the first time in living cells that PKCα rapidly converts between various states of membrane interactions and that there are at least two pools of PKCα in the membrane after agonist stimulation. Human PKCα was C-terminally fused to GFP or YFP in a custom-made pcDNA3-EGFP or -EYFP fusion plasmid, as previously described (). For the generation of DsRed2 (BD Biosciences) and mRFP () fusion proteins, we used pcDNA3-DsRed2 and -mRFP fusion plasmids, respectively. In line with the published results of , ) on PKCα mutants, we performed site-directed mutagenesis () with the primer 5′-TGTTTTGTGGTCCACAAGGCCTGCCATGAATTTGTTACTTTTTCTTG-3′ in the C1a domain and the primer 5′-CATGAAGTCATTCCTTGTCGTACGATCCCAGTTCCAGATTTCTACAGACA-3′ in the C2 domain of the human PKCα-EGFP fusion protein. We generated R77A (PKCα[C1a-mut]-EGFP) to suppress DAG binding to the C1a domain and D246N (PKCα[C2-mut]-EGFP) to suppress Ca binding to the C2 domain. All constructs were confirmed by sequencing. HEK293 and COS1 (CRL-1573 and CRL-1650; American Type Culture Collection) cells were grown in DME (Invitrogen) supplemented with 5% FBS (PANBiotech), 50 U/ml penicillin, and 50 U/ml streptomycin (both from Invitrogen). Cells were seeded on 20-mm glass coverslips in 12-well plates and transfected the following day at 20–30% confluency with Lipofectamine 2000 (Invitrogen) and appropriate amounts of plasmid DNA per well. All experiments were performed in both cell types and the results shown are representative of both cell types. It must be mentioned that for all the findings we verified in COS and HEK cell lines, we never found qualitative differences between the cell types. Localization and translocation of the fluorescent fusion proteins, as well as changes in the fluorescence of Ca indicators such as fluo-4 and Fura red (Invitrogen), were analyzed by laser scanning confocal microscopy. We either used a Nipkow disk–based QLC100 (VisiTech International) equipped with a dual-port adaptor for monitoring two emission wavelengths simultaneously (ORCA-ER cameras; Hamamatsu) or the kilobeam 2D-array scanner–based VTinfinity (VisiTech International) equipped with an ORCA-ER camera for single emission recordings. Both confocal systems were mounted to either an inverted microscope (TE-2000U; Nikon) or an upright microscope (E600; Nikon) and controlled by VoxCellScan software (VisiTech International). Excitation of the fluorescent fusion proteins was performed with the 488-nm line (EGFP or EYFP) or the 568-nm line (for DsRed2) of an Ar/Kr multiline laser (LaserPhysics). Emission was imaged at wavelengths >500 nm for EGFP and EYFP, and at wavelengths >580 nm for DsRed2. In experiments where we used fusion proteins and Ca indicators, we used two different setups. For the first setup, EGFP or EYFP together with Fura red were only excited at 488 nm, while the emissions were imaged at 520 and 600 nm, respectively. For the second setup, fluo-4 and DsRed2 required excitation at two wavelengths (488 and 568 nm), and the emissions were imaged at 520 and 600 nm. Individual images or image pairs were recorded at acquisition speeds ranging from 1 to 10 Hz through 40× (PlanFluor, NA 1.3, DIC H), 60× (PlanApo, NA 1.4, DIC H), or 100× oil immersion objectives (PlanApo, NA 1.4, Ph3 DM; all Nikon). All experiments were performed at room temperature (20–22°C) 2–4 d after transfection. In experiments with Ca indicators, the dyes (1 μM fluo-4 AM or 3 μM Fura red AM) were loaded into the cells for 30 min, with an additional 20 min after wash-out for deesterification. For rapidly increasing or decreasing the intracellular Ca concentration, we loaded the cells with 7.5 μM NP-EGTA–AM or diazo-2–AM (Invitrogen), respectively. A UV flash II (Till Photonics) was attached to an upright microscope (E600; Nikon) via an additional epifluorescence input. The UV was directed into the excitation light path via a UV dichroic mirror (HQ450LP; OptiLab). We used a water immersion objective (40×, NA 0.8; Nikon). This allowed for simultaneous recording of the cellular fluorescence and application of the UV flash. The timing between imaging and the flash was programmed within the VoxCellScan confocal software. Because both caged compounds decreased resting Ca to an unknown degree, quantitative measurements of Ca were not possible. In addition, the amount of Ca photoreleased from NP-EGTA largely depends on the amount of (a) photolysed compound and (b) Ca loading of NP-EGTA. The latter effect explains the different Ca steps in caused by UV flashes of identical amplitude. is increased to 1 mM); thus, the amount of photoreleased Ca was rather low, but growing with increasing “resting” Ca before the subsequent flashes (higher occupancy of NP-EGTA by Ca). With AM loading, such effects are largely unavoidable because we could not “preload” the caged compound as is possible when perfusing the cell in the whole-cell configuration of the patch clamp technique used elsewhere (). is increased upon UV illumination from 2.2 to 73 nM. All experiments were performed in a Hepes (Merck)-buffered salt solution (extracellular medium) composed of (in mM): 135 NaCl, 5.4 KCl, 2 MgCl, 10 glucose, 2 CaCl, and 10 Hepes adjusted to pH 7.35 with NaOH. ATP (Sigma-Aldrich) and PMA (Calbiochem) were dissolved as stock solutions in the appropriate solvents and diluted in the external salt solution at the given concentrations before each experiment. Rapid solution switch (exchange time <1 s) was achieved by means of a solenoid-driven custom-made perfusion system. With this system a laminar flow of ∼200 μm in diameter was applied to the entire field of view; thus, concentration changes of agonists display global changes occurring simultaneously over the field of view. After storage of the image, we analyzed the data using ImageJ (W. Rasband, National Institutes of Health, Bethesda, MD). ROI fluorescence over time data were transferred into IGOR software (Wavemerics) and further processed for data display. To account for inhomogeneous dye distribution, we calculated self-ratio images as indicated. For this, we divided subsequent images of a video by a mean image obtained after averaging the 5–10 resting images of each movie (,). Color-coded 2D images were constructed in ImageJ as 8-bit grayscale images to which the given color look-up table was applied. When given in the figures, Ca concentrations were calculated as described earlier, assuming a resting concentration of 100 nM (). 3D surface representations of individual images or entire movies were calculated with Imaris4 software (Bitplane). All results of our live-cell imaging were analyzed in cells displaying a wide range of expression levels. For the quantitative analysis of LTEs presented in , we had identified criteria that local fluorescence changes had to meet to be treated as LTEs. We identified fluorescence changes as LTEs when their spatial spread was <6 μm, their duration was at least 2 images (or 250 ms) above twofold the standard deviation of the resting signal, and their amplitude was at least two times the standard deviation of the resting fluorescence noise (horizontal dashed line in ) for at least 2 images (vertical dashed line in ). Fig. S1 exemplifies HEK and COS cells expressing PKCα fusion proteins showing Ca and PKCα oscillations and graded Ca and PKCα responses. Fig. S2 shows a model of the mechanisms possibly leading to the generation of LTEs. Video 1 shows the tight relationship between Ca oscillations (Fura-Red) and PKCα-EYFP translocation. Video 2 displays the spatiotemporal properties of a PKCα translocation wave propagating through HEK cells stimulated with ATP. Video 3 illustrates the spatiotemporal properties of a long-lasting LTE in COS1 cells upon stimulation with threshold concentrations of ATP. Online supplemental material is available at .

Intercellular channels, which cluster together to form gap junctions, are involved in various physiological functions (e.g., adaptation of retinas to the dark, conduction of excitation in the heart, and suppression of cell proliferation in cancer tissues). For vertebrates, gap junctions are formed by connexins, a multigene family of which 20 members have been identified in humans (). A new family of gap junction molecules, which are unrelated to connexins, has been identified in insects and nematodes and named innexins (). We have recently demonstrated the presence of innexin homologues in various taxonomic groups, including vertebrates (; ). Given the ubiquitous distribution of this protein family in the animal kingdom, we termed these proteins “pannexins” (PROSITE accession number ; ). Three genes, pannexin-1 (PanX1), -2 (PanX2), and -3 (PanX3), have been cloned from the human and mouse genome, and the pattern of their expression in various tissues has been studied (; ; ; ). It has been found that the human PanX1, which encodes mRNAs, are ubiquitously, although differentially, expressed in normal tissues. Human PanX2 is a brain-specific gene (; ). Recently, it was demonstrated that in paired oocytes rodent PanX1, alone and in combination with PanX2, induced the formation of intercellular channels (). When expressed in a single oocyte, PanX1 hemichannels were shown to be functional in plasma membrane (; ). Pannexin membrane channels are mechanosensitive conduits for ATP (). This type of nonjunctional function has been previously reported for connexins (for review see ). However, it is not clear if pannexins simply duplicate connexin functions or play some special physiological role. In this work, we investigated the pannexin function in human cell lines transiently or stably transfected with pannexin (human PanX1). Our results demonstrate that overexpression of PanX1 enables the formation of Ca-permeable gap junction channels between adjacent cells, thus, allowing direct intercellular Ca diffusion and facilitating intercellular Ca wave propagation. Furthermore, we obtained evidence that strongly indicate that, in addition to the gap junction function, PanX1 overexpression increases the Ca permeability of the ER membrane and thereby affects intraluminal ER Ca concentration ([Ca]). PanX1 overexpression dramatically reduces the intraluminal Ca content of the ER, which was directly measured by a fluorescent Ca indicator, Mag-fura-2. Endogenous PanX1 depletion by antisense and siRNA strategy in human prostate cancer cells increased the Ca content of the ER. Therefore, it seems likely that pannexins, which are structurally similar to gap junction–forming molecules, may also be involved in intracellular calcium homeostasis via the formation of the ER Ca-leak channels. These results give new insight into the mechanisms of the basal ER Ca leak, which has remained poorly understood until now. Thus, the results of our study imply that where vertebrates are concerned, the pannexin family of gap junction proteins not only facilitates an intercellular Ca movement but also represents one of the mechanisms responsible for ER Ca leak. #text Our study demonstrates for the first time that human PanX1, a protein encoded by a member of a new family of genes with yet unknown physiological role or roles, is a Ca-permeable ion channel that is localized on both the ER and plasma membrane and participates in two physiologically important processes: the ER Ca leak and intercellular Ca movement. Intercellular gap junction channels provide the primary pathway for communication between cells, which is crucial for coordination of tissue metabolism and sensitivity to extracellular stimuli. Where vertebrates are concerned, the integral membrane proteins forming intercellular channels are referred to as connexins. In insects and nematodes, this function has been attributed to proteins named innexins (). By homology with these invertebrate gap junction proteins, we predicted that another protein fam-ily, pannexins, might also form gap junctions (). Connexin channels have been demonstrated to clus-ter in maculae known as gap junctions and to allow cell–cell diffusion of ions (predominantly monovalent cations; for re-view see ) and small molecules. Pres-ently, it is commonly agreed that connexins provide two major pathways for intercellular calcium signaling. The first one, an “intracellular” pathway, involves the passage of a Ca2+-mobilizing messenger, such as IP3, through gap-junctional connexins (; ). The second one, an “extracellular” pathway, involves the release of a purinergic messenger, such as ATP, through connexin hemichannels, and a subsequent activation of P2Y receptors in a paracrine way (; ; ; ). Although Ca ions are recognized as second messengers within individual cells, their role as diffusible messengers in intercellular signaling has largely been overlooked because elevated [Ca] has been shown to reduce gap-junctional conductance in several systems, including insects and vertebrates (; ; ). Indeed, it is now generally agreed that even if a small amount of Ca can diffuse across gap junctions, it probably does not play a significant role in intercellular calcium signaling through connexin channels, which is mediated mainly by IP or other small signaling molecules (; ; ). This view is also supported by the recent observation that, in articular chondrocytes, intercellular calcium waves evoked by mechanical stimulation were abolished by incubation with TG and the phospholipase C inhibitor (). In contrast, in PanX1-transfected LNCaP cells an abrupt increase in [Ca] evoked by dialysis of one cell through the patch pipette with solution containing high Ca concentration (such an elevation of [Ca] is expected to uncouple gap junction channels formed by connexins) caused an elevation of [Ca] in adjacent cells, which was consistent with cell–cell Ca diffusion via gap junction channels formed by pannexin. Indeed, intercellular Ca movement was observed under conditions where the contribution of both the intracellular (or IP-dependent) and the extracellular pathway was eliminated by the inhibition of PLC and IPRs, depletion of ER, and removal of Ca from extracellular medium. Furthermore, the inability of LNCaP cells to respond to external ATP application (unpublished data) also argues against any possible involvement of an ATP-dependent extracellular pathway. Thus, our results suggest that pannexin proteins may form Ca-permeable channels providing a pathway for intercellular Ca diffusion. Even more intriguing is our finding that pannexins may also function as “leak channels” in the ER membrane. Indeed, the ER is the major calcium store (; ), and the Ca-filling status of the ER controls many physiological processes, ranging from gene expression to apoptosis and proliferation (). Under resting conditions, steady-state [Ca] is determined by the dynamic equilibrium of two components; an active Ca uptake mediated by ATP-dependent Ca pumps of the SERCA family and passive Ca efflux via leak channels. Even though this pump–leak cycle appears to be a common property of Ca-storing organelles, little is known about the molecular nature of the Ca-leak pathway. Several mechanisms involving quite different proteins have been previously suggested to explain the basal Ca leak from ER (for review see ), namely: (a) reverse Ca flux through the pumps (), (b) Ca leak in neutral complexes with small molecules by translocon channels (; ), (c) the fluxes of Ca through “natural” ionophores, such as bile acids (; ), (d) an antiapoptotic protein Bcl-2–mediated Ca leak (; ; ), and (e) IP3R- or RYR-mediated Ca leak (). However, as concluded by , “the drawing of these mechanisms is only a fantasy map of the leak terra incognita and discovery of the exact mechanisms of calcium leak remains a challenge to scientists working in the calcium signaling field.” The results of our study strongly suggest that the heterologous expression of PanX1 in LNCaP and HEK cells dramatically reduces the ER Ca content and alters the Ca permeability of the ER membrane, which is consistent with an ion leak-channel function of PanX1 in the ER membrane. To estimate the potential role of pannexins in endogenous ER basal Ca leak, we used the siRNA and antisense hybrid depletion strategy for the endogenous PanX1 protein. Interestingly, in PanX1-depleted cells, the ER Ca content was found to be ∼40% higher than in control cells. Moreover, the rate of the ER Ca leak (unmasked by inhibition of SERCA-mediated Ca uptake with TG) was substantially reduced in PanX1-depleted cells, thus, suggesting an important contribution of endogenous PanX1 to the global ER basal Ca leak. One may speculate that reduced resting concentration of calcium in the ER associated with the PanX1 overexpression could be caused by the following: (a) the modified level of the BCL-2 family of proteins with ER localization and known to play an important role in the regulation of the calcium leak from the ER, (b) the increased level of the antiapoptotic protein BCL-2 (; ; ), and/or (c) the deficiency for two “multidomain” proapoptotic proteins Bax and Bak (; ). Our results show that the levels of these proteins expression were not changed by PanX1 overexpression, thereby suggesting that the PanX1 may mediate Ca leak by itself, independently of other potential ER leak modulators. In this respect, it would be interesting to investigate whether the function of a pannexin, such as the ER Ca-leak channel, is specific to vertebrates, or if, in fact, some invertebrate innexins (which are pannexin homologous) share this function. In conclusion, this study directly demonstrates the involvement of PanX1 in intra- and intercellular Ca signaling, thus, illustrating the multifunctional role of a single molecule. PanX1 cDNA encompassing the entire coding region was synthesized by PCR amplification of the cDNA clone () using two gene-specific primers. A sense primer, containing a HindIII site at its 5′-end (PanX1F: 5′-TGTAAGCTTGCCATGGCCATCGCTCAAC-3′), and an antisense primer, containing an EcoRI site at its 5′-end (PanX1R: 5′-TGTGAATTCCCAGAAGTCTCTGTCGGGC-3′), were used. 500 ng of the pEGFP-N1 vector (CLONTECH Laboratories, Inc.) and PCR product were digested by HindIII and EcoRI. The digested vector and the PCR products were gel purified, ligated together, and cloned. The cDNA insert was sequenced to verify its identity and absence of mutation. Cx43-EGFP was a gift from B. Rose (Marine Biology laboratory, Woods Hole, MA), and Cx32 was a gift from M. Mesnil (University de Poitiers, Poitiers, France). The procedures for culturing and preparing LNCaP cells (American Type Culture Collection) for fluorescence measurements are detailed elsewhere (). HEK-293 cells were grown in DME, containing 10% FCS, 2 mM L-glutamine, and 100 μg/ml kanamycin at 37°C in a humidity-controlled incubator with 5% CO HEK-293 cells (50% confluent) were transiently transfected by 2 μg of plasmid (pEGFP-N1 or pannexin-pEGFP-N1, using a transfection reagent (Gene Porter 2; Gene Therapy Systems, Inc.) for 8 h in a serum-free medium. DME with 10% FBS was added overnight. Chimera expression was assessed by GFP fluorescence. Stable cell lines expressing PanX1-EGFP protein were constructed by transfection with 2 μg of pEGFP-N1/PanX1 plasmid in a 6-well plate for 6 h using a Gene Porter 2 reagent, following the manufacturer's recommended protocol. The cells in culture were then maintained under selected pressures with 700 μg/ml G418 for 4 wk. Colonies expressing GFP were identified under fluorescence microscope, subcloned, and maintained under the selected pressure for at least 3 wk. LNCaP, LNCaP-PanX, or HEK cells were transfected overnight by either 5 or 100 nM siRNA anti-PanX1 mRNA (siRNA PanX), using Gene Porter 2 transfection reagent in 35-mm dishes for electrophysiological purpose or in 60-mm dishes for either RNA or protein extraction. Ready-to-use siRNA-AR (processing option: A4) was synthesized by Dharmacon, Inc. Location of either siRNA or ODNs refer to PanX1 cDNA from NM_015368.3. The sense sequence of siRNA-PAnX1 and siRNA mTRPC6 used were 5′-ACGAUUUGAGCCUCUACAA(dTdT)-3′ (1362–1380) and 5′-UAUUGCCGAGACCG UUCAU(dTdT)-3′ (1591–1609, accession no.: NM_013838.1), respectively. Phosphorothioate ODNs were produced by Eurogentec and used at 0.5 μM. Antisense sequences used were 5′-TATGCAGCCACAGTGGGAGG-3′ (685–704) and 5′-TCAGATACCTCCCACAAACT-3′ (929–948), although the sense sequence used was 5′-CCTCCCACTGTGGCTGCATA-3′. Total RNA was isolated from different cell lines using the guanidium thiocyanate-phenol-chloroform extraction procedure (). After a DNase I (Life Technologies) treatment to eliminate genomic DNA, 2 μg of total RNA was reverse transcribed into cDNA at 42°C using random hexamer primers (Perkin Elmer) and MuLV reverse transcriptase (Perkin Elmer) in a 20-μl final volume, followed by PCR. The PCR primers used to amplify pannexin cDNAs were designed with Gene Runner 3.05 (Hastings Software). Primers for the human pannexin synthesized by Life Technologies were as follows: forward 5′-CCCAATTGTGGAGCAGTACTTG-3′ (963–984) and reverse 5′-AGACACTTGTATGAC TTGACCTCAC-3′ (1403–1427). The expected DNA length of the PCR product generated by these primers was 465 bp (NM_015368, National Center for Biotechnology Information database). PCR was performed on the RT-generated cDNA using a thermal cycler (GeneAmp PCR System 2400; Perkin Elmer). To detect pannexin cDNAs, PCR was performed by adding 1 μl of the RT template to a mixture of (final concentrations): 50 mM KCl, 10 mM Tris-HCl, pH 8.3, 2.5 mM MgCl, 200 μM of each dNTP, 600 nM of sense and antisense primers, and 1 U AmpliTaq Gold (Perkin Elmer) in a final volume of 25 μl. DNA amplification conditions included an initial 5-min denaturation step at 95°C (which also activated the Gold variant of Taq Polymerase) and 35 cycles of 30 s at 95°C, 30 s at 60°C, 40 s at 72°C, and finally, 5 min at 72°C. LNCaP-PanX cells (vehicle or siRNA-transfected) were harvested and pelleted in PBS and then sonicated in ice-cold buffer, pH 7.2, containing the following: 10 mM PONa/K buffer, 150 mM NaCl, 1 g/100 ml sodium deoxycholate, 1% Triton X-100, 1% NP-40, a mixture of protease inhibitors (Sigma-Aldrich), and a phosphatase inhibitor (sodium orthovanadate; Sigma-Aldrich). Samples were electrophoretically analyzed on a 10% polyacrylamide gel using the SDS-PAGE technique. The proteins were then transferred for 1 h (50 mA, 25 V) onto a nitrocellulose membrane using a semidry electroblotter (Bio-Rad Laboratories). The membrane was then cut into thin, equally sized strips and processed for Western blot. The strips were blocked in 5% TNT-milk (15 mM Tris buffer, pH 8.0, 140 mM NaCl, 0.05% Tween 20, and 5% non-fat dry milk) for 30 min at room temperature, washed three times in TNT, soaked in primary antibody anti-GFP (CHEMICON International, Inc.), anti-actin (MS-1295-P; Neomarkers), anti-Cx32 (CHEMICON International, Inc.), or anti-Cx43 (CHEMICON International, Inc.), and then diluted 1:500, 1:500, 1:200, and 1:200, respectively, in TNT-milk for 1 h at room temperature. After three washes in TNT, the strips were transferred into the IgG horseradish peroxidase–linked secondary antibodies (CHEMICON International, Inc.), and diluted in TNT-milk (1:20,000) for 1 h. After three 10-min washes in TNT, the strips were processed for chemiluminescent detection using chemiluminescent substrate (Supersignal West Pico; Pierce Chemical Co.) according to the manufacturer's instructions. The blots were then exposed to X-Omat AR films (Eastman Kodak Company). LNCaP cells were transfected at 60% confluency, as described in Transient expression in HEK-293 cells, in 6-well plates. After two washes in PBS, cells were fixed with 4% formaldehyde-1X PBS for 15 min. After two washes in PBS, the slides were mounted with Mowiol. Fluorescence imaging was performed in HBSS solution containing 142 mM NaCl, 5.6 mM KCl, 1 mM MgCl, 0–10 mM CaCl, 0.34 mM NaHPO, 0.44 mM KHPO, 10 mM Hepes, and 5.6 mM glucose. The osmolarity and pH of external solutions were adjusted to 310 mosM l and 7.4, respectively. Cytosolic calcium concentration was measured using fura-2–loaded cells. LNCaP cells were loaded for 45 min at room temperature with 2 μM fura-2-AM prepared in HBSS and, subsequently, washed three times with the same dye-free solution. The coverslip was then transferred into a perfusion chamber on a microscope (IX70; Olympus) equipped for fluorescence. Fluorescence was alternatively excited at 340 and 380 nm with a monochromator (Polychrome IV; TILL Photonics) and was captured after filtration through a long-pass filter (510 nm) by a 5 MHz charge-coupled device camera (MicroMax; Princeton Instruments). Acquisition and analysis were performed with the Metafluor 4.5 software (Universal Imaging Corp.). The intracellular calcium concentration was derived from the ratio of the fluorescence intensities for each of the excitation wavelengths (F340/F380) and from the equation of . All recordings were performed at room temperature. The cells were continuously perfused with the HBSS solution, and chemicals were added via the perfusion system. The flow rate of the whole-chamber perfusion system was set to 1 ml/min, and the chamber volume was 500 μl. ([Ca] was monitored using Mag-fura-2 as previously described (). The HEK-293 and LNCaP cells were grown on coverslips and transfected with pannexin-pEGFP-N1. Fluorescence imaging was performed using a confocal scanner (488 nm excitation for GFP; LSM 510; Carl Zeiss MicroImaging, Inc.) based on an Axiovert 200 M motorized inverted microscope with a plan-Apochromat 63×, 1.4 NA, oil immersion objective (Carl Zeiss MicroImaging, Inc.). The confocal microscope software used was AIM 3.2 (Carl Zeiss MicroImaging, Inc.). Confocal [Ca] imaging in LNCaP cells was performed using Ca-sensitive indicators fuo-4 or rhod-2. Fluo-4 was loaded by 20-min exposure of the cells to 5 μM fluo-4 AM (diluted from a stock containing 2 mM fluo-4 AM and 0.025 [wt/vol] pluronic F-127 in dimethyl sulphoxide) at room temperature, followed by a 40-min wash to allow time for deesterification. Rhod-2 was loaded by 10-min incubation of the cells with 15 μM rhod-2 AM (diluted from a stock containing 1 mM rhod-2 AM and 0.025 [wt/vol] pluronic F-127 in dimethyl sulphoxide) at room temperature, followed by a 60-min wash. Rhod-2 fluorescence was excited by the 543-nm line of a 5-mW HeNe ion laser and the emitted fluorescence was captured at wavelengths above 560 nm. Fluo-4 and GFP fluorescence were excited by the 488-nm line of a 20 mW argon ion laser and the fluorescence emitted was detected at wavelengths >505 nm. For both desired laser lines, the illumination intensity was set with an acoustooptical tunable filter. In the experiments on the imaging of intercellular Ca movement, one of the two adjacent LNCaP cells was dialyzed through the patch pipette with solution containing either 1 μM Ca (pipette solution: 145 mM KCl, 10 mM Hepes, 5 mM glucose, 125 μM MgCl, 3.8 mM HEDTA, and 0.3 mM CaCl; pH 7.4 with KOH) or 100 μM Ca (pipette solution: 145 mM KCl, 10 mM Hepes, 5 mM glucose, 2 mM MgCl, and 100 μM CaCl; pH 7.4 with KOH), while changes in the fluorescence of the Ca-sensitive indicator were monitored in the dialyzed and adjacent cells using x–y confocal imaging. To unmask Ca diffusion from the dialyzed to the adjacent cell, Ca entry from extracellular media was prevented by incubation of the cells in Ca-free solution supplemented with 300 μM EGTA, (bath solution: 140 mM NaCl, 5 mM KCl, 2 mM MgCl, 0.3 mM NaHPO, 0.4 KHPO, 4 mM NaHCO, 5 mM glucose, and 10 mM Hepes; pH adjusted to 7.4 with NaOH), whereas Ca release from the ER was prevented by the following: (a) ER depletion with 0.5 μM TG, (b) blocking RyRs with 100 μM ryanodine, (c) IPRs blocking with 100 μM 2-APB, and (d) PLC inhibition with 5 μM U-73122. Spatial organization of the ER was visualized using selective ER marker BODIPY 558/568 brefeldin A (; ; ). The ER was stained by 10–30 min incubation of the cells at room temperature in the solution containing 0.1–0.5 mM of the dye. The fluorescence was excited by the 543-nm line of a 5-mW HeNe ion laser, and the emitted fluorescence was captured at wavelengths >560 nm. To visualize the fine spatial pattern of the BODIPY 558/568 brefeldin A and EGFP fluorescence, the fluorescent signal was collected from the confocal optical slice below 0.5 μm with x–y frame size of 2048 × 2048 pixels, and the final images were obtained as a result of averaging of four sequential images taken in multitrack (line-by-line acquisition) configuration of the confocal scanner followed by low-pass filtering (7 × 7 pixels; LSM 510 software) to improve the signal-to-noise ratio. All chemicals were obtained from Sigma-Aldrich, except for fura-2-AM and TG, which were purchased from Calbiochem, and Mag-fura-2-AM, Fluo-4-AM, Rhod-2, and BODIPY 558/568 brefeldin A, which were obtained from Invitrogen. Each experiment was repeated several times. The data were analyzed using Origin 5.0 (Microcal) software. Results were expressed as the mean ± the SEM where appropriate. The test was used for statistical comparison of the differences, and P < 0.05 was considered significant.

Apoptosis is a critical process in development and normal tissue homeostasis and results in immediate removal of dying cells, either by neighboring cells or by professional phagocytes, such as macrophages or dendritic cells. The engulfment of apoptotic cells is one of the most primitive forms of phagocytosis, and many of the molecules implicated in the phagocytosis of apoptotic cells appear to have a high degree of conservation from the nematode to mammalian cells (). Phagocytosis of apoptotic cells is distinguished from the evolutionarily much younger Fc receptor (FcR)–mediated phagocytosis in terms of receptors, signaling molecules, and cytoskeletal as well as plasma membrane reorganizations (). In , at least two partially redundant processes are important in the engulfment of dying cells, one that involves CED-2, -5, and -12 and another involving CED-7, -1, and -6, and subsequently both pathways converge at CED-10 (). In the second pathway, CED-7 may be required for the function of CED-1 (), although recent data indicate that CED-7 may have alternative roles during the removal of apoptotic cells (). For all of these ced genes, mammalian orthologues have been proposed (for review see ). For CED-7, the ATP-binding cassette transporter A1 (ABCA1) has been suggested as the mammalian orthologue, based on sequence conservation and in vivo and cell culture studies suggesting a role of ABCA1 in the phagocytosis of apoptotic cells (; ; ). The low-density lipoprotein receptor–related protein 1 (LRP1) or CD91 has been suggested as a protein with function similar to CED-1 (). LRP1 is a multifunctional scavenger and signaling receptor (). The suggestion that LRP1 has a function similar to CED-1 is based on intracellular sequence similarity and an important role of LRP1 in the phagocytosis of apoptotic cells (; ; , ). Moreover, both LRP1 and CED-1 can interact via a phosphotyrosine binding motif (NPXY motif) within their cytoplasmic tails with the adaptor proteins CED-6 and its mammalian orthologue GULP (engulfment adaptor protein), respectively (). However, LRP1 and CED-1 show very limited overall sequence homology, and it is possible that the apparent similarity in their functions represents convergent evolution. ABCA7, a close homologue of ABCA1, is a 220-kD protein expressed in a variety of tissues, including macrophages (). We have shown that ABCA7 binds apolipoprotein A-I (apoA-I) and promotes the efflux of phospholipids but not cholesterol to apoA-I in ABCA7-transfected 293 cells. In resting macrophages, ABCA7 is found intracellularly and does not contribute to phospholipid or cholesterol efflux (; ). The physiological role of ABCA7 in these cells remains unknown. In this paper, we show that ABCA7 has a high sequence similarity to CED-7 and is required for efficient phagocytosis of apoptotic cells. We were unable to demonstrate a role of ABCA1 in the phagocytosis of apoptotic cells as reported previously (; ). ABCA7 appears to facilitate the cell surface localization of LRP1 and associated signaling via extracellular signal–regulated kinase (ERK), thereby promoting the phagocytosis of apoptotic cells. Together, these data suggest that ABCA7 and not -A1 is a functional orthologue of CED-7. An alignment of CED-7 and mouse ABCA7 showed that 25% of amino acids were identical and 43% were similar. A similar result was obtained if CED-7 was aligned to mouse ABCA1, resulting in 24% identity and 42% similarity. These homologies are in the range reported for known orthologous proteins (e.g., CED-6 and GULP; ). Compared with ABCA1 and -A7, CED-7 has a gap in the amino acid sequence starting at amino acid 198 (Fig. S1 a, available at ). This difference is more pronounced for ABCA1 (Fig. S1 b, dashed line). No other mammalian genes are more homologous to CED-7 than ABCA7 or -A1. Based on these sequence similarities, both ABCA7 and -A1 could be considered as potential orthologues to CED-7; thus, we undertook a study of the function of ABCA7 in the phagocytosis of apoptotic cells. The first observation indicating that ABCA7 might be involved in phagocytosis came from fluorescence microscopy experiments. We previously reported that “resting” macrophages show mainly intracellular staining for ABCA7 (). We confirmed this finding by confocal microscopy and saw mainly intracellular staining for ABCA7 (). Some cells also showed staining for ABCA7 at the leading edge (, solid arrows). Affinity-purified preimmune serum showed no staining in macrophages. Colocalization studies in resting macrophages indicated partial overlap of ABCA7 with the Golgi marker Golgi 58K and an early endosome marker (rab4; unpublished data). The distribution of ABCA7 altered in macrophages undergoing phagocytosis with redistribution of ABCA7 into phagocytic cups (, top left) and enrichment in membrane ruffles (, top left, solid arrows). This redistribution of ABCA7 was not specific for phagocytosis of apoptotic cells, as it was also seen with IgG-coated latex beads fed to J774 macrophages (Fig. S2, bottom left, arrows; available at ). Based on the report that CED-7, a possible orthologue of ABCA7, may be required for the function of CED-1 () and the suggestion that LRP1 could have a function similar to CED-1 (), we also compared the localization and functions of ABCA7 and LRP1. Studies in “resting” macrophages showed very little colocalization of ABCA7 (, red) and LRP1 (green). However, confocal microscopy of macrophages (, open arrows) engulfing an apoptotic cell showed that ABCA7 (red) and LRP1 (green) colocalized within the phagocytic cup (solid arrow), as shown by the yellow in the overlay. Colocalization of ABCA7 and LRP1 was also observed within membrane ruffles after activation of macrophages by C1q (, bottom right). To test whether ABCA7 is of functional relevance in phagocytosis, we performed genetic knockdown experiments of ABCA7 in mouse peritoneal macrophages. Six different siRNA constructs complementary to ABCA7 were tested to down-regulate ABCA7. The best three constructs were used in subsequent experiments. Each of them suppressed ABCA7 by ∼60–80% (). The control siRNAs did not affect ABCA7 protein expression (unpublished data). shows that all three siRNA constructs for ABCA7 significantly inhibited the phagocytosis of apoptotic cells by 60–70%. Knock down of ABCA7 did not impair binding of apoptotic cells to macrophages (unpublished data). In some experiments, the second control siRNA (control siRNA2; , white column with horizontal lines) also slightly reduced the phagocytic index (PI). Therefore, for further experiments the first control siRNA (control siRNA1) was used. As it has been suggested that ABCA1 plays a role in phagocytosis of apoptotic cells (), we also compared the relative importance of ABCA7 and -A1 in this process. Surprisingly, the knock down of ABCA1 did not decrease the phagocytosis of apoptotic cells (), even though the siRNA constructs used for ABCA1 were as effective as the constructs used for ABCA7 (reduction of ABCA1 protein by 60–80%; Fig. S3, available at ). We further tested the effect of genetic suppression of ABCA7 on FcR-mediated phagocytosis. Even though ABCA7 localizes in phagocytic membranes during uptake of IgG-coated latex beads (Fig. S2), down-regulation of ABCA7 had no effect on FcR-mediated phagocytosis (). We have obtained mice with a targeted deletion of the ABCA7 gene in exon 21. In contrast to the recently published ABCA7 knockout mouse, created using a different targeting strategy (), we have not obtained any viable ABCA7 −/− mice. Although we do not know the reason for this discrepancy, a difference in the genetic background of the mice is unlikely, as we were unable to obtain ABCA7 −/− mice in either the C57Bl6 (87 animals screened) or in a C57Bl6 × 129 F2 hybrid background (23 mice screened: 7 +/+, 16 +/−, and 0 −/−). Thus, we used ABCA7 +/− mice for the experiments. As we demonstrated previously, peritoneal macrophages from ABCA7 +/− mice showed a 50–60% reduction in ABCA7 protein levels (). Phagocytosis of apoptotic neutrophils was reduced by 41% in macrophages from ABCA7 +/− mice (), and similar results were obtained with apoptotic Jurkat T cells (). In contrast, macrophages from ABCA7 +/− animals had no defect in the FcR-mediated phagocytosis of viable neutrophils () or IgG-coated latex beads (). We also compared the phagocytosis of apoptotic cells in ABCA1 +/+, +/−, and −/− macrophages. As shown in , no difference could be detected between the ABCA1 +/+, +/−, and −/− mice. In regard to ABCA1, our data contradict earlier work that reported a role of ABCA1 in the phagocytosis of apoptotic cells (; ). Therefore, we repeated the phagocytosis assays using a different microscopic method that counts red (prelabeled) apoptotic cells () within green-labeled LAMP1-positive phagolysosomes to score phagocytosis. Using this phagocytosis assay, we found that phagocytosis of apoptotic cells was reduced by 39% in ABCA7 +/− macrophages compared with ABCA7 +/+ control cells (), but ABCA1 −/− macrophages did not show a defect in phagocytosis (). Thus, using two different genetic approaches and two different microscopic phagocytosis assays, we found an essential role of ABCA7 in phagocytosis of apoptotic cells but not in FcR-mediated phagocytosis and no comparable function of ABCA1. A specific defect in the phagocytosis of apoptotic cells but not in FcR-mediated phagocytosis makes it unlikely that ABCA7 has a general role in the reorganization of the cytoskeleton. A significant general defect could be excluded further, as no difference in membrane ruffling was seen after stimulation of ABCA7 +/+ and +/− macrophages (Fig. S4, available at ). To test whether the clearance of apoptotic cells is also diminished in ABCA7-deficient mice, we used two established in vivo models of phagocytosis. In the first model (; ), we assessed the clearance of exogenously instilled apoptotic cells in the lungs of ABCA7 +/− and +/+ mice. Mice were challenged intratracheally with exogenous apoptotic cells, and clearance was assessed by bronchoalveolar lavage. In this model, defective phagocytosis of apoptotic cells is suggested either by decreased uptake by alveolar macrophages (i.e., decreased PI) or by increased recovery of apoptotic cells in the lavage (; ). ABCA7 +/− mice showed a significant reduction in the PI (). There was also a trend toward an increased recovery of apoptotic cells for ABCA7 +/− mice compared with their +/+ age-matched controls (). In the second in vivo model, ABCA7 +/− and +/+ mice were challenged intratracheally with lipopolysaccharide (LPS), and the inflammatory response and the clearance of endogenous apoptotic cells was assessed by bronchoalveolar lavage at day 0 (control group without LPS) and 3 d after LPS instillation. Without LPS (day 0), the PI was small and no significant difference between groups was observed (PI for ABCA7 +/+, 1.0 ± 0.6 [ = 6]; PI for ABCA7 +/−, 0.6 ± 0.4 [ = 4]). However, 3 d after LPS instillation, the PI was increased and was significantly smaller in the ABCA7 +/− mice compared with their +/+ age-matched controls (). Total numbers of cells in the bronchoalveolar lavage fluids as well as the differential white cell count at baseline and after LPS were similar for ABCA7 +/+ and +/− mice (unpublished data). In the experiments described so far, we have shown that ABCA7, like LRP1 (), is required for the phagocytosis of apoptotic cells but not for FcR-mediated phagocytosis. Further, we have shown that both ABCA7 and LRP1 are redistributed from an intracellular pool to the phagocytic cup (an extension of the plasma membrane) during phagocytosis (). We next asked whether there might be a defect in cell surface expression of ABCA7 and LRP1 during phagocytosis in ABCA7 +/− macrophages. As the movement of ABCA7 and LRP1 to the cell surface during phagocytosis is difficult to quantitate by microscopy, we developed a biochemical assay, i.e., cell surface biotinylation followed by pull down with streptavidin beads and immunoblotting for ABCA7 and LRP1. shows a 12-fold and sustained enrichment in cell surface levels of ABCA7 and a fourfold, more transient increase of cell surface LRP1 in ABCA7 +/+ macrophages after exposure to apoptotic cells. The enrichment of ABCA7 and LRP1 was markedly attenuated in ABCA7 +/− macrophages (2.5-fold for ABCA7 and 1.5–2-fold for LRP1; , representative result of three independent experiments). Overall levels of LRP1 and ABCA7 in cell lysates were not affected by the uptake of apoptotic cells (unpublished data). An early event during phagocytosis is binding of C1q on apoptotic cells to calreticulin/LRP1 on phagocytes, leading to signaling and uptake of apoptotic cells (). To see if the interaction of C1q with LRP1 was sufficient to induce cell surface expression of ABCA7, macrophages were serum starved and then exposed to C1q. Consistent with the earlier result showing C1q-induced appearance of ABCA7 in membrane ruffles (), this led to a substantial increase in cell surface expression levels of both ABCA7 and LRP1 (, representative result of six independent experiments). In ABCA7 +/− cells, the increase in ABCA7 was greatly reduced, and LRP1, although readily detectable in the cell surface, showed minimal response to C1q treatment (). This suggests that the increased cell surface expression of ABCA7 and LRP1 in membrane ruffles and during phagocytosis of apoptotic cells at least in part depends on signaling through LRP1 as well as on ABCA7 function. The failure to detect a substantial increase of ABCA7 and LRP1 in the cell surface in ABCA7 +/− cells after stimulation may in part result from a relatively high constitutive cell surface expression of these molecules. As activation of LRP1 by different ligands (; ) as well as C1q (unpublished data) leads to phosphorylation of ERK, we tested the signaling function of LRP1 by evaluating ERK phosphorylation after stimulation of macrophages with C1q. As shown in , after stimulation with C1q for 10 and 20 min, ERK phosphorylation was reduced by 57 and 65% in ABCA7 +/− macrophages, respectively (similar results were obtained in four experiments). In contrast, after stimulation of the FcR with aggregated IgG, the magnitude and the time course of ERK phosphorylation were very similar in ABCA7 +/+ and +/− macrophages (Fig. S5, available at ). Macrophages were also stimulated more physiologically by apoptotic cells (). Apoptotic cells caused a rapid increase in ERK phosphorylation in phagocytes that was sustained to the end of the experiment. Compared with wild-type cells, in ABCA7 +/− cells, basal levels of ERK phosphorylation were slightly higher, there was less initial increase in ERK phosphorylation, and the response was transient, returning to the baseline level after 20–40 min ( shows quantification of three independent experiments). To see whether attenuated ERK phosphorylation in ABCA7+/− macrophages could explain the decreased phagocytosis of apoptotic cells, the effect of two inhibitors of ERK phosphorylation on the phagocytosis of apoptotic cells was tested. Viability of macrophages was not affected by the use of inhibitors as assessed by annexin V staining (unpublished data). As shown in , both inhibitors reduced phagocytosis of apoptotic cells and C1q-coated apoptotic cells, whereas the effect on FcR-mediated phagocytosis was not significant. This may explain the reduced phagocytosis of apoptotic cells by ABCA7 +/− macrophages, as they show a defect in ERK phosphorylation (). Finally, we addressed the question of whether ERK phosphorylation is required for the increased ABCA7 and LRP1 expression at the cell surface upon stimulation with apoptotic cells. Inhibition of ERK prevented the increase of ABCA7 and LRP1 expression at the cell surface after stimulation with apoptotic cells (). At baseline (without addition of apoptotic cells), cell surface expression levels of ABCA7 and LRP1 were higher after inhibition of ERK phosphorylation, similar to findings in ABCA7 +/− macrophages, where basal cell surface expression of these molecules was relatively high, but showed only limited increase upon stimulation with apoptotic cells or C1q. The clearance of apoptotic cells occurs throughout the lifespan of multicellular organisms and is important for development during embryogenesis, the maintenance of tissue integrity and function, and the resolution of inflammation (). Here, we report that macrophage ABCA7 enhances the clearance of apoptotic cells in vitro and in vivo. ABCA7 and LRP1 move to the cell surface after stimulation with C1q or apoptotic cells and localize to membrane ruffles or phagocytic cups, respectively. However, ABCA7 also localizes to phagocytic membranes during FcR-mediated phagocytosis, in which ABCA7 levels are not rate limiting. More important, ABCA7 is required for optimal ligand-induced signaling through LRP1, as shown by C1q-induced ERK phosphorylation and for sustained ERK phosphorylation during phagocytosis of apoptotic cells. Finally, ERK phosphorylation itself is shown to be essential for phagocytosis of apoptotic cells but not for FcR-mediated phagocytosis, suggesting a link between the defect in ERK phosphorylation and the impaired phagocytosis of apoptotic cells observed in ABCA7 +/− cells. A variety of different lines of evidence, using siRNA and gene targeted mice, demonstrate a significant role of ABCA7 in the phagocytosis of apoptotic cells. A 60–80% reduction of ABCA7 protein by siRNA resulted in a 60–70% reduction in the phagocytosis of apoptotic cells, and a 50–60% reduction in ABCA7 in ABCA7 +/− mice led to a 40% reduction in the phagocytosis of apoptotic cells, indicating a major role of ABCA7 in this process. By comparison, macrophages treated with antibodies against LRP1, C1q, and mannose binding lectin or in C1q −/− mice, phagocytosis of apoptotic cells is reduced by ∼50–75% (; ). Furthermore, LPS-induced lung inflammation in CD44 −/− mice (; the same model used in this study []) resulted in a reduced PI, similar to that reported here. Surprisingly, parallel experiments could not confirm a role of ABCA1 in the phagocytosis of apoptotic cells, and this clearly contrasts with previous data (; ). The reason for the apparent discrepancy is unclear. It could be related to methodological issues. quantified uptake of apoptotic cells in macrophages from a small number of mice using Cr-labeled apoptotic cells. This method may not distinguish between apoptotic cells that have been ingested by phagocytes or are adherent to the phagocytes. Although more tedious, both microscopic methods used herein count only apoptotic cells clearly within phagocytes or LAMP1-positive phagolysosomes. identified ABCA1 as a phagosomal protein and suggest a potential role of ABCA1 for phospholipid efflux from the phagosomal compartment. Recent studies in our laboratory have shown that ABCA1 is induced in phagocytes ingesting apoptotic cells, suggesting that ABCA1 might have a role in the disposal of lipids during phagocytosis (). Together, these studies indicate that ABCA1, although not directly involved in the phagocytosis of apoptotic cells, may help to recover after phagocytosis. The failure to obtain ABCA7 −/− mice indicates a role of ABCA7 during embryonic development. It is plausible that a defect in the phagocytosis of apoptotic cells could lead to embryonic lethality, and several knockout mice for other proteins involved in the phagocytosis of apoptotic cells are lethal, notably the LRP1 and calreticulin knockout mice (; ). In contrast to our findings, a recent report by found a Mendelian genotype distribution in a different line of ABCA7 −/− mice. One difference in the targeting strategies is that our mice have a β-gal gene inserted into the locus so that production of any functional protein is unlikely, whereas used a replacement vector that could perhaps lead to the formation of small amounts of active protein products. The targeting vector in our mice is multicistronic with an internal ribosome entry site upstream of the LacZ sequence and, thus, a fusion protein of ABCA7 and Lac Z is unlikely to be formed. However, we cannot rule out completely the production of a dominant-negative ABCA7 peptide upstream of the LacZ-Neo-Cassette, which would not be detectable by our antibody. However, the proportionate reduction in the PI resulting from gene targeting and knock down by multiple siRNAs makes this unlikely. ABCA7 was shown to be important for ERK phosphorylation (), which in turn is required for efficient phagocytosis of apoptotic cells but not for FcR-mediated phagocytosis (). showed a role for both ERK and p38 MAPK signaling in the phagocytosis of apoptotic cells by Sertoli cells but no effect on the phagocytosis of latex beads. For FcR-mediated phagocytosis, ERK activation has been shown to be required in human neutrophils () but not in macrophages (). This suggests involvement of different MAPK signaling pathways depending on cell type and phagocytic mechanism. The requirement of ERK activation for phagocytosis may be related to the involvement of ERK in the control of actin reorganization (), the regulation of focal adhesion disassembly (), or cell spreading and migration (; ). The reduced ERK phosphorylation upon stimulation with C1q indicates defective signaling through LRP1. ERK phosphorylation after activation of LRP1 has been shown using thrombospondin binding to calreticulin and after stimulation of LRP1 with connective tissue growth factor (, ; ). As C1q also binds to calreticulin, the signaling pathway may be similar to the one reported for thrombospondin, calreticulin, and LRP1. Stimulation of macrophages with apoptotic cells increases ABCA7 and LRP1 levels in the cell surface of ABCA7 +/+ macrophages, and this response is markedly reduced in ABCA7 +/− cells (). Similarly, stimulation of LRP1 with C1q results in an enrichment of ABCA7 and LRP1 in the cell surface (), and this increase is almost abolished in ABCA7 +/− macrophages. ABCA7 may be involved in the trafficking of ABCA7 and LRP1, and this may depend at least in part on ERK phosphorylation as indicated by the abolished increase of ABCA7 and LRP1 in the cell surface after inhibition of ERK phosphorylation (). The increase of ABCA7 and LRP1 in ABCA7 +/+ macrophages and its failure in ABCA7 +/− cells raises the possibility of a positive-feedback loop. That is, ABCA7 enables maximal LRP1-dependent ERK phosphorylation, which in turn leads to increased expression of ABCA7 and LRP1 at the cell surface, further increasing LRP1 signaling. This model indicating that ABCA7 is required for optimal function of LRP1 may also explain why ABCA7, analogous to LRP1 (), is only required for the phagocytosis of apoptotic cells but not for FcR-mediated phagocytosis. On a more fundamental level, the putative role of ABCA7 in signaling events during the phagocytosis of apoptotic cells may be related to an underlying lipid translocase activity. Based on our previous finding that ABCA7 can promote efflux of phosphatidylcholine and sphingomyelin from cells to apoA-I (), we hypothesize that ABCA7 could be involved in the translocation of phospholipids from the inner to the outer leaflet of the plasma membrane, as it has been shown for ABCB4 or MDR2 (multidrug resistance 2; ). Such a “floppase” activity could lead to the formation of specialized membrane microdomains, which may be important for the clustering of receptors and signaling molecules such as LRP1 in the phagocytic cup. In summary, the data herein show that ABCA7 but not -A1 is required for efficient phagocytosis of apoptotic cells. The functional properties of ABCA7, together with the sequence similarity to CED-7, suggest that ABCA7 may be a mammalian orthologue of CED-7. Mechanistically, ABCA7 is important for maximal and sustained ERK phosphorylation, which in turn is shown to be essential for the phagocytosis of apoptotic cells. At least in part this is explained by the role of ABCA7 in optimal ligand-induced signaling through LRP1 (). These findings shed new light on the fundamental issue of the phagocytosis of apoptotic cells in mammals. 1 mg/ml of purified human C1q was obtained from Quidel Corp. The antibody against ABCA7 has been described previously (). Antibody against ABCA1 was from Novus. The antibody against LRP1 for immunoblotting was provided by J. Herz (University of Texas Southwestern Medical Center, Dallas, TX). The anti–integrin β1 subunit antibody was provided by E.E. Marcantonio (Columbia University, New York, NY). For fluorescence microscopy of LRP1, monoclonal antibodies 5A6 and/or 8B8 (Molecular Innovations) were used. Rat anti–mouse monoclonal antibody (clone 1D4B) against LAMP1 was obtained from BD Biosciences. Antibody against phospho-ERK1/2 and total ERK were obtained from Cell Signaling. For the siRNA experiments, C57/B6 mice from The Jackson Laboratory or Taconic were used. ABCA1 heterozygous mating pairs were obtained from The Jackson Laboratory. ABCA7-null/LacZ knockin heterozygous mice were provided by G. Gao (Lilly Research Laboratories, Indianapolis, IN). Originally, these mice were generated by Deltagen, Inc. For in vivo experiments, mice were in C57/B6 background (total number of backcrosses, = 10). ABCA7 +/− mice were obtained by breeding ABCA7 +/+ females with ABCA7 +/− males. The genotype distribution in the offspring was as expected with ∼50% ABCA7 +/+ and 50% ABCA7 +/−. Animal protocols were approved by the Institutional Animal Care and Use Committee of Columbia University and the National Jewish Medical and Research Center, respectively. Thioglycollate-elicited mouse peritoneal macrophages were plated in 24-well plates and maintained in DME with 10% (vol/vol) heat-inactivated FBS, penicillin, and streptomycin under a humified atmosphere of 90% air and 10% CO. Control siRNA1 was designed by scrambling the sequence of a siRNA targeting mouse ABCA7 (control siRNA1, 5′-AAAACTCCGACTACCGAAACT-3′), and control siRNA2 was obtained from QIAGEN (control siRNA2, 5′-AATTCTCCGAACGTGTCACGT-3′). Three siRNA targeting mouse ABCA7 (mABCA7) were used (mABCA7 siRNA1, CAGGGACTTGACCAAGGTTTA; mABCA7 siRNA2, 5′-GCCTTCCTAGCTATGCAGACT-3′; mABCA7 siRNA3, 5′-AAGGCCGTGGTGCGTGAGAAA-3′) and two siRNA targeting mouse ABCA1 (mABCA1) were used (mABCA1 siRNA1, 5′-TCGGTTGACATCATTAAATAT-3′; mABCA1 siRNA2, 5′-CTGGATGTATAATGAGCAGTA-3′). Transfection with siRNA diluted in Opti-MEM I at a concentration of 180 nmol was performed using Oligofectamine (Invitrogen). 1 d after transfection, medium was changed. If RNAi was done for ABCA1, cells were grown in the presence of 2 μM TO901317 (Sigma-Aldrich) for the last 48 h. Healthy adult human subjects donated 400 ml of whole blood under a protocol approved by the National Jewish Medical and Research Center's institutional review board. Neutrophils were separated as described previously (). Before phagocytosis, the cells were irradiated at 312 nm (Fotodyne, Inc.) for 10 min and then cultured in RPMI with 1% BSA at 5 × 10 cells per ml at 37°C plus 5% CO for 2–3 h to induce apoptosis. Immunoblot analysis was performed as described previously (). Relative intensities of bands were determined by densitometry. For RNAi experiments, macrophages were plated at 0.5 × 10 cells per well, and for all other experiments, cells were plated at 0.1–0.2 × 10 per well (24-well plates). Phagocytosis experiments were done 2 d after RNAi or 2 d after harvesting of cells (if no RNAi was performed). If not otherwise indicated, phagocytosis assays were performed with human neutrophils. For some experiments, Jurkat T cells were used. For experiments with C1q-coated apoptotic cells, Jurkat T cells were incubated with 10 μg/ml C1q followed by spin down and uptake of cells in new medium. Uptake conditions and assessment were almost identical as previously described (). In brief, apoptotic cells were taken up in DME with 10% FBS and 10 cells were added to macrophages in 24-well plates. After 90 min, cells were washed twice in PBS before they were fixed and stained with modified Wright's Giemsa. Using 40× light microscopy, macrophages were examined for uptake of apoptotic cells by counting triplicate or quadruplicate wells (200 macrophages/well) in a blinded fashion. The PI was calculated as the number of cells ingested per the total number of macrophages × 100. For measurement of uptake of apoptotic Jurkat T cells into LAMP1-positive phagolysosomes, Jurkat T cells were prelabeled with CellTracker red (Invitrogen) according to the manufacturer's instructions before induction of apoptosis by UV radiation, and apoptotic cells were incubated with macrophages for 45 min before samples were stained as indicated in the following section. For FcR-mediated phagocytosis, neutrophils were incubated with mouse anti–human CD18 monoclonal antibody (Immunotech), 10 μl per 10 cells (0.2 μg/μl), for 2–3 h. If Jurkat T cells were used they were coated with mouse anti–human CD3 (BD Biosciences). Phagocytosis assays with IgG-coated 4-μm Aldehyde/Sulfate latex beads (Invitrogen) were performed as previously reported (). J774 macrophages were used for the experiments using PD 98059 (preincubation for 20 min, 20 μM) and U 0126 (preincubation for 20 min, 10 μM). For ruffling experiments, macrophages were incubates for 4–5 h in serum-free DME medium before stimulation with 0.05 μg/ml macrophage inflammatory protein 1 α (PeproTech) for 5 min. Quantification of ruffling was performed as described previously (). For immunofluorescent microscopy, glass-bottomed cover dishes were used. Experiments were performed as for phagocytosis assays. After washing with PBS, cells were fixed with 3.2% paraformaldehyde, permeabilized with 0.1% Triton X-100, blocked with 10% goat serum in PBS, and incubated with peptide affinity-purified rabbit anti-ABCA7 antibody and/or mouse monoclonal anti-LRP1 antibodies at a dilution of 1:100 for 1 h at room temperature or at 4°C overnight. As a secondary antibody, goat anti–rabbit Cy3 (Jackson ImmunoResearch Laboratories) and/or goat anti–mouse Alexa 488 (Invitrogen) were used. The experiments using latex beads were performed with J744 macrophages. Beads were added for 20 min at 37°C. Confocal pictures were taken at room temperature with imaging medium PBS using a multiphoton, upright confocal microscope (LSM 510 NLO; Carl Zeiss MicroImaging, Inc.) with a 63×/0.9 water lens (primary acquisition software obtained from Carl Zeiss MicroImaging, Inc.). For fluorescent microscopy, Axiovert 200 (Carl Zeiss MicroImaging, Inc.) with a Plan-Neofluar 63×/1.25 lens was used attached to a camera (Orca ER; Hamamatsu). Brightness was adjusted with Photoshop 6.0 (Adobe). In vivo phagocytosis assays were performed exactly as described by . For LPS administration, 0111: B4 LPS (List Biological Laboratories) was suspended in PBS at a concentration of 2.5 mg/ml. 200 mcg of LPS was instilled intratracheally. 72 h later, bronchoalveolar lavage was performed. Thioglycollate-elicited mouse peritoneal macrophages were plated in 12-well plates at 1.5 × 10 cells/well. The medium was changed the next day. After 2 d, cells were incubated with apoptotic Jurkat T cells for the indicated time points followed by three washes with ice cold PBS. For ERK phosphorylation assays with clinical grade, 100 μg/ml of aggregated IgG macrophages were cultured for the last 18 h in the presence of 100 U/ml γ interferon followed by incubation for 3–4 h in serum-free DME. For C1q experiments, macrophages were incubated for 3–4 h in serum-free DME before stimulation. Biotinylation was performed with 0.5 mg/ml Sulfo-NHS-SS-Biotin (Pierce Chemical Co.) for 30 min on ice. Cells were lysed on ice, and the lysis buffer was supplemented with Tyr and Ser/Thr phosphatase inhibitor cocktail (Upstate Biotechnology). 100–200 μg of cell lysates were used for pull down with 25 μl of streptavidin beads (Pierce Chemical Co.) followed by two washes in lysis buffer. 25 μg of protein was used for immunoblotting according to the protocol provided with the ERK1/2 antibody. Data are expressed as means and SD if not otherwise stated. Significance of differences was always calculated with a two-sided unpaired test. For box blots, StatView was used. For pairwise alignments, the program NEEDLE with scoring matrix EBLSOUM40 was used, and for multiple alignments, the ClustalW program was used. Fig. S1 depicts the alignment of CED-7, mouse ABCA7, and ABCA1 amino acid sequences. Fig. S2 shows localization of ABCA7 in phagocytic cups/phagosomes during FcR-mediated phagocytosis. Fig. S3 shows a representative immunoblot for ABCA1 and β-actin after knock down of ABCA1 in peritoneal macrophages. Fig. S4 shows membrane ruffling with quantification of ABCA7 +/+ and +/− macrophages after stimulation. Fig. S5 shows immunoblot and quantification of ERK phosphorylation after stimulation of ABCA7 +/+ and +/− macrophages with aggregated IgG. Online supplemental material is available at .

Plectin is an intermediate filament (IF)–associated cytolinker protein of very large size (M > 500,000) and with versatile functions (for review see ). It is expressed in a wide variety of mammalian cells and tissues and has important functions in mediating interactions between different cytoskeletal network systems and their anchorage at cell–cell and cell–matrix junctional complexes. Plectin's IF-binding site was mapped to a stretch of ∼50 amino acid residues linking two (5 and 6) of its six C-terminal repeat domains. At its N terminus, plectin harbors a functional actin-binding domain that also serves as an integrin β4- and an additional vimentin-binding site (). Plectin's association with microtubules and their cross-linking to vimentin IFs has been previously reported (); however, the molecular mechanism of this interaction is still unknown. There is increasing evidence that, apart from acting as a cytoskeletal linker protein, plectin serves an important function as a scaffolding platform of proteins involved in cellular signaling. Strong support for this idea comes from the recent identification of several novel interaction partners of plectin with links to signaling, such as the nonreceptor tyrosine kinase Fer and AMP kinase, and the observation that plectin deficiency affects their enzymatic activities (; ). The scaffolding function of plectin was recently confirmed when it was shown that the protein binds and sequesters the receptor for activated C kinase 1 (RACK1) to the cytoskeleton, and thereby affects PKC signaling pathways (). This, together with previous studies showing that plectin is involved in the regulation of actin filament dynamics and influences Rho/Rac/cdc42 signaling (), suggests that plectin, and likely cytolinkers in general, provide a crucial link between cytoskeleton dynamics and signaling machineries. The prominent localization of plectin at hemidesmosomes (HDs), desmosomes, Z-line structures and dense plaques of striated and smooth muscle, intercalated discs of cardiac muscle, and focal contacts implied a role for the protein in linking the cytoskeleton to plasma membrane junctional complexes (). This was supported by studies showing that defects in plectin expression lead to the skin disease epidermolysis bullosa simplex (EBS). In these patients, as well as in plectin-deficient mice generated by targeted gene inactivation (), the link of the keratin cytoskeleton to HDs was dramatically affected (), with keratin filaments appearing less tightly bundled at their insertion into the inner plate structure of HDs. Moreover, earlier ultrastructural studies and in vitro reconstitution of IFs had shown that plectin was primarily located at branching points of IFs (). However, the important question of whether plectin, indeed, affects IF network cytoarchitecture and its dynamics has not been addressed so far, except for a study showing that a recombinant fragment containing plectin's C-terminal IF-binding site inhibits IF formation in vitro in a dose-dependent manner (). Investigating what impact plectin deficiency has on the organization and dynamic properties of IFs in keratinocytes, we found a direct link between plectin-controlled IF cytoarchitecture and MAP kinase signaling cascades involved in cell migration and stress response. To investigate whether plectin plays a role in IF network organization, we first compared the general appearance of such networks in plectin wild-type (ple [+/+]) and plectin-deficient (ple [−/−]) basal keratinocytes using immunofluorescence microscopy. In interior cytoplasmic regions, both primary and p53-deficient ple (+/+) keratinocytes exhibited dense networks of keratin filaments. The filaments showed partial colocalization with discontinuous plectin-positive structures, which in superimposed images often resembled beads on a string (). However, at the cell margins, hardly any IFs were found, leaving a peripheral ring–shaped, filament-free zone densely packed with plectin structures. Keratin filaments extended to the inner circumference of this ring (, A and G, arrows), as if peripheral plectin acted in an inhibitory manner on filament extension. In support of this view, the keratin network system of wild-type cells seemed to be more densely packed around the cell center compared with that of ple (−/−) cells, where it extended further to the cell periphery (, compare B and H with E and K, respectively). This phenotype was noticed independent of whether the anti-pan keratins K5, K6, and K18 or anti-K5 or K6 antibodies alone were used (; unpublished data). Furthermore, particularly in subconfluent cell cultures (∼60% confluence), ple (−/−) cells exhibited significantly enlarged mesh size of keratin networks at their periphery compared with wild-type cells (, compare B with E and H with K, along with their corresponding magnifications). A quantitative analysis of subconfluent cell populations revealed that >70% of ple (−/−), but only <8% of ple (+/+), keratinocytes displayed such a phenotype. Regarding expression levels of different keratins, no differences were observed between ple (+/+) and (−/−) keratinocytes, as revealed by immunoblotting analysis (Fig. S1, available at ). Interestingly, the vimentin filament network of plectin-deficient fibroblasts showed similar characteristics. It appeared more bundled and less delicate compared with wild-type cells, creating a network with wider spaces between the filaments (unpublished data). In both ple (+/+) and (−/−) cells, actin filaments exhibited a subcortical organization typical of subconfluent keratinocyte cultures (, M and P; ). However, ple (−/−) keratinocytes showed a slight increase in the number of stress fibers and, moreover, were extending small lamellipodia (, arrows), which is indicative of an ongoing transition from a stationary to a migratory phenotype. Furthermore, no gross differences were observed between ple (+/+) and (−/−) cells in the organization of microtubules (). Peripheral integrin (INT) α6 clusters were formed in both cell types (), and in ple (+/+) cells, partial colocalization with plectin was observed (Fig. S2, A–C, available at ). In addition, keratin staining partially overlapped with the ring formed by integrin clusters (Fig. S2, D–F). Transmission electron microscopy revealed a prominent lateral bundling of keratin filaments in peripheral regions of ple (−/−) cells, resulting in larger spaces between individual and bundled filaments (). In some regions, filaments showed a complete lateral collapse, appearing as one massive filamentous bundle (, arrow). Bundling to such an extent was never observed in ple (+/+) cells, where corresponding areas were dominated by numerous fine, and apparently short, filamentous structures, with orthogonal cross-bridges filling the space between the filament bundles (). To demonstrate that the observed changes in keratin network organization were directly connected to plectin deficiency, the phenotypic rescue potential of plectin was examined by transient transfection of ple (−/−) keratinocytes with an expression plasmid encoding full-length mouse plectin isoform 1a, which is the major isoform normally expressed in this type of cells (). Similar to endogenous plectin of corresponding wild-type cells (), reexpressed plectin 1a showed a punctated distribution pattern (). A quantitative analysis revealed that 90% of ple (−/−) cells expressing plectin 1a at significantly high levels displayed keratin networks with mesh sizes as small as those of ple (+/+) cells (). This strong rescue potential of plectin 1a clearly demonstrated a direct link between the absence of plectin and the observed alterations of the keratin network cytoarchitecture. Osmotic and heat-shock assays serve as useful tools in monitoring keratin network properties and their alterations in EBS caused by keratin mutations (; ). Therefore, we examined whether the altered keratin filament cytoarchitecture of ple (−/−) keratinocytes affected their response to changes in osmolarity and temperature. Upon urea treatment, INTα6-positive retraction fibers, which are a characteristic feature of migrating and mitotic cells (), became very pronounced at the rear of both cell types (). Most probably, the general shrinkage of the cells, which started shortly after exposure to urea (), triggered the elevated formation of these structures. INTα6-positive retraction fibers of ple (−/−) cells, however, were significantly longer than those of ple (+/+) cells, possibly reflecting a stronger shrinkage of plectin-deficient cells (). In accordance with recent studies showing increased urea-induced bundling of keratin filaments (), the IF network appeared less filamentous after urea treatment and accumulated around the cell center, indicating that it had collapsed onto the nuclei (). Interestingly, at the leading edge of cells, which is an area devoid of retraction fibers, keratin bundles of ple (+/+) cells displayed a regular form and were closely associated with peripheral integrin clusters, whereas in ple (−/−) cells filament bundles were conspicuously tangled and distant from the cell periphery (). The appearance of such “torn” filaments in ple (−/−) cells suggested a reduction in their INTα6β4 anchorage. This would be consistent with the reduced attachment of keratin filaments to HDs reported for EBS patients (). To address this issue biochemically, we prepared cytokeratin-enriched cell fractions from ple (+/+) and (−/−) keratinocytes grown on collagen I and compared their INTα6β4 content using antibodies to INTβ4. As expected, such cell fractions were highly enriched in keratins 5 (unpublished data) and 14 () and, in the case of wild-type cells, contained considerable amounts of plectin. INTβ4 was, however, completely absent from cytokeratin fractions of ple (−/−) cells. The absence of INTβ4 from the cytokeratin fraction of ple (−/−) keratinocytes correlated well with the reduced number of HD-like structures found in these cells (, compare O with R; ). If, in the absence of plectin, the rigidity of IF networks is reduced and filaments are more loosely bound, or not bound at all, to the outer membrane, one might expect the network to be disassembled more readily in ple (−/−) compared with ple (+/+) cells. To test this, we monitored by immunofluorescence microscopy the kinetics of IF disassembly upon treatment of ple (+/+) and (−/−) cells with the serine/threonine phosphatase inhibitor okadaic acid (OA), which is known to selectively cause the disruption of IFs (). After a 2-h treatment of keratinocytes, the well-spread keratin network had formed thick bundles of filaments that seemed to be retracting toward the nucleus (, arrowheads). At later time points (4 and 6 h), a progressive breakdown of filaments was observed, with keratin granules forming first at the cell periphery (, arrowheads), followed by their collapse into a dense perinuclear ring (, arrow) that eventually became fragmented into numerous granules of various sizes (, arrow). The initial bundling of filaments appeared to occur more efficiently in ple (−/−) cells, as indicated by the increased mesh size of keratin networks visualized in the majority of these cells compared with ple (+/+) cells (, arrows). Moreover, at the 4- and 6-h time points, the proportions of cells with keratin granules were significantly higher in ple (−/−) compared with ple (+/+) cells (). For a statistical analysis of keratin filament disassembly, cells were classified into three categories (1–3), where category 1 represented cells with no granules, category 2 represented cells with granules and residual filaments (, cell marked with asterisk), and category 3 represented cells in which complete granulation, including that of the perinuclear ring structure, had occurred (, cell marked with arrow). At the 2-h time point, ∼15% of ple (−/−) cells already fell into category 2, whereas only a mere 4% of ple (+/+) cells did (). After 4 h, the majority of the ple (−/−) cell population had their keratin network either partially (30%) or completely (50%) disassembled, whereas ∼70% of ple (+/+) cells were still without any granules (). After 6 h, the difference between ple (+/+) and (−/−) cells became less pronounced, but ∼30% of ple (+/+) cells still did not show any granules, versus only ∼7% in the case of ple (−/−) cells (). Based on this, we concluded that the disassembly of IFs upon OA treatment is significantly accelerated in ple (−/−) compared with wild-type keratinocytes. In agreement with a previous study (), even after 6 h of OA treatment, no disruption of microtubules or microfilaments was observed. In fact, actin stress fiber formation was found to be significantly increased in ple (−/−) cells, and to some extent, also in ple (+/+) cells (unpublished data). Plectin itself appeared to dissociate from IFs upon OA treatment of keratinocytes. Immunofluorescence microscopy revealed a strong increase in nonfilamentous (diffuse) plectin staining in keratinocytes at the 2-h time point (Fig. S3 A, available at ), whereas no colocalization with residual filaments nor with keratin granules was observed at later time points (; unpublished data). Consistent with this, in cell fractionation experiments, plectin was no longer detectable in the cytokeratin fraction beyond the 2-h time point (Fig. S3 B). Monitoring detergent-soluble keratin pools during OA treatment of keratinocytes, we found the level of soluble keratin proteins to already be elevated by approximately twofold in plectin-negative compared with wild-type cells before drug treatment (, 0 h). This difference further increased to approximately threefold within 2 h of drug treatment. Thereafter, keratin solubility in ple (−/−) cells stayed about level, while that in ple (+/+) cells further increased, approaching a level similar of that of ple (−/−) cells (, 4 h). Previous studies have shown stress-activated p38 MAP kinase to be one of the major candidates for mediating the effects of OA on vimentin and keratins (; ). Therefore, we examined whether the OA-induced changes in network appearance and solubility of keratins were paralleled by changes in p38 activity. Using anti–phospho-p38 antibodies to monitor the activation status of p38 kinase, we found no significant differences between ple (+/+) and (−/−) keratinocytes under basal conditions (). Upon addition of OA to ple (−/−) cells, a moderate increase in p38 activity during the first hour was followed by a steep increase during the second hour and a sharp decline thereafter. In contrast, p38 kinase activity levels in ple (+/+) keratinocytes showed only a modest increase during the first 2 h, staying constant thereafter. Thus, the maximum of p38 activity measured in plectin-negative cells was more than twice as high as in wild-type cells. To assess whether the elevated response of p38 kinase to OA in ple (−/−) cells was specific, the activity of another MAP kinase, Erk1/2, was monitored in a similar fashion. Unexpectedly, we found the basal phosphorylation of Erk1/2 kinases to be already significantly elevated in ple (−/−) keratinocytes () compared with ple (+/+) cells. Erk1/2 activities in ple (+/+) and (−/−) cells showed a similar response to OA treatment, however, reaching maxima after 1 h, followed by a decrease to levels below those observed before the treatment (). Hence, the accelerated OA-induced activation of p38 kinase in plectin-deficient, compared with wild-type, keratinocytes seemed to be specific for this MAP kinase, correlating with the observed tendency for faster keratin solubilization in these cells. Plectin-mediated attachment of the keratin cytoskeleton to INTα6β4 has been shown to play a crucial role in stabilizing adhesion of keratinocytes to the matrix, thereby inhibiting cell migration (). Plectin-deficient keratinocytes, showing no association of INTα6β4 with keratins (), together with their up-regulation of Erk1/2 (see previous section), which is a kinase that positively regulates keratinocyte migration (), prompted us to compare the migratory potentials of ple (+/+) and (−/−) keratinocytes using an in vitro wound-healing assay. Average migration distances measured for ple (−/−) cells were almost twice as long as those of ple (+/+) cells (). Interestingly, the mesh size of the keratin network in ple (−/−) keratinocytes along the wound edge was much larger compared with that of cells at a distance from the wound and, in these regions, differences to the keratin network of ple (+/+) cells became most prominent (). This was consistent with our finding that an increased keratin network mesh-size characteristic of ple (−/−) keratinocytes was particularly evident in subconfluent cell cultures (). Furthermore, in migrating wound-edge keratinocytes, plectin's localization changed from basal integrin cluster– to keratin filament–associated (), highlighting the importance of plectin in organizing keratin cytoarchitecture during cell migration. Because we had observed a correlation between the enhanced migration of ple (−/−) keratinocytes and the up-regulation of Erk1/2, we asked whether pharmacological inhibition of MEK1/2, which are the upstream kinases of Erk1/2, would decrease migration of ple (−/−) keratinocytes. As shown in , when ple (−/−) keratinocytes were exposed to PD98059, which is a specific inhibitor of MEK1/2 activities, their migration distances were reduced by almost a factor of two, bringing their level close to that of untreated wild-type cells. Wild-type cells showed a similar drug response. The observed decrease in migration of drug-treated cells directly correlated with inhibition of Erk1/2 activities, as demonstrated by analysis of Erk1/2 phosphorylation (). A similar analysis showed that other MAP kinases, in particular JNK, were unaffected under these conditions (unpublished data). Importantly, although MEK1/2 inhibition decreased the migration rate of ple (−/−) keratinocytes, it had no effect on their aberrant keratin network organization (), clearly placing plectin in the MAP kinase cascade upstream of Erk1/2. These data established a causal relationship between plectin deficiency and accelerated migration of keratinocytes, showing hyperactivation of Erk1/2 to be a consequence of plectin deficiency. PKCδ and c-Src both have been suggested as major players in signaling pathways responsible for migration of keratinocytes (; ), and both have been shown to be upstream activators of Erk1/2 (; ). Therefore, we next investigated activation of these kinases in membrane and cytosolic fractions of ple (+/+) and (−/−) keratinocytes. As shown in , both, PKCδ and c-Src kinase, exhibited increased phosphorylation (corresponding to higher activities) in the membrane fraction of ple (−/−) keratinocytes compared with wild-type cells. Although total c-Src levels in membrane and cytosolic fractions from both cell types were comparable, those of PKCδ were lower in the membrane fraction of ple (−/−) compared with ple (+/+) cells. In the cytosolic fractions, total PKCδ signals were hardly detectable in any of the two cell types. To assess whether up-regulation of c-Src was related to enhanced Erk1/2-dependent migration of ple (−/−) keratinocytes, cells were treated with the Src family kinase inhibitor PP2, a suppressor of cell motility and Erk activation (), before scratch wound assays. As expected, this treatment led to suppression of cell motility (unpublished data) and a dose-dependent decrease in phosphorylation (activity) of Erk1/2 (). In a previous study, we revealed a role of plectin as a cytoskeletal regulator of PKC signaling and possibly other signaling events (). We proposed that plectin sequesters RACK1, which is a receptor and scaffolding protein of activated PKC and a direct binding partner of plectin, to the cytoskeleton when PKC is inactive. Because of the lack of its cytoskeletal docking site in the absence of plectin, in plectin-deficient cells RACK1 accumulates (together with PKC) at the periphery of cells, similar to the situation in wild-type cells after activation of PKC. According to this model, one may expect that the forced expression of RACK1 in wild-type keratinocytes mimics the situation in ple (−/−) cells, leading to their characteristic phenotypes. To test this we analyzed the migration potential of keratinocytes expressing an EGFP-RACK1 fusion protein using time-lapse video microscopy. In accordance with migration distances measured in scratch wound closure assays (), ple (−/−) keratinocytes displayed a migration velocity (1.58 μm/min) approximately two times as high as that of ple (+/+) cells (0.82 μm/min), when observed 2–6 h after plating (not depicted). As shown in (controls), 14–18 h after plating, ple (−/−) cells migrated three times as fast as ple (+/+) cells (1.53 vs. 0.49 μm/min). Expression of EGFP-RACK1 led to an increase in the average migration rates of both cell types. Transfected wild-type cells (1.30 μm/min) migrated 2.6 times faster than untransfected control cells, reaching 85% of the speed of untransfected ple (−/−) cells, whereas transfected ple (−/−) cells (2.02 μm/min) migrated 1.3 times faster than their untransfected counterparts. Similar experiments were performed with the cytoplasmic nonreceptor tyrosine kinase Fer, which, like RACK1, directly binds to plectin and thereby is inhibited in its activity (). In this case the speed of wild-type cells was increased by approximately twofold (unpublished data). In contrast, expression of an EGFP-plectin isoform 1a (full-length) fusion protein in ple (−/−) cells led to a significant slowdown of the cells, reducing their average speed to 0.98 μm/min (). This was equivalent to a slightly >50% rescue potential of the fusion protein, taking the values of control ple (+/+) and (−/−) cells into account. The lower rescue potential of plectin 1a in this assay compared with restoration of keratin network cytoarchitecture () may reflect the requirement of other major isoforms expressed in keratinocytes, such as plectin 1c and 1 (), for full phenotype restoration. Thus, whereas overexpression of plectin-controlled signaling proteins, such as RACK1, led to downstream mechanisms boosting cell motility (), reexpression of a major plectin isoform in ple (−/−) cells led to the partial reversal of their aberrant migration. This study provides evidence that plectin plays a crucial role in the appropriate organization of IF networks in keratinocytes. In plectin's absence, these networks are less delicate, their mesh size is increased, and individual filaments appear bundled and straighter. Plectin may control the properties of other IF systems as well, as alterations of IF networks resembling those described here for keratinocytes were also observed in plectin-deficient fibroblasts (unpublished data). Furthermore, a recent study suggests that plectin regulates both the organization and solubility of GFAP in astrocytes (). Based on ultrastructural analysis, we suggest that the mechanism behind the observed phenotype is a reduction of orthogonal cross-linkages between individual keratin IFs in the absence of plectin. In cross-linking individual filaments at high angles, plectin's interaction with IFs would resemble that of filamin with microfilaments. Such a mode of action would be consistent with earlier studies showing plectin to be predominantly localized at crossover and branching points of IFs (). Similar to filamin, plectin was shown to exist in dimeric and tetrameric states. Tetrameric structures are assumed to be formed by antiparallel alignment of two parallel plectin dimers (). Thus, exposed C-terminal IF-binding sites on both ends of such structure have the ability to cross-link two filaments. Even individual parallel dimers may have a cross-linking capacity, as an additional N-terminal vimentin-binding site resides in the N-terminal actin-binding domain of plectin (). Oligomers of plectin tetramers, which are generated by the head-to-head fusion of dumbbell-shaped plectin molecules (), have been shown to form at IF branching points (). Cross-linking functions of plectin have also been clearly demonstrated by , who used immunogold electron microscopy to show that plectin is organized in millipede-like structures around the core of individual IFs, with plectin sidearms frequently making Y contacts with each other. Conceivably, such Y-shaped structures may emerge from head-to-head fusion of plectin molecules, allowing cross-linking at high angles. Moreover, these authors reported that plectin was not localized regularly all along IFs, but was more concentrated at their distal ends, which is consistent with our finding that defects in IF organization were most prominent in the peripheral regions of ple (−/−) cells. The urea-based osmotic shock assay revealed a stronger stress response of ple (−/−) versus (+/+) keratinocytes, reflected by increased cell shrinkage and considerably longer retraction fibers. Suggesting an increased plasticity of ple (−/−) cells, these data also implied a role of plectin in stabilizing membrane surfaces. By interlinking IFs into properly organized three-dimensional networks, and by connecting these to the plasma membrane, plectin probably provides cell membranes with the required resistance against deformations, such as those induced by osmotic shock. A similar model has been proposed by to explain the increased deformability of filamin-deficient melanoma cells. Thus, plectin and filamin could have similar modes of action in respect to different cytoskeletal filament systems. We assume that plectin increases the stiffness of IFs by introducing orthogonal cross-links between filaments and, thus, acts as a stabilizer opposing their disassembly. Moreover, the protein acts as a linker, anchoring keratin filaments to hemidesmosomal INTβ4, as shown on ultrastructural and biochemical levels (; ). Further support for an IF-stabilizing role of plectin stems from differences in ple (+/+) and (−/−) keratinocytes during the early stages of OA-induced disassembly of IFs, which in wild-type cells correlates with the dissociation of plectin from keratin IF networks. The events after OA treatment can be viewed to parallel those of a more physiological process, i.e., mitosis, as in both cases plectin dissociates from IFs during their disassembly (). Thus, it seems that the release of stabilizing proteins such as plectin is a requirement for the efficient disassembly of IFs. However, if the faster IF disassembly in ple (−/−) cells was strictly caused by the diminished mechanical stability of IFs, why did we observe a higher increase in p38 activity in these cells after OA-induced IF disassembly? It is unlikely that faster IF disassembly was caused by elevated levels of p38 kinase activity because, in this case, increased phosphorylation of keratins in ple (−/−) compared with wild-type cells should have been detected using p38 kinase site-specific phosphokeratin antibodies (). This, however, was not the case (unpublished data). Furthermore, there were no differences in the activity levels of p38 kinase between wild-type and ple (−/−) cells under basal conditions. Recently, an association of simple epithelial keratins 8/18 with Raf kinase and its disruption after treatment of cells with OA has been reported (). Based on this, the authors suggested a role of keratins in sequestering a population of Raf and thereby regulating its signaling potential. In a similar fashion, keratins might regulate either p38 directly or one of its upstream effectors. Thus, we speculate that it is the faster disassembly of IFs and larger pool of their soluble subunits proteins in ple (−/−) compared with ple (+/+) cells that, via an unknown positive feedback mechanism, affects the activity of p38, rather than the other way around. It is the currently accepted view that ligation of integrins triggers the activation of Erk, via either the adaptor protein Shc or some other mechanism (). This has been shown to be the case for integrins of fibroblasts, as well as the integrin specific for keratinocytes, α6β4 (). In some studies, the importance of integrin linkage to the cytoskeleton has been emphasized. For instance, in fibroblasts and skeletal muscle cells, it was shown that the disruption of microfilaments with cytochalasin D blocked the integrin-mediated activation of MAP kinases (). Similarly, a mutation in INTβ4 that prevents plectin–INTβ4 interaction leads to accelerated migration of keratinocytes (). Fully in line with these studies, we demonstrate that plectin-deficient keratinocytes, showing no association of INTα6β4 with keratins, have an elevated migration potential. Most importantly, the migration of plectin-deficient cells was significantly reduced when plectin was reexpressed. An involvement in migration has recently been reported for another cytolinker family member, kakapo/short-stop (). By regulating Notch receptor localization and activity, short-stop was shown to be essential for the movement of proventricular cells during the invagination of foregut epithelium into endodermal midgut layers. Our study shows that the increased migration potential of ple (−/−) compared with ple (+/+) keratinocytes is directly linked to elevated states of Erk kinase phosphorylation. Erk has been implicated in the migration of numerous cell types (), and it has been shown to be the sole kinase responsible for ECM-initiated migration of keratinocytes (). On the other hand, enhancement and directionality of growth factor signaling is mediated by both Erk and p38 kinases, whereas JNKs were reported to be uninvolved in keratinocyte motility (). By pharmacological inhibition of Erk's upstream kinases MEK1/2, we were able to restore the aberrant high migratory potential of ple (−/−) keratinocytes to normal levels, but were unable to rescue the abnormal keratin network organization of these cells. Therefore, we feel it is safe to conclude that hyperactivation of Erk1/2 is a result of keratin network alterations caused by plectin deficiency, rather than the opposite. In support of this, keratinocytes from EBS patients, with mutations in keratins leading to spontaneous formation of keratin aggregates, migrate significantly faster in comparison to control cells (). Although Erk activities were not investigated in this study, elevated basal levels of the stress-activated kinase SAPK/JNK found in these cells () were implicated in their faster migration. This raises the intriguing question of whether distinct alterations in keratin network organization, such as aggregation in keratin-related EBS versus bundling in EBS caused by plectin deficiency, may lead to the up-regulation of distinct signaling pathways, such as SAPK/JNK versus MAPK. Studying human keratinocytes from an EBS-MD patient, reported unaltered migration using phagokinetic track measurements. As these cells very likely expressed rodless isoforms of plectin, contrary to the plectin-null cells used in our study, it is difficult to compare both studies. The signaling pathway leading from plectin-related keratin network alterations to hyperactivation of Erk1/2 still remains elusive. Our analysis conducted so far shows that the activities of two key proteins known to be involved in the regulation of keratinocyte migration, c-Src, and PKCδ are up-regulated in the membrane fraction of plectin-deficient cells, and that PP2-inhibition of c-Src indeed down-regulates Erk1/2 activities. Thus, membrane-associated c-Src and PKCδ are likely candidates for mediators of signals from plectin to Erk1/2. For the transduction of signals from the IF network to the membrane, our live-cell imaging data of transfected migrating keratinocytes, expressing EGFP-RACK1 fusion proteins, offer a plausible mechanistic explanation. Similar to a model proposed for PKCδ regulation through plectin-sequestration of RACK1 on IFs of fibroblasts (), we propose that in keratinocytes regulatory (trigger) proteins of PKC, c-Src, and/or other upstream affectors of Erk1/2, are sequestered on IF-associated plectin molecules in a wild-type scenario, but are unbound to IFs and have free access to the membrane because of their missing anchor in plectin-deficient cells. The faster migration of cells overexpressing RACK1, which is shown in this study, is consistent with such a model. It will be of interest to characterize in more detail on the molecular level how c-Src and PKCδ become activated through signaling proteins such as RACK1 (and possibly Fer kinase) and what consequences this might have in different cell types. shows a scheme depicting the model proposed. Primary keratinocytes were isolated from 1-d-old ple (+/+) and (−/−) mice according to the protocol described by . Cells were cultured on laminin 5–enriched matrices, which were prepared from 804G cell cultures (see below), and used for experiments without further passage of cells. Immortalized p53 (−/−) basal keratinocytes were derived from ple (+/+)/p53 (−/−) and ple (−/−)/p53 (−/−) mice () cultured in KGM (Cambrex) on collagen I –coated (Sigma-Aldrich) plastic dishes. These cells were used at passage numbers 10–15, expressing keratins exclusively (Fig. S1). All initial key studies (IF phenotype and OA exposure experiments) were performed using primary keratinocytes and immortalized cell lines in parallel, with identical results. Because of the inability of primary keratinocytes to grow on collagen (), all migration-related experiments ( and ) were performed with immortalized cells. To obtain subconfluent cultures (used in all experiments except where indicated otherwise), cells were seeded at 8 × 10 cells/ml. Cells were exposed to 0.1 μg/ml OA (Sigma-Aldrich) for different time periods (for 1, 2, 4, and 6 h). For transient transfections we used Fugene reagent (Roche) according to the manufacturer's instructions. Expression plasmids used encoded, tagged versions of full-length plectin isoform 1a (pGR245 and pVP37) or RACK1 (pSOS33). pGR245 and pVP37 were generated by subcloning plectin 1a cDNA into pEGFP-N2 (CLONTECH Laboratories, Inc.) or a modified pEGFP-N2 where EGFP was replaced by an Myc tag, respectively. pSOS33 was generated by subcloning RACK1 cDNA (provided by D. Mochly-Rosen [Stanford University, Stanford, CA] as plasmid pDM31) into pEGFP-C1 (CLONTECH Laboratories, Inc.). For preparation of laminin 5 matrices, rat bladder carcinoma 804G cells were cultured to confluence on plastic dishes in DME, washed once with PBS, and lysed with 20 mM NHOH for 5 min. Dishes were then washed thoroughly with water and kept in PBS (plus 10% DMSO) at −20°C. Osmotic shock stress assays were performed following the protocol of . For immunoblotting, the following primary and secondary antibodies were used: anti-K5 and -K6 antisera (PRB-160P and PRB-169P, respectively; Covance), mAbs LL001 to K14 (provided by J.M. Leigh, Royal London School of Medicine and Dentistry, London, England; ), mAbs LP34 (DakoCytomation) to K5, K6, and K18 (pan-keratin), a mixture of mAbs to K18 and K8 (Ks 18.04 and Ks 8.7, respectively; Progen), anti-plectin antiserum #9 (), anti-INTβ4 antiserum (provided by F.G. Giancotti, Memorial-Sloane Kettering Cancer Center, New York, NY; ), affinity-purified goat anti-vimentin antiserum (provided by P. Traub, University of Bonn, Bonn, Germany), mouse mAbs sc-535 to p38 (Santa Cruz Biotechnology, Inc.), rabbit mAbs 3D7 to phospho-Thr/Tyr p38 (Cell Signaling Technology), mAb D-2 to Erk2 (Santa Cruz Biotechnology, Inc.), mAbs E-4 to phospho-Tyr Erk1/2 (Santa Cruz Biotechnology, Inc.), anti–c-Src antiserum (Santa Cruz Biotechnology, Inc.), anti-phospho Y418 Src antiserum (Biozol), anti-phospho Thr PKCδ antiserum (Cell Signaling Technology), mAb P36520 to PKCδ (BD Biosciences), anti-caveolin antiserum (BD Biosciences), goat anti–rabbit IgG, goat anti–mouse IgG, and donkey anti–goat IgG (all from Jackson ImmunoResearch Laboratories), all conjugated to horseradish peroxidase. For immunofluorescence microscopy the following primary antibodies were used: anti-plectin antiserum #46 (), pan-keratin (see above), anti-K5 antiserum (see above), and rat mAbs to INTα6 (CD49f; BD Biosciences), mAbs B-5-1-2 to α-tubulin (Sigma-Aldrich), and affinity-purified antiserum and mAbs to actin (A 2066 and AC-40, respectively; Sigma-Aldrich). As secondary antibodies, we used goat anti–rabbit IgG Alexa Fluor 488 (Invitrogen), goat anti–rat IgG Texas red (Accurate Chemical & Scientific Corporation), goat anti–mouse IgG Texas red, donkey anti–mouse Rhodamine red-X, and donkey anti–goat Cy2 (all from Jackson ImmunoResearch Laboratories). Cells grown overnight (∼12 h) were methanol-fixed, washed with PBS, mounted in Mowiol, and viewed in a laser-scanning microscope (LSM 510; Carl Zeiss MicroImaging, Inc.) at room temperature. Images were visualized with either a Plan-Apochromat 63×, 1.4 NA, or a Plan-Apochromat 100×, 1.4 NA, objective lens (Carl Zeiss MicroImaging, Inc.) using LSM software and processed using the Photoshop CS2 (Adobe) software package. For electron microscopy, cells grown on glass coverslips were washed three times with 0.15 M Sorensen's buffer (SB), pH 7.4, before a 1-h fixation in 3% glutaraldehyde in SB. Cells were then washed twice with SB and postfixed in 1% OsO4 in SB for 30 min. Subsequently, they were dehydrated in ethanol and flat-embedded in epoxy resin (Agar 100). Glass coverslips were removed from the Epon block by immersion in liquid nitrogen and subsequent warming. Thin sections (60–80 nm) were cut parallel to the plane of the cell layer, using an ultramicrotome (Leica). They were then mounted on copper grids, contrasted by uranyl acetate and lead citrate, and viewed at 60 kV in an electron microscope (JEM-1210; JEOL). After ∼12 h of adhesion, keratinocytes were washed twice with PBS and lysed directly with 50 mM Tris-HCl, pH 6.8, 100 mM DTT, 2% SDS, 1 mM NaVO, 1× phosphatase inhibitor cocktail 1 (Sigma-Aldrich), 1% bromophenol blue, and 10% glycerol (sample buffer). Aliquots of cell lysates containing equal amounts of total proteins were separated by SDS-PAGE and, after immunoblotting using peroxidase-coupled secondary antibodies, protein bands were visualized by exposure to x-ray film. Quantitation of bands was performed as previously described (). Triton X-100 or high-salt extract fractions of keratinocytes were prepared according to . Membrane and cytosolic fractions were prepared according to a digitonin-based extraction protocol (). ple (+/+)/p53 (−/−) and ple (−/−)/p53 (−/−) basal mouse keratinocytes were grown in parallel until reaching confluence (∼48 h). They were then treated with 10 μg/ml mitomycin C for 2 h. Subsequently, a scratch wound was introduced into the monolayer using a yellow Gilson pipette tip. Cells were washed three times with growth medium and further incubated for 24 h. For MEK1/2 or c-Src inhibition, 30 μM PD98059 (Cell Signaling Technology) or PP2 (Calbiochem), as indicated, were added to the growth medium 1 h before scratching and throughout the scratch closure period. Before fixation, a reference wound was inflicted to determine the original wound size. Cells were then fixed with methanol and processed for immunofluorescence microscopy using mAbs to actin, anti-plectin

The expression of heparan sulfate (HS) in the developing brain is crucial for the correct formation of several brain structures. This was recently demonstrated with a knockout of the enzyme responsible for HS polymerization in proteoglycans (PGs; ). The described phenotype indicates a general organizing role for HS and HS-binding growth factors in the brain. Because of the dramatic phenotype, it is difficult to tell how the individual HSPGs contribute to the brain morphology and what mechanisms are responsible for the phenotypic changes. The syndecan family forms a rather diverse group of transmembrane HSPGs with a wide array of ligands and distinct cell signaling capabilities (; ). Although in some cases the family members have overlapping functions, syndecans and their HSs can bind distinct ligands and produce cellular responses unique to the particular syndecan in question (; ). In addition, the in vivo expression pattern of each syndecan can differ greatly from the others. Thus, individual HSPGs are suspected to have unique functions in brain development. N-syndecan and heparin-binding (HB) growth-associated molecule (GAM; pleiotrophin) act as a receptor–ligand pair in neurite outgrowth (, ; for reviews see ; ). ECM-associated HB-GAM specifically binds the HS glycosaminoglycans present in the perinatal brain N-syndecan. Ligation of N-syndecan by HB-GAM triggers a signaling cascade involving the phosphorylation of c-Src and cortactin, thus affecting the actin assembly in the growth cones of neurites (for review see ; ). To date, their cofunction in neural migration has not been reported, although HB-GAM is known to mediate haptotactic migration of perinatal rodent forebrain cells via the receptor-type tyrosine phosphatase β/ζ, which is another transmembrane receptor of HB-GAM (). In addition, HB-GAM promotes osteoblast migration, which may be at least partially caused by its binding to N-syndecan (). In this study, we present findings suggesting an important role for N-syndecan in radial neural migration and in the rostral migratory stream (RMS) of the brain. Haptotactic migration caused by the binding of N-syndecan to its ligand HB-GAM offers a mechanistic explanation for the role of N-syndecan in neural migration. Furthermore, N-syndecan cooperates with the EGF receptor (EGFR) to regulate neural migration. In the initial characterization of the N-syndecan knockout mice (), no changes were found in the gross anatomy of the brain. However, stereological analysis revealed an increased cell density in deep cortical areas and a decreased cell density in superficial layers of the cortex (). Defective radial migration of neurons in the N-syndecan knockout brain is one possible mechanism to explain the phenotypic finding. This explanation appears plausible because N-syndecan is expressed in the neurites of embryonic brain neurons (), and the major ligand of N-syndecan that enhances cell migration, HB-GAM, forms radially oriented streaks during and after the midgestation period in the rodent cortex (for review see ). Thus, we examined the migration of newly born neural cells in vivo in the N-syndecan knockout cortex by labeling the cells with BrdU at embryonic day (E) 15 and tracing the BrdU cells from cortical sections at E18, postnatal day (P) 1, and P10. The population that is labeled by BrdU in the ventricular zone (VZ) and sub-VZ (SVZ) of the dorsal cerebral cortex at E15 contains neurons that will end up in the most superficial laminae of the adult cerebral cortex. We determined the relative scattering of the BrdU cells in the cortical layers to reveal the possible differences in cell accumulation to the cortical plate (CP). At E18 in the N-syndecan knockout pups, the proportion of BrdU cells in the VZ/SVZ was higher than in the wild-type samples. Correspondingly, the amount of BrdU cells was lower in the CP in the knockouts (). The relative shift of BrdU cells from VZ/SVZ to the CP was clearly smaller in the knockouts than in the wild types (). We continued to follow the fate of the labeled cell population, and at P1 in the N-syndecan knockout brains, the relative distribution of BrdU cells in layers in the VZ/SVZ and CP was still clearly different from the wild type but had normalized somewhat (). At P10 both in wild types and knockouts, all BrdU-labeled cells had shifted from below the subplate to the CP. The mature layer structure of cortex is already quite recognizable at P10, and we estimated the number of BrdU cells in the superficial layers II/III and in the deep layer VI. N-syndecan knockout pups displayed a similar phenotype as the adult knockouts, namely that layers II/III had less BrdU cells than the wild-type layers II/III and that layer VI in knockouts was denser with BrdU cells than in the wild types (). N-syndecan has been proposed to act as a coreceptor in FGF-2 binding to high affinity FGF receptor (). Neural stem cells/progenitor cells are highly responsive to FGF-2, and N-syndecan deficiency could thus lead to decreased cell proliferation and neuron numbers in the cortex. We compared the cortical neural proliferation and differentiation between N-syndecan knockout and wild-type mice and found no differences. For the results and methods used, see supplemental material (available at ). Radial neuronal migration can be recapitulated in brain slices and is stimulated by EGF (). We followed this protocol and labeled the cells with BrdU in E15 living brain slices. In this assay, the scattering of BrdU cells in the different cortical layers was examined after 24 h of culture. We could detect a relative increase in the number of BrdU cells in the CP of the wild-type slices and a decrease in the number of cells in the VZ/SVZ, when EGF was present in the culture medium. The relative shift of cells from the VZ/SVZ to the CP was ∼10% higher in the EGF-treated slices than in nontreated slices. Interestingly, in the N-syndecan knockout slices, we could not see an EGF-induced shift of BrdU cells to the CP (). Results of this semi–in vivo experiment support the view that the cortical structure changes in the N-syndecan knockout mice depend on radial migration defect and, furthermore, suggest that the role of N-syndecan in neuronal migration may be coupled to EGFR activation. A major neuronal migration route, the RMS, feeds the olfactory bulb (OB) with interneurons throughout life. A previous study has localized both N-syndecan mRNA and protein to migrating neuron clusters of the olfactory region and to the embryonic and adult OB (), but expression in the RMS has not been studied. In addition, no functional evidence has been presented for a role of N-syndecan in the migrating neurons. Therefore, we used P3 living slices of RMS, in which N-syndecan is detected by immunohistochemistry in the migratory stream of cells (). For functional studies, cells in the rostral part of the RMS were labeled with the lipophilic dye DiI (). This method effectively labels only cells migrating along the RMS and not the cells that originate from the VZ of the OB itself (). We cultured the injected slices for 24 h. During this time period, N-syndecan wild-type slices accumulated twice as many DiI-labeled cells in the volume of the OB as the N-syndecan knockout slices (). The volume of the N-syndecan knockout OB was not different from the wild type (unpublished data). HB-GAM occurs in developing tissues as an ECM-associated ligand and is spatiotemporally coexpressed with N-syndecan (; ). Thus, N-syndecan may be a receptor of matrix-associated HB-GAM in migration. We tested this hypothesis in a transfilter migration assay using E15 forebrain cells, the most relevant cell type from the viewpoint of the knockout phenotype (see first section of Results). N-syndecan–deficient forebrain cells migrated very poorly to HB-GAM in this assay (). However, they did display elevated migration at higher HB-GAM concentrations, suggesting the presence of another migration-associated receptor for HB-GAM. The other receptor could very well be the receptor-type protein tyrosine phosphatase β/ζ () because we could not completely abolish this migration in the knockout cells by degrading HS chains with heparitinase (unpublished data). To rule out possible deficiency in the general migration ability of the knockout cells, we used EGFR activation to induce migratory phenotype in the cells. EGFR functions as a scatter factor in neural cell migration, and EGFR activation induces the migratory behavior of all neural cell types and, in addition, mediates chemotactic migration in the brain (; ; ). EGFR is also known to activate the c-Src–cortactin pathway in a ligand-specific manner, which links EGFR signaling to the modulation of the actin cytoskeleton (). First, we wanted to verify that EGFR activation alone is enough to potentiate migration in vitro. In the haptotactic cell migration assay, exogenous EGF enhanced the migration of wild-type mouse E15 forebrain cells to HB-GAM () and to other haptotactic guidance molecules like midkine (; and unpublished data). The increase in the number of migrating cells was not the result of EGF chemotaxis because the exogenous ligand was homogeneously available to the cells in the assay medium. Unexpectedly, N-syndecan knockout cells completely failed to respond to EGF stimulus and did not show any enhancement in cell migration (), suggesting that N-syndecan may interact with EGFR or some component required for EGFR-dependent migration. The receptor-mediated induction in the wild-type cells was blocked with AG1478, a specific blocker for EGFR. The effect was also highly dependent on Src activity, which was revealed by the Src-blocking agent PP2, whereas blocking the MAPK pathway with U0126 did not affect the EGF-induced migration enhancement (). HB-GAM–induced activation of c-Src through N-syndecan has been demonstrated with N18 cells transfected with N-syndecan and adhered on HB-GAM matrix (). We followed the same experimental procedure with E15 forebrain primary neurons. The cells were plated on HB-GAM–coated plates and lysed after defined time intervals. An increase in c-Src phosphorylation was clear in the wild-type cells within the first 60 min (). c-Src phosphorylation was not induced in the N-syndecan knockout forebrain cells in a similar manner (). The level of phosphorylation in the knockout cells gradually increased toward the 60-min time point but never reached the levels observed in the wild-type cells. HB-GAM has other signaling receptors but seems to induce c-Src phosphorylation specifically via N-syndecan. The slight, gradual increase in phosphorylation in knockout samples could be explained by some other weak receptor activity. HB EGF-like growth factor (HB-EGF) is suggested to specifically induce the migration of cortical neurons in a chemotactic manner (), and its binding to EGFR is heparin dependent (). We prepared living brain slice cultures from E18 N-syndecan knockout and wild-type brains to estimate the chemotactic response of N-syndecan–deficient cells in semi–in vivo conditions. 1 d before slice preparation (E17), the mothers received BrdU injections, which should label neuronal cells from the last dividing stem cell population. These cells then migrate to the superficial cortical layers (). A 1-d gap between the injections and slice preparations prevents false positive results, as there is no free BrdU left to label possibly damaged or late-proliferating cells in the brain. We placed HB-EGF–soaked agarose beads unilaterally in layer I of the cortical slices. The other cortical hemisphere in each slice was used as an internal control with a BSA-soaked bead. Using the density of BrdU-positive cells accumulating around the bead in 24 h, we estimated the relative chemotactic activity of HB-EGF in normal and N-syndecan–deficient brain. In the wild-type slices, HB-EGF induced a clear accumulation of BrdU-labeled cells around the beads (). The density around the BSA beads stayed the same as the density elsewhere in the same layer. In the N-syndecan knockout slices, the accumulation of cells was much weaker around the HB-EGF beads (only 30% of the wild-type accumulation; ). These results suggest a physiologically significant role for N-syndecan in HB-EGF chemotaxis. We modified our migration chamber assay to measure more accurately the strength of chemotaxis in N-syndecan–deficient neurons. An insulating layer of matrigel was added to separate the compartment with the chemoguidance protein from the compartment containing the neural cells. Within 24 h, HB-EGF induced strong migration through the insulating layer in normal embryonic forebrain cells but not in N-syndecan–deficient cells (). If the binding of HB-EGF to HS chains is required for EGFR stimulation in this assay, HB-GAM, which binds HS chains at the neuron surface () but is not chemotactic itself, might interfere in HB-EGF–induced chemotactic migration. In fact, HB-GAM in the assay medium clearly inhibited HB-EGF–induced chemotaxis, with a half-maximal inhibition at about threefold molar excess compared with HB-EGF (). EGFR is normally expressed in the N-syndecan knockout forebrain at E15 (unpublished data). However, the low amount of EGFR prevents the detection of differences in receptor activation between knockout and wild-type brains. Therefore, we used the A431 cell line with abundant EGFR expression, induced EGFR phosphorylation in the cells with HB-EGF, and examined the effect of soluble HB-GAM on EGFR activation. HB-GAM inhibited HB-EGF–induced EGFR phosphorylation at a 10-fold molar concentration (). If N-syndecan were able to sensitize EGFR signaling in the forebrain cells by direct interaction, it should be found in the same cellular compartments. We double immunostained forebrain cells plated on HB-GAM with N-syndecan and EGFR antibodies. Cells grown on HB-GAM showed limited colocalization of N-syndecan and EGFR initially, but N-syndecan was found more in the base of the growing neurite and filopodia, whereas EGFR was more diffusely distributed near the cell soma (unpublished data). Cells treated with EGF started spreading rapidly, and EGFR relocalized closer to the edges of the growing cells and away from the cell soma (). N-syndecan and EGFR clearly colocalized in the growing edges and neurite base of the EGF-treated cells. At least some of the EGFR and c-Src signaling occurs in detergent-resistant membrane domains (), which are also known as lipid rafts (). To determine whether N-syndecan and its signaling partners are targeted into the same membrane fractions, we used flotation sucrose gradients to isolate lipid rafts from prenatal rat forebrains and detected the proteins of interest with specific antibodies. In agreement with previous studies (for review see ; ), EGFR was present in the lipid raft fraction. In addition, a limited amount of the EGFR ligand HB-EGF was detected in the same fraction, although the majority of differentially processed HB-EGF forms was distributed in soluble fractions and pelleted material (). N-syndecan and its ligand HB-GAM were enriched in the lipid raft fraction as well as Fyn kinase, which was also used as a raft fraction marker. We also examined the localization of cortactin, which is specifically phosphorylated by Src family kinases to regulate cell motility (). Furthermore, cortactin is involved in both the EGFR and N-syndecan signaling pathways. To some extent, cortactin was localized in the lipid raft fraction, although the majority was diffusely present in soluble fractions. More significantly, cortactin phosphorylated at tyrosine residue 421 was almost exclusively present in the lipid rafts. Thus, all of the key molecules appear to be localized in the lipid raft membrane regions, greatly increasing the likelihood of their association. Fluorescence resonance energy transfer (FRET) analysis reveals the physical proximity of two different fluorochromes (; ). This method is successfully used to demonstrate physical coupling and oligomerization of receptors and receptor complexes by tagging the receptors with selected fluorochromes, such as CFP (donor) and YFP (acceptor), and quantifying donor and acceptor fluorescence changes upon their close encounter. We used this method to examine in more detail the possible coupling of N-syndecan and EGFR at the cell surface. A CFP–YFP fusion vector was used as a positive FRET control, whereas cotransfection of unmodified CFP and YFP vectors was used as a negative control (). When HEK293T cells transfected with the fluorescent fusion receptors were plated on HB-GAM–coated coverglass, FRET quantification clearly showed the clustering of N-syndecan and EGFR at the sites of matrix (HB-GAM) contact (). To examine whether HB-EGF was capable of inducing further receptor association, we added HB-EGF to the culture medium and measured FRET after 20 min. There was a visible increase in the FRET signal, indicating that HB-EGF binding induces the close coupling of N-syndecan and EGFR (). Interestingly, the most intense FRET signal is seen at the base of the processes of 293T cells; this area may be analogous to the base of the neurite, where the most intense colocalization of N-syndecan and EGFR is observed in neurons (). That clustering should occur to some extent even in the absence of EGFR ligand, which speaks for the receptor coupling via the common signaling complexes of N-syndecan and EGFR (see Discussion). When the cells were plated on laminin, a weaker FRET signal was visible for a short period of time in the presence of HB-EGF (). This indicates that laminin does not induce such a strong clustering of N-syndecan and EGFR and that the HB-EGF–induced complexes are not very stable on laminin. HEK293 cells may handle their membrane traffic in a manner not comparable with neurons, and we repeated this FRET experiment in hippocampal primary neurons. With electroporation, we could transfect the neurons with high enough efficiency to measure FRET, although the expression level of either N-syndecan or EGFR construct was not very high and declined rapidly in culture. Nevertheless, we could essentially repeat our results obtained with HEK293 cells, as we saw the clustering of N-syndecan and EGFR after HB-EGF stimulus (). Interestingly, the clustering took place in almost exactly the same locations where we observed the colocalization of these receptors with immunostaining, namely in the base of the growing neurite and in the neurite itself. Migration assays and FRET analysis suggest that interactions of N-syndecan and EGFR depend, at least in part, on a shared ligand such as HB-EGF. We have previously used N-syndecan ectodomain to determine the detailed interactions between the N-syndecan and HB-GAM (). In this study, we have used the same recombinant PG fragment to elucidate the binding of HB-EGF to the N-syndecan ectodomain in an ELISA-type assay (). HB-EGF indeed binds the N-syndecan ectodomain with a similar affinity as HB-GAM. We estimated the dissociation constant to be 12 nM for HB-GAM and 15 nM for HB-EGF. The binding was found to be competitive (), indicating an interaction with the same or overlapping HS sequences. These findings provide the apparent explanation for why N-syndecan knockout cells fail to migrate in HB-EGF–induced chemotaxis assays and why soluble HB-GAM inhibits HB-EGF–induced chemotaxis and EGFR phosphorylation. #text N-syndecan knockout mouse production and genotyping has been described previously (). Methods used in animal studies were approved by the board of laboratory animal experiments at the University of Helsinki. The mice were back bred to the C57/BL6 mouse strain (>10 generations) to obtain an inbred line of mutant animals. Timed pregnant females received an i.p. injection of 50 μg/g BrdU at E15. For in vivo migration assay, brain samples were collected at E18, P1, or P10, fixed with 4% PFA, and embedded in paraffin. For slice culture migration assay, the females were killed 30 min after the injection, and the embryos were collected. The brains of the embryos were prepared as slice cultures. The slices were cultured for 24 h with or without 20 ng/ml EGF. The slices were fixed and embedded in paraffin. 5-μm–thick sections were cut from the slice culture and E18 paraffin blocks, whereas 20-μm sections were cut from the P1 and P10 blocks. The sections were stained with anti-BrdU antibody. The density of BrdU cells in the SVZ/VZ and CP was estimated in relation to the other laminae from photographs or with Stereo Investigator (MicroBrightField). Haptotactic migration assays were performed on Transwell plates with a 12-μm pore size (Costar). The chamber-side membrane was coated with HB-GAM, laminin, or 0.01% poly--lysine. 150,000 primary neurons were plated per well in the assay medium. In the assay for EGFR-promoted migration, EGF was added to the medium (2, 20, or 50 ng/ml; Promega). The medium was freely diffusing across the membrane. Mouse P3 brains were collected and prepared as slice cultures. 100 ng/ml DiI in dimethylformamide (Invitrogen) was injected with a glass capillary using a stereomicroscope (SZX9; Olympus) to the area of the anterior RMS (). The injected slices were cultured for 24 h, rinsed with PBS, and cryopreserved with 15% sucrose in PBS. The slices were frozen and cut to 20-μm–thick sections with cryotome (Microm) and mounted with water-based mounting glue. Embryonic brain samples were collected from timed pregnant females; the day of detection of the vaginal plug was determined as E0. Females were killed with CO and by cervical dislocation, the uteri were dissected out, and the embryos were collected and dissected. The torsos were used for genotyping, and the heads were either further dissected for cell culture or fixed in 4% PFA/PBS overnight. Pups under P10 were killed by decapitation, and the heads were washed in PBS. DNA from the abdominal organs was used for genotyping the pups. The brains were dissected out of the skull on a petri dish and kept on ice. The whole brain was either fixed with 4% PFA in PBS overnight or used for cell culture experiments. Adult brains were collected by killing the mice with CO and by cervical dislocation. The brains were rinsed in cold PBS and fixed with 4% PFA in PBS overnight. Primary forebrain neurons were collected from E15 mouse embryos. The frontal lobes of the brain were dissociated and triturated with a 20-G needle and sterile syringe. Hippocampal cells were prepared from E18 rat brains in a similar manner. Primary cells were allowed to recover in a cell incubator with medium containing 10% FCS for 30 min before use in the assays. Assay medium was serum free with 10 mg/ml BSA. After recovery, the cells were centrifuged and suspended to the assay medium. The cell preparation was left to stand for 5 min to sediment the undissociated material. The cleared cell suspension containing 80–90% of neurons or neuronal precursors () was then collected and plated for the assays. DME (Invitrogen) with 10% FCS (Invitrogen) was used with added -glutamine and penicillin-streptomycin (Invitrogen) in primary cell preparations and in cell line cultures. Cells were cultured at 37°C in 5% CO. In migration assays and assays involving the application of recombinant proteins, FCS was omitted, and 10 mg/ml BSA (Sigma-Aldrich) was used instead in the assay medium. In the slice culture assays, 2% glucose was added to the assay medium. Primary forebrain neurons were collected from E15 mouse or E18 rat embryos. HEK293T and A431 cells were purchased from the American Type Culture Collection and cultured according to the supplier's instructions. Human recombinant EGF was purchased from Promega; human recombinant HB-EGF was obtained from R&D Systems; poly--lysine was purchased from Invitrogen; and laminin was purchased from Sigma-Aldrich. HB-GAM and N-syndecan ectodomain IgG fusion protein (ENS-IgG) were produced as described previously (). Affinity-purified chicken anti–N-syndecan was produced against N-terminal peptide () by AgriSera. Rabbit anti–HB-GAM production has been described previously (). Goat anti-EGFR, rabbit anti–c-Src, rabbit anti-Fyn, goat anti–HB-EGF, and rabbit anticortactin were purchased from Santa Cruz Biotechnology, Inc. Rabbit anti–β-III–tubulin and rabbit antineurofilament were purchased from Chemicon. Mouse anti-BrdU and detection reagents were purchased from GE Healthcare. Agarose-conjugated antiphosphotyrosine antibody was obtained from Cell Signaling Technologies, and unconjugated mouse antiphosphotyrosine was purchased from Sigma-Aldrich. The cell samples were lysed in 1% NP-40 PBS with protease and phosphatase inhibitors (1 μg/ml aprotinin, 1 mM PMSF, and 1 mM Na-o-vanadate). The cell lysates were homogenized by triturating with a 20-G needle. The homogenized samples were immunoprecipitated with either agarose-conjugated or soluble antibodies at 4°C overnight. Complexes formed with soluble antibodies were further collected with agarose-conjugated protein G (Sigma-Aldrich) for 2 h at RT. The precipitates were washed twice with the lysis buffer and suspended in SDS-containing loading dye. The samples were separated by denaturing PAGE, transferred to nitrocellulose filters, and blotted with antibodies of interest. HRP-conjugated secondary antibodies were detected with ECL (GE Healthcare). All quantifications of band intensities in films or cell number estimates from photographs were made using ImageJ software (National Institutes of Health; ). Thin (5 μm) paraffin sections were photographed at 63× magnification (NA 1.4) with a camera (AxioCam; Carl Zeiss MicroImaging, Inc.) mounted in a microscope (Axioplan; Carl Zeiss MicroImaging, Inc.). The relative density of immunostained or hematoxylin-eosin–stained cells in systematic random sample was estimated using the selector method, a variant of optical fractionator suitable for thin sections (). The number of immunostained or DiI-labeled cells in thick (20 μm) sections was estimated with Stereo Investigator (MicroBrightField) and a microscope (BX51; Olympus) using an optical fractionator (). Petri dishes were coated with 10 μg/ml HB-GAM overnight at 4°C. The plates were washed with sterile PBS and equilibrated with the assay medium for 30 min before use. E15 forebrain neurons were prepared and plated on the HB-GAM–coated plates for 20, 40, and 60 min, after which the medium was carefully removed from the plates and the cells were washed and scraped off in the immunoprecipitation lysis buffer. The samples were immunoprecipitated with agarose-conjugated antiphosphotyrosine antibodies and blotted with anti–c-Src antibodies. E16 forebrain cells were collected from N-syndecan knockout and wild-type embryos. The cells were plated on 5-cm plates coated with poly--lysine (2.5 million cells per plate). The cells were allowed to attach to the plates for 2–3 h. 20 ng/ml EGF or an equal volume of PBS was added on the cells for 15 min, the plates were carefully washed with PBS, and the cells were lysed in the immunoprecipitation lysis buffer. The samples were immunoprecipitated with anti-EGFR antibody and blotted with antiphosphotyrosine antibody and anti-EGFR. Rat forebrains were dissected from E17 embryos and homogenized on ice by passing through a 20-G needle in a small amount of 1% Triton X-100, TNE buffer (20 mM Tris, pH 7.4, 150 mM NaCl, and 5 mM EDTA) with complete protease inhibitors (Roche), and 1 mM NaVO. After 15 min on ice, the lysate volume was adjusted to 0.5 ml, mixed with 0.5 ml of 80% sucrose, and overlayed with discontinuous sucrose gradient (2 ml of 36% sucrose in TNE and 1.5 ml of 5% sucrose in TNE) in a 5-ml polyclear centrifuge tube (Beckman Coulter). The samples were centrifuged for 16–20 h at 100,000 in a rotor (SW55; Beckman Coulter) at 4°C. Fractions of 0.5 ml were collected from the top, and preliminary identification of the lipid raft fraction was performed based on the density and presence of white scattering material. Fraction aliquots containing equal protein amounts were precipitated with 13% TCA, resuspended in SDS-PAGE sample buffer, and resolved by 4–15% PAGE. Protein content was analyzed by Western blotting using antibodies against syndecan-3 (rabbit IgG against the cytoplasmic domain), EGFR, Fyn, HB-EGF, and HB-GAM. For production and transfection of control and assay cDNA vectors, see supplemental material. HEK293T cells or hippocampal primary neurons were grown in 35-mm glass-bottom culture dishes (MatTek) in DME with 10% FCS. Petri dishes were coated by 20 μg/ml poly--lysine, 50 μg/ml HB-GAM, 10 μg/ml laminin, or 100 μg/ml BSA. After 24 h, cells were washed with serum-free medium and incubated overnight without serum. During image acquisition, the temperature was decreased from 37 to 22–24°C to reduce the movements of cellular organelles. Cells were incubated with or without 20 ng/ml HB-EGF for up to 30 min. FRET signal was quantified with three filter sets (three-cube method): CFP channel (CFP; excitation of D436/20x, emission of D480/40m, and dichromatic long-pass filter [DCLP] 455); FRET channel (FRET; excitation of D436/20x, emission of D535/30m, and 455 DCLP); and YFP channel (YFP; excitation of HQ500/20x, emission of HQ535/30m, and 515 DCLP; Chroma Technology Corp.). The image was recorded in live transfected cells in a platform (Cell Imaging Systems) consisting of a 60× planApo oil immersion objective (Olympus), CCD camera (C9260-905; Hamamatsu), inverted microscope (IX71; Olympus), and illumination system (MT20; Olympus). The excitation light source was a 150-W Xenon lamp (3.42% of intensity). The images were acquired in binning 2 × 2 modes to increase the signal to noise ratio and 300–500-ms integration times. The background subtraction was made before the FRET calculations. The signals measured in the FRET channel were corrected for cross talk between CFP and YFP channels using a linear spectral unmixing algorithm (; ). Corrected FRET (Fc) was calculated on a pixel by pixel basis with the following formula:where a and b are correction factors for donor and aceptor. For HEK293 cells, factors a and b were found to be 0.39 ± 0.009 ( = 30) and 0.022 ± 0.002 ( = 30), respectively, and for primary neurons were found to be 0.3975 ± 0.0537 ( = 12) and 0.0281 ± 0.0053 ( = 8). EGFR-CFP and N-syndecan–YFP were expressed separately in the cells, and, for each fluorophore, the emission from the FRET channel was divided by the emission measured with either the CFP or YFP channels as follows: a = FRET/CFP, and b = FRET/YFP. FRET-corrected images are displayed in pseudocolor mode (red areas represent the high values of FRET, and blue areas represent the low values of FRET). The N-syndecan ectodomain was produced as a human IgG fusion protein as previously described (). The concentration of the fusion protein was determined with a dot blot using human IgG protein as a standard and HRP-conjugated anti–human IgG antibody for detection. 1 μg/ml HB-GAM or HB-EGF was diluted in TBS and bound to nonimmune 96-well plates for 1 h (100 μl per well). The wells were blocked with 2% BSA in TBS, and a dilution series of N-syndecan ectodomain was pipetted to the wells starting from 10 pmol/well. After 1 h, the excess of ectodomain was washed away, and the amount of bound ectodomain was estimated using human IgG as a control protein and HRP-conjugated anti–human IgG for detection. The K values for both HB-GAM and HB-EGF were estimated from the binding curves using Scatchard analysis. HB-EGF was diluted to 10 μg/ml in PBS with 0.1% BSA. 100 AffiGel agarose beads (Bio-Rad Laboratories) were collected and washed with micropipette in PBS. The beads were incubated with diluted HB-EGF or only dilution buffer for 1 h in 37°C. The beads were washed with PBS and placed in a drop of PBS on a glass plate. E17 timed pregnant N-syndecan heterozygote females received one 50-μg/g BrdU injection, and, 1 d later, cortical brain slice cultures were prepared from embryos. Embryos were genotyped afterward. Under a stereomicroscope (SZX9; Olympus), the HB-EGF beads were placed to the marginal zone of the cortex unilaterally, and a dilution buffer bead was inserted to the other hemisphere. After 24 h in culture, the slices were processed for BrdU immunostaining. The area for cell density estimation was defined to be within the longest radius still containing higher cell density than the layer mean. Online supplemental material provides analysis methods for neural cell proliferation. Table S1 provides data on cortical layer differentiation. Table S2 presents data on neural cell differentiation in vitro, and Table S3 provides the corresponding results received from the N-syndecan knockout mouse strain. Online supplemental material is available at .

Precursor cells that generate the various differentiated cell types of the central nervous system (CNS) are generally located in defined areas of the developing embryo, which frequently correspond to the ventricular zones of the developing nervous system. During a large part of the development of the embryo, neural progenitors proliferate to renew the precursor pool and, in parallel, give rise to postmitotic cells that move out of the progenitor areas (). The balance between the generation of precursor and postmitotic cells shifts during development but has to be tightly controlled to guarantee the formation of the appropriate neuronal cell numbers and the size of the tissue domains. Nervous tissue in the vertebrate embryo arises through complex processes of neural induction of the ectoderm, the delineation of neurogenesis-competent domains within the neural plate, and differentiation of neural precursors within these regions (; ; ). Simultaneously, the expansion of each neuronal compartment takes place until its definitive size and form are reached. At this point in time, the neurons undergo terminal differentiation, thereby acquiring the characteristics necessary for participation in nervous signal propagation while simultaneously losing their capacity for further proliferation. As the mechanisms controlling the spatio-temporal timing of the sequence of proliferation and differentiation of neuronal-committed cells are still only partly understood, we performed a differential display of mRNAs designed to isolate genes that are differently expressed during neuronal differentiation as well as genes responsible for the proliferation of the stem cell compartment in vivo. We used a differential display screen to discover genes that are involved in control of neural stem cell compartment proliferation. To this end, zebrafish embryos were either left untreated or were pulse treated with 0.5 μM all-trans retinoic acid (RA) between 50 and 60% epiboly when neural induction and patterning are initiated. Subsequently, mRNA was isolated from wild-type and treated embryos at 90% epiboly, which, after conversion into cDNA, was used for a differential display using the transcripts from the untreated embryos as driver cDNA and the transcripts from the treated embryos as tester cDNA. The resulting differential fragments were tested using whole-mount in situ hybridization at different developmental stages. One fragment was singled out for detailed characterization based on its restricted spatio-temporal expression pattern during late gastrulation and early somitogenesis. Obtaining a full-length sequence of the down-regulated fragment revealed that the gene was most homologous to mammalian ASB11 (ankyrin repeat and suppressor of cytokine signaling [SOCS] box–containing protein 11), thus identifying this gene as (sequence is deposited at the National Center for Biotechnology Information under GenBank/EMBL/DDBJ accession no. ; , supplemental material, and Fig. S1, available at ). ASB family members contain a relatively divergent N-terminal domain followed by a varying number of ankyrin repeats and a C-terminal SOCS box acting as a part of a ubiquitin ligase complex (; ; ; ) and are expressed in members throughout the chordate phylum (; ; ). was not maternally expressed because the transcripts were not detected by in situ hybridization or Northern blotting in 2.5 h postfertilization (hpf) embryos (unpublished data). The expression of was by in situ hybridization first detected at 4 hpf and was ubiquitous throughout the blastoderm (), and such uniform expression was maintained until late gastrula, when transcripts were restricted to the polster (). At the tailbud, neuroectodermal expression extends as two stripes along the margins of the neural plate (), most likely overlapping with the posterior lateral neural plate expression of and the lateral neural plate expression of (Figs. 2 and 3 in ) and being complementary to proneural domains expressing (). At the level of the prospective mid/hindbrain boundary, two bilateral domains of expression are detected (). The expression is transient, as this pattern is maintained until 12 somites. From 12 somites to at least 3 d after fertilization, there was no expression detectable by in situ hybridization, although low levels of mRNA were still detected in 24 hpf embryos by Northern blotting (unpublished data). In RA-treated embryos, the expression of is absent in the anterior half and down-regulated in the posterior half of the embryo, which is consistent with its isolation as a down-regulated fragment from the differential display screen (). Thus, the expression of may indicate its possible function in the spatio-temporal timing of proliferation and differentiation in the developing nervous system. Direct in vivo evidence for a role of in zebrafish development was obtained from d-Asb11 knockdown by the injection of morpholinos (MOs) designed to inactivate (d-Asb11–MO; ). We generated an antibody against d-Asb11, tested it on transfected cells, and confirmed it by Western blotting of embryo extracts upon MO injection of zygotes, which resulted in embryos devoid of d-Asb11 protein (supplemental material and Fig. S2 C, available at ). We assessed the effects of two different ATG MOs on development, and both gave similar phenotypes (see Materials and methods section MOs). At 3 d after fertilization, morphants were typically smaller with a shortened trunk, sometimes accompanied by a downward-directed curly tail and a hyperpericardium (). Rescue experiments were performed in which d-Asb11 knockdown by MO that targets sequences upstream of ATG was rescued with the overexpression of myc tag () mRNA lacking these sequences. Based on the morphology, the overexpression of (MO+) rescued 50% of the phenotype (MO), as shown by a decrease in phenotype from 50 to 25% (). Next, d-Asb11 knockdown experiments were performed to determine changes in the expression of the proneural basic helix loop helix gene () that actively promotes neuronal differentiation, inhibits gliogenesis, and is also expressed in still-dividing neuronal precursors (; ). Strikingly, d-Asb11 knockdown led to the expansion of that was first observed at the three-somite stage, when in the morphants, was expressed at a higher level and more contiguously in an otherwise unaffected pattern (). In the neural tube of morphants, lateral and dorsal expression domains of were substantially expanded, as assessed by image quantification of the -positive surface in transversal sections of representative embryos (). Furthermore, strong ectopic expression appeared at the center of the neural tube in the proliferative zone (). These data show that knockdown of d-Asb11 results in a premature commitment to the neuronal cell lineage. Next, we investigated whether the inhibition of translation affects the generation of terminally differentiated neurons by analyzing the expression of , a marker for postmitotic neurons. Again at the three-somite stage, expression was increased in the morphants () as compared with embryos injected with inverted MO. A comparable, albeit more subtle, phenotype was obtained by labeling of a subpopulation of the primary neurons by . As in morphants, neurons were slightly more numerous with what appeared to be higher levels of transcript than in the controls (). The increase of HuC-positive cells in the neural tube as evaluated by antibody to HuC protein is maintained at 24, 48, and 72 hpf (). The proportion of HuC+ cells in the neural tube was at all times higher in morphants as compared with wild-type embryos and, at 72 hpf, occupied even most of the ventricular zone known to consist of the proliferating cells (). However, low numbers of HuC-negative cells were at 72 hpf still present in the ventricular zone, suggesting that the proliferating compartment was not entirely depleted by loss of function d-Asb11. To quantify the proportion of postmitotic neurons relative to presumably proliferating unlabeled precursors that reside in the ventricular zone of the CNS, a 3D reconstruction of serial sections from a 24 hpf wild-type embryo and a representative 24 hpf morphant was made (). The measurements show that HuC+ cells contribute to 28% of the total volume of the neural tube in the wild-type embryo and 36% in the morphant, confirming that the proportion of postmitotic neurons is increased in the morphant embryo (). These data suggest that the targeted knockdown of d-Asb11 results in premature postmitotic neuronal commitment of a subpopulation of neural precursors. At 24 hpf, the expanded compartment in the morphants could hold both terminally differentiated neurons as well as still-dividing + precursors. To verify whether the HuC+ cells were indeed nondividing terminally differentiated neurons, we performed double fluorescent whole-mount antibody labeling with HuC and a mitotic marker anti–phosphohistone-3 (PH3) antibody. Examination of nine morphants and seven wild-type embryos revealed that there was no colocalization of the HuC+ and PH3+ signals (see Whole-mount immunolabeling). The data show that the knockdown of d-Asb11 resulted in a relative and premature increase of terminally differentiated nondividing neurons already at 24 hpf (). It is very well possible that a fraction of the supernumerary + cells in the morphants was still proliferating and, therefore, was not expressing HuC. The comparison of numbers of mitotic PH3+ cells between morphants and controls at 24 hpf showed that in the morphants, there was a slight but significant (P < 0.03) increase in proliferation that was most likely caused by ectopic proliferating + neurons (). These data suggest that d-Asb11 functions to prevent neuronal commitment possibly by sustaining proliferation. The reduction of the definitive neuronal compartment may be the result of the depletion of neural progenitors through decline in proliferation. This is supported by a highly significant (P = 0.008) decrease in PH3+ mitotic cells at 48 hpf (). Concurrent to the decrease in proliferation, the apoptosis of prematurely committed precursors could also contribute to a reduction in compartment size. To study this possibility, we analyzed whether the increase of HuC+ cells in the morphants was accompanied by apoptosis in the neural tube. We found slightly enhanced apoptosis at 24 hpf morphants using acridine orange as an indicator of apoptosis (). Analysis of apoptotic cells was extended using TUNEL assay (). Transversal sections of the neural tube of wild-type embryos showed apoptotic Rohon-Beard neurons, whereas throughout the entire neural tube of the morphants, there was an increase in apoptosis (). This suggests that cell survival of the prematurely committed neuronal precursors may be compromised as a consequence of d-Asb11 knockdown. Because d-Asb11 appears to function in preventing neuronal commitment of the progenitors, we studied whether knockdown of d-Asb11 altered the expression of neural progenitor genes or of the SoxB1 high mobility group (HMG) box transcription factors. The expression of and through evolution correlates directly with ectodermal cells that are competent to acquire neural fate and subsequently may correlate with the commitment of cells to a neuronal fate (; ). The expression of and in the zebrafish developing CNS is found in the regions containing neural precursors as well as those containing neurons (). The situation is complex, as neither nor is uniformly expressed in all neural-competent cells of the neural plate. This is because there are other SoxB1 genes expressed in the developing CNS of zebrafish and because the kinetics of expression during the transition between the neural precursor and neuron has thus far not been elucidated. Strikingly, the knockdown of d-Asb11 affected the expression of both and , albeit differently. At 20 hpf, expression in the neural tube was reduced in 50% of the morphants (), suggesting that d-Asb11 may function to maintain specific levels of transcripts that may be required for precursor cell maintenance. If this is the case, d-Asb11 does so, most likely, in conjunction with other factors because the expression of was weaker but still present. This difference in the intensity of expression between morphants and controls is already clear in whole-mount in situ hybridizations (). At 20 hpf in the neural tube of controls, transcripts were cellularly condensed and appeared to be localized in specific compartments of the cell (). In contrast, in 40% of the morphants, transcripts were diffuse and appeared scattered evenly throughout the cell (). The data suggest that d-Asb11 may be required for subcellular condensation of transcripts. We speculate that such condensed transcripts may be representative of a specific localization, and, in this form, may be required for maintenance of the precursor cell fate. Such a mechanism of segregating and subcellularly localizing molecular determinants in precursor cells through, for instance, interactions with the cytoskeleton has extensively been studied in neuroblast development in (). These data show that the loss of d-Asb11 may mediate the reduction of a population of neural precursors through affecting neural competence factors of SoxB1 HMG group transcription factors. Confirmation for d-Asb11 function in vivo and the possibility of its involvement in the transition of neural precursor to neuron was obtained in experiments in which we investigated whether the misexpression of mRNA would interfere with primary neurogenesis by evaluating expression, a marker for primary neurons, as well as the expression of neuronal markers and (). Indeed, in 70% of the embryos, expression was affected (). Most prominent was a reduction or total absence of expression in the lateral sensory Rohon-Beard neurons. Occasionally, defects in the ventromedial motor neurons or the cranial ganglia were observed. Immunolabeling demonstrated that MT–d-Asb11 was present in the domains that did not contain labeling. Comparable results were obtained with and (). To verify whether aberrant patterning of the neural plate could underlie abnormalities in primary neurogenesis, we evaluated the expression of regional neural plate markers , , and upon the misexpression of Next to the wild-type expression pattern, we observed a reduction of marker gene expression in embryos (). Most interestingly, we observed embryos in which marker expression was indicative of a split or otherwise deformed neural plate (), suggesting that in regions misexpressing MT–d-Asb11, neural plate formation was disrupted. To investigate whether the loss of -positive neurons and/or abnormalities in , and expression were secondary to neural plate defects, we studied the effect of misexpression on the expression of germ cell nuclear factor (GCNF) as a neural plate marker (; ). The regional overexpression of MT–d-Asb11 as detected by anti-MT antibody labeling at the three-somite stage overlapped with the region of the neural plate where GCNF was absent (). This suggests that d-Asb11 negatively interfered with the establishment of the neural plate and that the absence of -, -, and -labeled neurons and aberrant , , and expression are the consequence of this event. To exclude the possibility that the misexpression of mediates nonspecific repression of gene expression, we evaluated several markers upon the misexpression of . Interestingly, the expression of the Notch target gene was ectopically expanded in embryos misexpressing (), whereas the expression of was unaffected (not depicted), ruling out the possibility that is a general transcriptional repressor. The aforementioned experiments suggest that d-Asb11 may, when misexpressed, function to block neuralization through forced maintenance of the precursor cell fate. In support of this notion, the phenotype of d-Asb11 morphants is characterized by altered expression of neural competence factors and . To establish whether d-Asb11 misexpression may mediate its effects in preventing neuronal differentiation through the induction of expression, we misexpressed mRNA. As a result, we sporadically observed scarce ectopic -positive (sox2+) cells (unpublished data). In an attempt to increase the inefficiency of mRNA misexpression, we microinjected DNA in embryos. Although the frequency of ectopic induction of + cells (∼5%) remained low, the ectopically induced cells were more numerous. The ectopic expression of coincided with MT–d-Asb11 protein expression (). These data show that when misexpressed, d-Asb11 is capable of inducing the neural precursor gene is expressed in neural precursors as well as in neurons. As d-Asb11 misexpression inhibited neurogenesis (), we propose that through its positive regulation of the gene, d-Asb11 may maintain neural-competent cells in their precursor state. We obtained functional insight into the molecular mechanism by which d-Asb11 acts from experiments in which we investigated the action of mutated versions of with a specific deletion of the SOCS box () or of the entire C-terminal part after the last ankyrin repeat () using as a read-out their effects on neurons or on neural plate patterning as revealed by . Importantly, both mutant molecules resulted in a marked decrease of an altered expression pattern of when injected into zebrafish embryos (). As we have already shown, the expression of at early somitogenesis was clearly affected by the misexpression of , as in ∼50% of the embryos, one half of the V-shaped expression domain at the mid/hindbrain boundary was shifted laterally or was diminished (). Likewise, when or mRNA was microinjected, the wild-type expression pattern of was maintained in a high percentage of embryos (). Thus, the SOCS box of d-Asb11 is required for its biological effects. As mentioned in the second paragraph of Results, SOCS boxes frequently act in ubiquitin ligase complexes, and because the SOCS box in d-Asb11 appears to play an important role during embryogenesis, we determined whether the SOCS box in d-Asb11 is also able to function as an ubiquitin ligase. Indeed, the overexpression of d-Asb11 either in zebrafish embryos or in HeLa cells led to the ubiquitination of a variety of cellular proteins in a SOCS box–dependent fashion (), demonstrating that d-Asb11 is capable of functionally participating in a ubiquitin ligase complex and again showing that d-Asb11 is a member of the ASB family (; ; ). Importantly, cells transfected with the control expression vector or with SOCS box–deficient d-Asb11 were incapable of ubiquitination (). Thus, d-Asb11 can act as an ubiquitin ligase in a SOCS box–dependent fashion. Ankyrin repeats are important mediators of protein–protein interaction, and, thus, we reasoned that if d-Asb11 has the capacity to act as an ubiquitin ligase, it is most likely acting on ankyrin repeat–containing proteins. Indeed, we observed that upon coexpression in HeLa cells, d-Asb11 coimmunoprecipitates with various ankyrin repeat–containing proteins, including I-κBα and Notch, but not recombining binding protein-Jκ, which does not contain ankyrin repeats (). The loss and gain of function experiments in zebrafish embryos suggest that d-Asb11 acts as a molecule that supports the maintenance of neuronal precursors possibly by safeguarding the proper identity and expansion of the precursor stem cell compartment. This notion was further tested using a specific neuronal progenitor cell line (PC12) and a pluripotent embryonic carcinoma cell line (Nt2-D1; ; ). Experiments on PC12 cells were performed using an NGF-induced in vitro differentiation of PC12 pheochromocytoma cells according to the standard protocols (; ). Upon NGF stimulation, the cells transfected with control DNA showed a decrease in proliferation, which was determined by MTT (3-[4,5-dimethyl-2-thiazolyl]-2,5-diphenyl-2H-tetrazolium bromide) activity, and the expression levels of proliferating cell nuclear antigen (PCNA) and neurites were formed (). Strikingly, when d-Asb11 was overexpressed in PC12 cells, the proliferation was sustained, and the amount of neurites was significantly reduced (). Importantly, PC12 cells overexpressing d-Asb11 appear to continue proliferating upon NGF-induced differentiation, as they form overlaying colonies in contrast to control cells that remained in a monolayer (). The lack of NGF-induced neurite extension in d-Asb11–expressing PC12 cells may be caused by the incapacity of these cells to initiate neuronal differentiation but, alternatively, may reflect a block further down in the execution of the neuronal differentiation program. To distinguish between these possibilities, we investigated the expression of growth cone–associated protein 43 (GAP-43), a marker already expressed in proliferating neuroblasts, and the neurofilament, a marker of terminally differentiated neurons (; ; ). As expected from the neurite extension experiments, transfection of into PC12 cells inhibited neurofilament expression as assayed either by immunoblotting () or immunofluorescence (). Importantly, however, d-Asb11 enhanced the NGF-dependent expression of GAP-43 (). To examine the effects that d-Asb11 may have upon the differentiation of uncommitted pluripotent embryonic carcinoma cells (, ), we chose Nt2-D1 cells that differentiate primarily into neurons upon RA treatment (). When d-Asb11 was overexpressed in the Nt2-D1 cell line, it inhibited the RA-induced differentiation into neurons, as shown by diminished neurofilament expression (). Importantly, as in PC12 cells, these cells were still mitotically active, as shown by increased PH3 expression, suggesting that these pluripotent cells are also blocked in a phase before terminal neuronal differentiation. Thus, the overexpression of d-Asb11 in both PC12 and Nt2-D1 cells enhances proliferation of the committed neural precursor in spite of differentiation stimuli. These in vitro data are therefore consistent with the inhibition of neurogenesis upon the misexpression of d-Asb11 in the embryo. We have shown that in the developing zebrafish embryo, is expressed in the neural plate margins and is complementary to and abutting the proneuronal zone, as defined by staining. Before early somitogenesis, is ubiquitously expressed in the pluripotent cells of the blastoderm (). Progressive restriction of to the lateral regions of the neural plate as well as its transient expression at that location suggests that d-Asb11 may indeed support the stem cell fate. Upon the knockdown or misexpression of , the size of the terminally differentiated neuronal compartment increases or decreases, respectively, suggesting that d-Asb11 is involved in regulation of the number of neurons that eventually arise. The relatively subtle minor initial increase in neurogenesis upon d-Asb11 knockdown, as not all dorsal or lateral precursors are induced to differentiate, may implicate redundancy with other ASB family members from this group (Fig. S1). Alternatively, the strictly localized and transient expression of could affect only a small subset of neural precursors, thereby resulting in this subtle phenotype. We propose that during zebrafish development, d-Asb11 maintains the undifferentiated neural-competent state of a subpopulation of cells in the neural plate until its expression is down-regulated by subsequent differentiation signals (e.g., RA; ). Because is only transiently expressed from 4 hpf until the 12-somite stage, upon its disappearance, cells devoid of d-Asb11 would become sensitive to secondary inductive signals and would be able to pursue the appropriate neuronal cell fate. Interestingly, the timing of extinguishing mRNA expression at ∼12 somites roughly corresponds with the final round of DNA synthesis occurring between 9 and 16 hpf in motor neurons (). As first secondary neurons are born ∼6 h after the first primary neurons, it may be that d-Asb11 functions to separate waves of primary and secondary neurogenesis by maintaining a pool of precursors to respond to inductive signals operating later in development. The loss of function of d-Asb11 causes premature neuronal commitment that is reflected in ectopic expression and followed by a relative increase of HuC+ terminally differentiated neurons. In parallel, a progressive loss of ventricular zone cells occurs, resulting in reduction of the definitive neuronal compartment. Importantly, d-Asb11 may mediate its effects through control of SoxB1 neural precursor genes, as the knockdown results in a reduced and aberrant expression of and respectively. Concurrently, when misexpressed, d-Asb11 prevents neuronal differentiation possibly through the induction of or yet another SoxB1-type factor. Both gain of function and loss of function of d-Asb11 phenotypes in zebrafish are in agreement with experiments in the chick in which the overexpression of Sox2 and/or Sox3 also inhibits neuronal differentiation of neural progenitors, which then maintain their undifferentiated state and capacity to proliferate. In contrast, dominant-negative Sox2 and/or Sox3 in neural progenitors results in premature exit from the cell cycle, and neuronal differentiation ensues (; ). Our data are consistent with a model in which d-Asb11 supports the maintenance of a proliferative subpopulation of neural precursors. We further propose that d-Asb11 may control this process through SoxB1 neural precursor genes. Consistent with such a model, it has been proposed that SoxB1 genes play a role in maintaining neural progenitors in the cell cycle and regulate the timing of their exit from the cell cycle. The findings that the inhibition of SOX2 function in chick neural progenitors () and rat oligodendrocyte precursor cells () leads to their premature exit from mitosis support this hypothesis and are in agreement with loss of function d-Asb11, as d-Asb11 knockdown similarly leads to the reduction of mitotic cell numbers after 24 hpf. The role of d-Asb11 in maintaining the proliferative state of precursors was also illustrated in in vitro differentiation models. When d-Asb11 is overexpressed in a neural-committed progenitor or a pluripotent cell line, these cells are unable to execute the terminal neuronal differentiation program, as shown by the inhibition of neurofilament expression upon exposure to stimuli that cause neuronal differentiation. Consistently, the proliferation was not inhibited in these cells as it was in the controls, indicating that d-Asb11 is able to block terminal neuronal differentiation and allow cells to proliferate. Because PC12 is already committed to neural crest fate, the overexpression of d-Asb11 cannot prevent the PC12 cell in activating its neuronal differentiation program (enhanced GAP-43 expression in d-Asb11–transfected cells), but it can hold these cells in a state of proliferating neuroblast precursors. Consistent with the proposed role of d-Asb11 in the maintenance of neural precursors, GAP-43 is reported to be expressed in proliferating neuroblasts, and its function has been implicated in the cell cycle, differentiation, and, eventually, also cell survival (; ; ). Importantly, PC12 cells overexpressing d-Asb11 appear to have a block in further execution of the neural differentiation program (as they have diminished neurites and lower neurofilament expression). Our data suggest that d-Asb11 acts to maintain cells in the neuroblast state independently of their primary status, being either pluripotent (Nt2-D1) or neural progenitor (PC12) cells. Consistently, the misexpression of XSox3 with XBF1 promotes the proliferation of neuroectodermal cells in embryos (). We speculate that the misexpression of d-Asb11 in zebrafish embryos perhaps through the induction of or another SoxB1-type factor may similarly affect the fate of zebrafish neuroectodermal cells, thereby abolishing neuronal differentiation. The in vitro data on Nt2-D1 and PC12 cells, a pluripotent and a neuronal-committed cell line, respectively, also suggest that d-Asb11 functions by maintaining the proliferative state of the neural precursors, thereby making them refractory to differentiation signals perhaps through regulating SoxB genes. Our findings show that d-Asb11 functions as a regulator of progenitor cell maintenance. More specifically, it keeps progenitors in a less differentiated, more proliferative state. Although little is known about the functions of ASB proteins, the fact that ASB5 (which is highly homologous to d-Asb11) has a role in arteriogenesis strengthens the notion that this class of ASB proteins may have a role in cell fate decisions (). Interestingly, a possible role for d-Asb11 in the development of the brain has recently indirectly been uncovered by the experiments performed by . They compared the genetic mutations between the human and chimpanzee genome, looking for genes with an unusually higher number of functional nucleotide changes between the two species, which would suggest a possible important role for such genes in the definition of being a chimpanzee or a human. Most of these genes were involved in the immune and reproductive systems, but ASB11 was among the highly altered genes between the human and chimpanzee genome despite its generally very high level of conservation in the chordate phylum. It is of course tempting to suggest that the obvious size differences between human and chimpanzee brains are also a result of altered ASB11 function. Zebrafish were kept at 27.5°C. Embryos were obtained by natural matings, cultured in embryo medium, and staged according to ; ). HEK293, HeLa, Nt2-D1, and PC12 cells were maintained in DME containing 10% FCS. The culture medium was supplemented with 5 mM glutamine and antibiotics/antimitotics. Cells were incubated in a 5% CO humidified incubator at 37°C. pCS2+MT (MT; myc-tagged expression vector), pCS2+MT–d-Asb11 (MT–d-Asb11), pCS2+MT–d-Asb11–ΔC (MT–d-Asb11–ΔC), and pMT2SM-HA–d-Asb11 (HA-tagged expression vector; HA–d-Asb11) were constructed as follows: an expression vector containing , the coding region, was cloned into the BamHI–XhoI sites of the pCS2+ expression vector. To obtain a myc-tagged version of (), the coding region was cloned into the NcoI–XhoI sites of pCS2+MT. An construct encoding a C-terminal deletion (aa 229–293) was generated by digesting with XbaI. All expression constructs were checked for errors by sequencing. For HA-tagged d-asb11 (HA–d-asb11), d-asb11–ΔC (HA–d-asb11–ΔC), and d-asb11–ΔSOCS (HA–d-asb11–ΔSOCS), the coding region was cloned into the XhoI–EcoRI sites of pMT2SM-HA, and the expression constructs were subsequently verified by sequencing. To obtain a construct for RNA synthesis of to control mRNA injections of MT–d-asb11 mRNA, the 1.6-kb cDNA was cloned into the BglII and XhoI sites of pT7TS+. A partial cDNA fragment of in pBluescript was used as a template to generate a riboprobe for in situ hybridizations. To misexpress wild-type or mutant versions of mRNA, zebrafish embryos were injected at the two-cell stage in one of the two blastomeres (300 pg mRNA), resulting in defects in the injected half of the embryo. To misexpress MT–d-asb11 DNA, 10 pg was microinjected into zygotes. Whole-mount in situ hybridizations were performed according to and for sox2 were performed according to . Double label whole-mount in situ hybridization was performed according to . For histological analysis, upon in situ hybridization, embryos were overstained for 48 h, subsequently fixed for 1 h in 4% PFA in phosphate buffer, dehydrated, embedded in plastic, and sectioned. and probes were provided by B. Appel (Vanderbilt University, Nashville, TN). Probes for and were provided by L. Bally-Cuif (Institute of Developmental Genetics, Neuherberg, Germany). U. Strahle (Institute of Toxicology and Genetics, Forschungszentrum Karlsruhe, Karlsruhe, Germany) provided the probes for and . V. Cunliffe (Centre for Developmental Genetics, University of Sheffield, Sheffield, United Kingdom) and S. Wilson (University College London, London, United Kingdom) provided the and probes, respectively. Acridine orange staining was performed as described previously () with modification of the acridine orange (Sigma-Aldrich) solution (0.078 μg/ml in embryo medium). TUNEL assay was performed as described previously by with modification of the staining reaction. Antidigoxigenin antibody was preabsorbed in blocking buffer (1% DMSO and 2% BSA in PBS) for 1 h at 4°C. Embryos were incubated overnight in blocking buffer with 1:2,000 antidigoxigenin antibody at 4°C, washed eight times with PBS, and stained according to the method used for in situ hybridization. A random primed λZAP neurula cDNA library prepared from 3 to 15 hpf embryos was used to screen 1.5 × 10 phages to obtain a full-length clone. The probe was labeled to high specific activity with α-[P]dCTP (GE Healthcare) using a Rediprime labeling kit (GE Healthcare). After in vivo excision, the longest positive clone was subcloned and sequenced. Antisense MOs (Gene Tools, LLC) were designed to complement the 5′ untranslated sequences of (d-asb11–MO-1; 5′-AGAAACCTCGCAGACAGCAACGGTC-3′) or sequences upstream and including the ATG start site (d-asb11–MO-2; 5′-CCATCTCTAAACTAAAACACAGCCA-3′), and, as a control, an inverted MO (d-asb11–MO; 5′-CTGGCAACGACAGACGCTCCAAGA-3′) was used. Approximately 9 ng/1 nl was injected into one-cell stage embryos, whereas as a control, we injected ∼15 ng/1 nl of inverted MO. For whole-mount immunohistochemistry, fixed embryos were rinsed in PBS and incubated in blocking solution for 1 h (PBS + 0.1% Tween 20 containing 1% BSA and 2% normal lamb serum). The primary antibody (anti-HuC; Sigma-Aldrich) was added, and embryos were incubated overnight at 4°C. After washing in PBS + 0.1% Tween 20, embryos were incubated overnight at 4°C in goat anti–mouse peroxidase-conjugated secondary antibody. After extensive washing, embryos were incubated in 0.5 mg/ml DAB and reacted in 0.003% HO. For the double fluorescent whole-mount antibody staining with anti-HuC and anti-PH3 antibody (Upstate Biotechnology), fixed embryos were permeabilized for 4 min in 2.5 mg/ml ice-cold trypsin in PBT (PBS + 0.1% BSA + 0.1% Tween 20; Worthington Biochemical Corp.) on ice, subsequently washed three times for 5 min each in PBT, and incubated for at least 1 h in blocking buffer (2% lamb serum, 0.1% BSA, 0.1% DMSO, and 1% Triton X-100 in PBS). To detect HuC and PH3, respectively, secondary antibodies were Cy3 and Cy5 conjugated. Immunofluorescently labeled embryos were analyzed in 100% glycerol using confocal laser scanning microscopy at ambient temperature. To determine numbers of PH3-positive cells, they were scored manually on the cumulative z series for each embryo (for see Results). The cells were counted at 30,000× magnification in a standardized square of 7-cm length and depth covering the extent of the neural tube defined by HuC antibody labeling. Comparable regions were chosen in morphants and wild types with reference to the yolk-tube extension. Volumes were calculated from a 3D reconstruction from 7-μm serial sections with an acquisition station (). This resulted in 123 section images for the control and 149 section images for the treated fish. The notochord and otic vesicle were included to have a frame of reference. Pictures were obtained using a microscope (Axioplan; Carl Zeiss MicroImaging, Inc.) with a 10× NA 0.30 plan Neofluar or a 40× NA 0.65 Achroplan lens on a Leica microscope (TCS NT). Images were digitized with a camera (DFC480; Leica) and processed with the IM500 Image Manager (Leica). A stereomicroscope (MZ FLIII; Leica) with a camera (DC 300F; Leica) was used for and . Digital pictures of transversal sections of in situ hybridizations were then quantified with Image Analysis software (EFM Software) by calculating the total signal intensity. The pictures of sections were corrected from embryo size by dividing the amount of staining by the size of the embryo surface. PC12 cells were transfected with MT alone or MT–d-Asb11 plus neomycin resistance plasmid (see supplemental material for plasmid construction). After 5 d, polyclonal neomycin-resistant PC12 cells were reseeded in 96-well plates and stimulated with 100 ng/ml NGF-2.5S (Invitrogen). After 3 or 6 d, 0.1 mg MTT (M5655; Sigma-Aldrich) was added to the medium. The cells were lysed by the addition of 50 μl MTT lysis buffer (20% SDS and 10% dimethylformamide) to the medium, and absorbance was measured at 570 nm. The results are the mean and standard error of 11 independent cultures obtained in two different experiments. The mean neurite length in was measured by blinded counting of the total amount of neurites divided by the amount of counted cells using the Image Analysis software package (EFM Software). Nt2-D1 cells were seeded in a 24-well plate 1 d before transfection using the MATra and IBAFect (IBA GmbH) transfection technique according to the standard supplier's protocol. After 24 h, the medium was refreshed and 10 M RA was added, and, after 48 h, the cells were lysed (Cell Signaling) and supplemented with phosphatase and protease inhibitors. GST-tagged d-Asb11 was purified from BL21 bacteria using glutathione–Sepharose beads. Rabbits were immunized with GST-tagged d-Asb11. Subsequently, the serum containing the polyclonal antibody directed against d-Asb11 was used in immunoblotting analysis. Embryos were lysed in cell lysis buffer (50 mM Hepes, pH 7.5, 150 mM NaCl, 1.5 mM MgCl, 1 mM EGTA, 10% glycerol, and 1% Triton X-100) at 4°C (7.5–12.5 μl/embryo). Loading buffer was added to the cell lysis buffer, and samples were boiled for 3 min. A total equivalent of five embryos was loaded per lane. Supplemental material provides information about the identification of d-Asb11 and testing of d-asb11 morphants. Fig. S1 presents the characterization of d-Asb11 and the Asb subfamily of proteins. Fig. S2 shows the generation of a d-Asb11 antibody and validation of d-Asb11 expression during embryogenesis. Online supplemental material is available at .

Endothelial cells are contact inhibited in their growth and lose the capacity to respond to growth factors when they reach confluence. This phenomenon is mediated by different concurrent mechanisms. Molecules at cell to cell junctions, such as cadherins, may transfer signals that reduce the capacity of the cells to respond to proliferative stimuli (; ). Cadherins are located at intercellular adherens junctions and are linked to different intracellular partners that include β-catenin, plakoglobin, p120, Src (), csk (), and density-enhanced phosphatase-1 (DEP-1)/CD148. β-catenin, plakoglobin, and p120 can also translocate to the nucleus and modulate cell transcription. In tumor cells, the negative effect of epithelial cadherin (E-cadherin) on cell growth is a result of its capacity to bind β-catenin and inhibit its translocation to the nucleus. This effect is detected in tumor cell lines in which cytosolic β-catenin ubiquitination and destruction is impaired (; ; ; ; ). Endothelial cells express a cell-specific cadherin called vascular endothelial cadherin (VEC). This protein exerts a negative effect on cell growth by binding VEGF receptor (VEGFR) type 2 and inhibiting its signaling activity (; ; ; ). VEGF is a major growth factor for endothelial cells and plays an important role in the formation of new vessels during embryogenesis and in proliferative diseases (; ; ). In blood endothelium, the activities of VEGF are mediated by its interaction with two tyrosine kinase receptors, VEGFR-1 (flt-1) and -2 (flk/KDR), as well as neuropilins. The growth signals are transferred, to a large extent, through the activation of PLC-γ, PKC, and subsequently p44/42 MAPK (, ; ; ). We found that in contact-inhibited endothelial cells, VEGFR-2 forms a complex with VEC that results in the inhibition of its tyrosine phosphorylation and, consequently, in the attenuation of MAPK activation. This effect was attributed to the phosphatase DEP-1/CD148 that, by binding β-catenin and p120, may associate with the cadherin–receptor complex and dephosphorylate the receptor (). In this study, we go further by describing another aspect of this phenomenon. Upon activation with specific ligands, growth factor receptors are internalized via clathrin-dependent and -independent pathways. In many cases, this process leads to signaling termination via degradation of the activated receptor complex. Therefore, internalization is considered an important mechanism through which cells may control the intensity and duration of signal transduction. However, more recent findings indicate that internalization is not just a sink through which receptors are degraded (; ). On the contrary, some receptors, such as TGF-β, EGF, or NGF receptors, can maintain their signaling activity from within intracellular compartments (; ; ; ). Little is known about the internalization pathways followed by VEGFR-2 or their functional significance (; ; ). It has been reported that cadherins may influence growth factor receptor internalization, but the extent to which they do depends on the cadherin or growth factor receptor. In tumor cell lines, N-cadherin forms a complex with FGF receptor 1 that inhibits its internalization and degradation. This causes a sustained FGF signaling and abnormal cell growth (). In contrast, E-cadherin cointernalizes with FGF receptor 1, which facilitates its nuclear translocation and signaling activity (; ). In this study, we analyzed the role of VEC on VEGFR-2 internalization and signaling in endothelial cells. We found that the receptor is internalized more rapidly and efficiently when VEC is absent or not clustered at intercellular contacts. Strikingly, internalization does not terminate receptor signaling, which instead continues in endosomes. This may explain why VEC-null cells present increased and uncontrolled growth. We first investigated whether the establishment of cell to cell contact modulates VEGFR-2 internalization. Using freshly isolated human umbilical vein endothelial cells (HUVECs) stimulated with VEGF, we observed that VEGFR-2 endocytosis, which was evaluated by immunofluorescence labeling of intracellular vesicular compartments, was significantly reduced by cell density (). Time course analysis revealed that in sparse cells, the number of receptor-positive vesicles increased more rapidly and to a larger extent than in confluent cells (, bottom). This first observation suggested that the establishment of cell to cell contact reduced VEGFR-2 internalization. Because VEC plays a role in VEGFR-2 signaling, we investigated whether VEC could be involved inVEGFR-2 internalization. We compared syngenic endothelial cell lines differing for the expression of VEC. These cells had been characterized previously in detail and presented superimposable levels of VEGFR-2 (see ; , ). As shown in , after the addition of VEGF, the number of VEGFR-2–containing vesicular compartments is markedly higher in the absence of VEC. Quantification of the amount of biotinylated receptor that was internalized, degraded, or recycled back to the plasma membrane is reported in . In VEC-null endothelium, the receptor is internalized more quickly and to a higher extent than in VEC-positive cells (). The overall amount of internalized receptor for the duration of the experiment is about fourfold more in VEC-null than -positive cells. Receptor degradation exceeds recycling by about fivefold in both cell types, but both parameters are significantly increased in the absence of VEC (). These data indicate that a higher amount of VEGFR-2 is internalized, degraded, and recycled in the absence of VEC. Internalization of growth factor receptors may follow clathrin-dependent or -independent pathways. Among the latter, caveolae have been shown to regulate receptor internalization directed toward degradation (; ). Our codistribution experiments of VEGFR-2 with early endosomal antigen-1 (EEA-1) and caveolin-1 show that VEGFR-2 internalizes mostly in EEA-1–positive early endosomes () and to a very low extent in caveolae. Colocalization of VEGFR-2 and the caveolar component PV-1 (; ) was also negligible (unpublished data). To control whether caveolae were expressed correctly and to a comparable extent in both VEC-null and -positive cells, we costained these structures with PV-1 and caveolin antibodies. As shown in Fig. S1 (available at ), the extensive and comparable colocalization of caveolin and PV-1 could be observed in both cell types, suggesting structural integrity of the caveolar compartment. The preferential distribution of VEGFR-2 in EEA-1–positive endosomes was further confirmed using immuno-EM (). In addition, silencing clathrin heavy chain expression either by siRNA (see and Fig. S3, available at ) or disrupting clathrin-coated pits by hypertonic medium blocked receptor internalization both in VEC-positive and -null cells (). In contrast, after incubation with filipin at a concentration able to fully disrupt lipid rafts and caveolae (; ), receptor internalization did not significantly change (). We first tested whether the internalized receptor retained tyrosine phosphorylation by cell fractionation on an iodixanol gradient (). Antibodies recognizing phosphotyrosine (PY) 1214– and PY1054/59–VEGFR-2 were used. As shown in , in the absence of VEC, a higher amount of phosphorylated VEGFR-2 is detected in intracellular fractions, whereas in the presence of VEC, the phosphorylated receptor remains preferentially in fractions corresponding to peripheral plasma membranes. Phosphorylation of VEGFR-2 tyrosine 1175 is required for binding and activation of PLC-γ, which is the major effector of VEGF-mediated cell proliferation (). By using antibodies specific for PY1175–VEGFR-2, we observed that in the absence of VEC, a higher amount of internalized receptor was phosphorylated at this specific tyrosine (). Consistently, more internalized PY1175–VEGFR-2 was found in sparse than in confluent HUVECs (Fig. S2, available at ). To further prove that PLC-γ could be activated by the internalized receptor, we stained the cells with antibodies directed to the active PY783–PLC-γ. As shown in , active PLC-γ codistributes with internalized VEGFR-2 more effectively in the absence than in the presence of VEC. To further test this hypothesis, we prevented receptor internalization by silencing clathrin with two specific siRNAs (, , and S3) that target independent sequences of clathrin heavy chain messenger. As expected, VEGF-induced phosphorylation of VEGFR-2 and activation of p44/42 MAPK were higher in the absence than in the presence of VEC (). When siRNA clathrin was applied to VEC-null cells, both receptor and MAPK phosphorylation dropped to values comparable with those of VEC-positive cells. In contrast, in the presence of VEC, the effect of siRNA clathrin was either weak or undetectable. Consistently, the inhibition of MAPK activation was also observed by treating the cells with hypertonic medium, whereas treatment with filipin did not modify MAPK activation in response to VEGF either in VEC-positive or -null cells (unpublished data). Overall, these results strongly suggest that internalization protects the receptor from dephosphorylation and, therefore, increases and prolongs its proliferative signaling. We then investigated the mechanism through which VEC inhibits VEGFR-2 internalization. In a previous study, we found that VEC forms a complex with VEGFR-2, and we analyzed the domains of VEC involved in this process (). VEC mutants lacking either the β-catenin– or p120-binding domains were unable or less efficient, respectively, to coimmunoprecipitate VEGFR-2. We found that these mutants were also unable to significantly prevent VEGFR-2 internalization (Fig. S4, available at ), suggesting that receptor internalization is reduced as a consequence of binding to VEC. Similar to E-cadherin (for review see ), VEC can be internalized through clathrin-coated pits (). We tested whether VEC codistributes with VEGFR-2 in intracellular compartments. As reported in upon VEGF activation, no significant codistribution of VEC with VEGFR-2–positive vesicles is detected in VEC-positive cells and HUVECs (Fig. S4). Only junctional colocalization can be observed (). The lack of codistribution in internal compartments was confirmed by immuno-EM (unpublished data). Collectively, these data suggest that the receptor internalizes upon dissociation from VEC. In previous studies, we showed that the phosphatase DEP-1/CD148 can reduce VEGFR-2 signaling. DEP-1/CD148 can associate with β-catenin and p120 (; ) and with the VEC–VEGFR-2 complex, reducing VEGFR-2 phosphorylation (). Therefore, we tested whether DEP-1/CD148 could also reduce receptor internalization. As reported in , in endothelial cells transfected with DEP-1/CD148 siRNA, VEGFR-2 internalization is significantly higher. This effect is accompanied by an increase in VEGFR-2 phosphorylation and MAPK activation (). These data suggest that retention of VEGFR-2 at the membrane by VEC allows its dephosphorylation by DEP-1/CD148 and limits its internalization and signaling. In this study, we report a novel aspect of the mechanism through which VEC expression and clustering inhibits VEGFR-2 proliferative signaling. We found that in the absence of VEC or in conditions in which VEC is not clustered at adherens junctions as in sparse cells, VEGFR-2 is endocytosed to a higher extent in intracellular compartments, from where it maintains its signaling activity. VEC could therefore reduce receptor activity by inhibiting VEGFR-2 internalization and promoting its inactivation at the cell surface. In our experimental conditions, VEGFR-2 is internalized in early endosomes mostly through a clathrin-dependent pathway. We were unable to detect caveolin-1–positive vesicles containing VEGFR-2, and we could not inhibit receptor internalization using a caveolae-perturbing drug such as filipin (; ). Other studies found the codistribution of VEGFR-2 with caveolin-1 (; ; ), and we cannot exclude that under different experimental conditions, VEGFR-2 may be internalized through caveolae. However, the observed association with caveolin-1 may also represent a mechanism of receptor compartmentalization at the plasma membrane. It was found that caveolin-1 would form a molecular complex with VEGFR-2 that inhibits receptor activation in resting cells. Upon activation of the cells with VEGF, caveolin-1 is phosphorylated, and the complex rapidly dissociates (). Thus, it is tempting to speculate that similar to TGF-β receptor (), once VEGFR-2 is released from the caveolin-1 complex, it becomes available for internalization in clathrin-coated pits. In agreement with this model, the overexpression of caveolin-1 in transgenic mice reduces permeability and angiogenic response to VEGF (). It has been reported that VEGFR-2 internalization and degradation are regulated by ubiquitination through a Cbl-dependent mechanism () or C-tail serine phosphorylation by activated PKC (). These mechanisms may coexist and may be responsible for the amount of receptor degradation reported here. Our observations support the hypothesis that internalized VEGFR-2 maintains its activity. These data are in agreement with recent publications indicating that signaling through growth factor receptors does not occur only at the cell membrane but may continue even more effectively from intracellular compartments (; ; ; ). Clathrin-dependent internalization of TGF-β in early endosomes, where the Smad2 anchor SARA is enriched, promotes TGF-β signaling (). EGF receptor is internalized shortly after ligand addition in intracellular compartments together with its downstream signaling factors shc, Grb2, and mSOS () and maintains its signaling activity (). Similarly, the specific activation of endosome-associated PDGF receptor leads to the activation of its major signaling pathways (). Thus, as for VEGFR-2, endocytic transport is important not only for receptor turnover but also for regulating signal transduction and for mediating the formation of specialized signaling complexes. A novel aspect of our work is that VEC inhibits VEGFR-2 internalization, thereby reducing its cell growth signaling activity. How can VEC inhibit receptor endocytosis? A likely hypothesis is that VEC retains VEGFR-2 at the membrane by binding to it. In addition to our studies, others have reported (; ; ; ) that VEGFR-2 couples with VEC. This process requires the binding of VEC to β-catenin and, to a lesser extent, to p120. In the present study, we found that mutants of VEC lacking the cytoplasmic domain responsible for binding either β-catenin or p120 and unable to associate with VEGFR-2 () do not prevent VEGFR-2 internalization. This supports the idea that VEC–VEGFR-2 coupling is required to inhibit receptor endocytosis. Cadherins themselves are endocytosed via several different routes, including clathrin-dependent (; ; ; ) and -independent pathways (; ). Therefore, cotrafficking of receptor and cadherin complexes is possible (for reviews see , ; ). However, under our experimental conditions, we could detect VEC in intracellular compartments (), but we could not see codistribution with the receptor. Thus, it is likely that the receptor dissociates from VEC before internalization (). In a previous study, the phosphatase DEP-1/CD148 was found to play a role in the inhibitory effect of VEC on VEGR-2 signaling (). This phosphatase associates with VEC through its binding to β-catenin and p120 and, in this way, reduces VEGFR-2 phosphorylation (). We report that DEP-1/CD148 could prevent VEGFR-2 internalization along with the reduction of receptor phosphorylation and signaling. Therefore, it is possible that VEC, by retaining VEGFR-2 at the cell surface, allows its dephosphorylation by DEP-1/CD148, which, in turn, inhibits its internalization and signaling. Besides VEC, VEGFR-2 was found to bind to integrins (). Another study reported that when cells are plated on collagen I, the phosphatase SHP2 can associate with VEGFR-2 and stimulate its internalization. SHP2 activates Src, which in turn activates dynamin II–dependent receptor internalization (). Interestingly, this phenomenon does not occur when cells are plated on vitronectin, and SHP2 does not bind to VEGFR-2 (). These observations suggest that the capacity of different adhesive proteins to complex with growth factor receptors and modulate their internalization and signaling may be a general paradigm. In this way, cells may modulate their growth and survival as a function of density and interaction with specific matrix proteins. In conclusion, the results reported in this study are consistent with the idea that VEGFR-2 proliferative signaling is increased by endocytosis. The inhibitory role of VEC is likely that of binding and retaining the receptor at the cell surface, preventing its endocytosis, and favoring inactivation by DEP-1/CD148. This suggests that the modulation of VEC–VEGFR-2 complex formation may be a novel strategy to regulate VEGF proliferative signaling and, therefore, to inhibit or stimulate angiogenesis. For the detection of VEGFR-2, anti–human VEGFR-2 (single chain recombinant; clone scFvA7 with E tag; RDI and Fitzgerald) and anti–mouse VEGFR-2 (rat clone Avas12α1; RDI and Fitzgerald) were used for immunofluorescence; rabbit polyclonal C-1158 (sc504; Santa Cruz Biotechnology, Inc.) was used for Western blotting. Antibodies to tyrosine-phosphorylated VEGFR-2 were rabbit polyclonal PY1214 and PY1054/59 (Biosource International) and rabbit polyclonal PY1175, which was provided by M. Shibuya (University of Tokyo, Tokyo, Japan). Antibodies to clathrin heavy chain were mouse monoclonal cloneX22 (Affinity BioReagents, Inc.) for immunofluorescence and mouse monoclonal clone 23 (R&D Systems) for Western blotting. Antibody to EEA-1 was goat polyclonal N-19 (sc-6415; Santa Cruz Biotechnology, Inc.); antibody to caveolin-1 was rabbit polyclonal N-20 (sc-894; Santa Cruz Biotechnology, Inc.); and antibody to VEC was goat polyclonal C-19 (sc-6458; Santa Cruz Biotechnology, Inc.) and mouse monoclonal BV6 and BV9 (produced in our laboratory; ). Antibody to PY783-PLCγ, total p42/44 MAPK, and phospho-p42/44 MAPK was rabbit polyclonal (Cell Signaling). Antibody to DEP-1/CD148/CD148 was goat polyclonal (R&D Systems), and antibody to PV-1 was rat monoclonal (provided by R. Stan, Dartmouth Medical School, Lebanon, NH). Endothelial cells with a homozygous null mutation of the gene (VEC null) and the cell lines derived from them through retroviral gene transfer and expressing wild-type (VEC positive) or various VEC mutant constructs were generated and characterized as described previously in detail (). For all of the experiments, 50,000 cells/cm (to reach confluence within 24 h) were seeded in complete culture medium and cultured without medium change for 72 h. Cells were then washed once with MCDB 131 (Life Technologies) and starved in 1% BSA in MCDB 131 (starving medium) for 18–20 h. 2 h before activation, cells were washed once with MCDB 131 and further incubated in fresh starving medium. Cells were treated with 80 ng/ml VEGF (human recombinant VEGF 165; PeproTech) in fresh starving medium (fresh starving medium alone was used for controls) for the indicated intervals at 37°C. If not otherwise indicated, VEGF treatment was for 10 min. HUVECs were cultured in MCDB 131 with endothelial cell supplements as described previously (). For the experiments 1,800 and 42,000 cells/cm were seeded to obtain sparse and confluent cultures, respectively. HUVECs were then treated as described in this section for mouse endothelial cells except that starving was reduced to 6 h before stimulation with VEGF. Caveolae organization was altered by treatment with 1 μg/ml filipin (filipin III from ; Sigma-Aldrich) for 1 h (; ). Clathrin pit–mediated endocytosis was inhibited using hypertonic medium (0.45 M sucrose in MCDB 131 with 1% BSA) for 30 min to affect as described previously (; ). Cells were then stimulated and assayed as indicated in the specific sections. ext-link sup #text Internalization, recycling, and degradation were measured as described previously by with the following modifications. Cells were put on ice and washed three times with ice-cold PBS containing Ca and Mg (Ca/Mg PBS). For surface biotinylation, cells in Ca/Mg PBS were treated with 0.5 mg/ml of thiol-cleavable Sulfo-NHS-S-S-Biotin (Pierce Chemical Co.) for 1 h on ice. They were then washed on ice twice with Ca/Mg PBS, once with MCDB 131, and once with 1% BSA MCDB 131. 80 ng/ml VEGF in fresh 1% BSA MCDB 131 was added, and cells were incubated at 37°C for the time indicated to allow internalization. The cultures were then put back on ice and washed three times with ice-cold Ca/Mg PBS. Samples were incubated twice for 20 min with 45 mM of the membrane-nonpermeable reducing agent GSH in 75 mM NaCl, with 75 mM NaOH and 1% BSA added just before use (stripping buffer). Cells were further washed twice on ice with Ca/Mg PBS and incubated for 15 min with iodoacetamide (in Ca/Mg PBS with 1% BSA; quenching buffer) to quench free sulfo-reactive groups. To evaluate total labeling, a sample for each cell type was not reduced with GSH. To control background, a sample was labeled and reduced without incubation at 37°C. For immunoprecipitation, cells were washed with Ca/Mg PBS and extracted on ice in 50 mM Tris-HCl, pH 7.6, 150 mM NaCl, 1% Triton X-100, 1% NP-40, and a cocktail of protease inhibitors (Set III; Calbiochem). Extracts were precleared for 90 min with protein A–agarose beads and incubated overnight with 5 μg anti–VEGFR-2 (rabbit sc-504), and the immunocomplexes were collected on protein A–agarose beads for 90 min. After five washes in extraction buffer (the last one containing 0.1% Triton X-100), proteins were eluted by boiling for 10 min in nonreducing laemmli sample buffer. Samples were analyzed by SDS-PAGE followed by Western blotting on nitrocellulose membrane and revealed by ECL chemiluminescence. Band intensity was quantified by ImageJ analysis (National Institutes of Health; freely available at ). To quantify VEGFR-2 recycling and degradation, cells were labeled as described in the first paragraph of this section, and endocytosis was allowed for 10 min in the presence of 80 ng/ml VEGF (peak time for VEGF-induced VEGFR-2 internalization both in VEC-null and -positive cells as determined in internalization experiments). Samples were then reduced as described in the first paragraph of this section to remove the label from the residual cell surface receptor. The internalized fraction was chased by reincubation at 37°C for 10 and 20 min in duplicate samples. One sample was reduced to evaluate the amount of VEGFR-2 that recycled back to the plasma membrane, and the other sample was left unreduced to measure degradation. The samples were then processed as described in the first paragraph of this section. VEGFR-2 degradation was calculated by subtracting the value of residual biotinylated receptor after incubation at 37°C without reduction (i.e., internalized + recycled − degraded) from the total pool of internalized receptor. VEGFR-2 recycling was calculated by subtracting both the degradation value and the value of residual biotinylated receptor after incubation at 37°C and reduction (i.e., internalized − recycled − degraded) from the total pool of internalized receptor. Cells were cultured in 35-mm diameter petri dishes as described in Cell types and culture conditions. After the treatments indicated in the specific sections, culture medium was removed, and cells were fixed in 1% PFA in 2.5 mM triethanolamine, pH 7.5, containing 0.1% Triton X-100 and 0.1% NP-40 for 25 min at RT (). Before staining, 0.5% Triton X-100 in PBS was added for 10 min at RT. In some experiments, immunofluorescence microscopy for VEGFR-2 was performed using the Avas12α1 antibody after cell fixation (in vitro staining). The fixation/permeabilization method applied () allows an optimal observation of VEGFR-2 in internal compartments with in vitro staining (primary antibody after cell fixation). Data obtained with in vivo (see Internalization assays) and in vitro staining were superimposable for both VEC's effect on VEGFR-2 vesicular labeling and the codistribution of VEGFR-2 with markers of specific compartments such as EEA-1 and caveolin-1. A comparison of VEGFR-2 vesicular distribution after in vivo staining (with and without acid wash before fixation) is shown in Fig. S5. For immunogold labeling, 4% PFA/0.4% glutaraldehyde in PBS was added in a 1:1 ratio to culture medium. After 2 h at RT, the fixative was discarded, and cells were scraped in 1% PFA in PBS, collected in Eppendorf tubes, and processed for ultrathin cryosectioning as described previously (). Double immunogold labeling was performed as described previously (). Subcellular fractionation on an iodixanol gradient was performed as described previously (). Cells were cultured in 150 cm flasks as described in Cell types and culture conditions. Three flasks per gradient were used. After the indicated treatments, cultures were put on ice and washed twice with ice-cold PBS. Cells were scraped in 1.2 ml of ice-cold isotonic buffer/flask (20 mM Hepes-KOH, pH 7.5, 0.25 M sucrose, 90 mM KO-acetate, 2 mM Mg-acetate, 0.5 mM Na-vanadate, 1 mM NaF, 10 mM pyrophosphate, 3 mM β-glycerophasphate, 1 mM pefabloc, 40 U/ml aprotinin, 10 μg/ml leupeptin, and 10 μg/ml pepstatin) and homogenized with a Dounce homogenizer. Nuclei and residual intact cells were pelleted by centrifugation at 1,400 for 5 min at 4°C. The supernatants were separated in three equal aliquots and mixed with iodixanol (OptiPrep; Axis-Shield) and homogenization buffer to generate 30, 20, and 10% iodixanol solutions. They were then loaded into 11.2-ml OptiSeal tubes (Beckman Coulter) and ultracentrifuged at 353,000 for 3 h in a rotor (VT 65.1; Beckman Coulter). 600-μl fractions were collected from the top of the gradient, and protein concentration (bicinchoninic acid reagent; Pierce Chemical Co.) and density (OD at 244 nm as indicated by the OptiPrep manufacturer) were determined. The fractions were then boiled in the presence of reducing laemmli sample buffer. Samples of each fraction containing the same amount of protein for the different cell types to be compared and representative of the total protein content of each fraction were analyzed by SDS-PAGE followed by Western blotting. Fig. S1 shows that PV-1 colocalizes with caveolin in VEC-null and -positive cells. Fig. S2 shows that cell confluence modulates VEGF-induced PY1175–VEGFR-2–positive compartments in HUVECs. Fig. S3 shows that clathrin siRNA inhibits the formation of clathrin-positive vesicular compartments. Fig. S4 shows that the cytoplasmic domain of VEC is required to modulate VEGFR-2 internalization from the plasma membrane. Fig. S5 shows that the antibody Avas12α1 does not modify either the basal or VEGF-stimulated tyrosine phosphorylation of VEGFR-2 and does not induce VEGFR-2 internalization. Online supplemental material is available at .

While studying physics, I became more and more fascinated by the tight interchange between theory, technology development, and experiment. But one of the things that I found frustrating was the scale of many of the experiments. They were so huge that they didn't have the sort of personal feel that experiments in cell or developmental biology have. So I was looking for ways to apply the same sort of approaches to things that are much more personal, and where the time between thinking of an experiment and getting a result was much shorter. So I did a biophysics Ph.D. I was attracted to biophysics because I saw it as a place where microscopy and instrumentation could have an impact. And then I made the mistake of looking at embryos, and it was a slippery slope, I was just pulled down into working on the embryo. Well, “down” is probably not the right word [laughs]. It was a little of both. To me, the cell is the quantum, and microscopy's the way to get to that. So it's very much the microscopy driving my interest in embryology and the embryology driving my interest in microscopy. It's a whole loop. Being at Irvine with a really strong community in developmental biology, and also a very strong community in biophysics, was an ideal incubator space for me as an assistant professor. There were people like Hans Bode, who was working on hydra, and Peter Bryant working on the imaginal wing disk. They had been defining formal rules and making predictions about how cells should interact, but it was hard to go from how they should interact to how they actually do interact. That's where the imaging tied in nicely; it allowed us to ask questions about cell interactions in the intact embryo. People like Gunther Stent and his colleagues at Berkeley had been using vital dyes to trace lineages in the leech. And similar work was going on in David Bentley's lab and Corey Goodman's lab on the patterning of the insect nervous system. In these systems, the cells are large and somewhat convenient for applying vital dyes. What we decided to do is to take those technologies and apply them to cells that are much more challenging. We chose inconvenient systems like the vertebrate nervous system or imaginal disk cells, or things of that sort. The cells of the imaginal disk, for example, are orders of magnitude smaller than the cells that had been studied in the leech. Well, it's really funny, in fact it was a riot trying to convince the microscope manufacturers to sell us the equipment. We needed a microscope with more room between the objective and the stage. The first microscope we used for this was one that had actually been built for the electronics industry, for inspecting big things like integrated circuits. The manufacturers said, “Oh, no, no, no, that's not for a biologist, that's for the electronics people,” and we had to go back and forth, and finally they agreed to sell it to me. It was sort of odd to have to convince them to let me spend my money. The way we do it is really painfully simple. There's a couple different ways you can window the egg. One of them is to just use a pair of fine scissors to actually cut a porthole right in the top. People had operated on chicken eggs before, like Nicole LeDouarin and my wife, Marianne Bronner-Fraser, who was working on chick neural crest. So we appropriated the technology, and what we added to it was going in and labeling cells, or only one cell, in some cases, so that we'd know that we could follow one lineage. I also collaborated with Andrew Lumsden, Roger Keynes, and Claudio Stern on their systems: on somites, the hindbrain, and the spinal cord. In each case, they provided this huge background knowledge of the snapshots of what might be going on. And we animated it. So we could then follow the individual cells, ask questions about what they do and how those snapshots relate to the real cell behaviors that are going on. I wanted to be able to look inside of a frog embryo—all the important interactions are happening deep down inside. The problem was that, with something the size of an embryo, you're completely blind to everything below the outer few tens of microns, if you're looking at it with a light microscope. If you want to see deep into an embryo, that's just like a neurologist wanting to look down into somebody's head, or like an orthopedist wanting to look into somebody's knee, so MRI jumps out at you. The problem is that, if you look at a frog embryo with normal MRI, you'd get a picture with one bright dot, because the resolution of the clinical MRI is almost exactly the same as the size of a frog embryo. That's when we started thinking, “Would it be possible? Could you push the resolution to the microscopic scale?” I should say that all of this MRI started because of a very, very good colleague at UC Irvine, Russell Jacobs, and because we shared a coffee pot. Once a day we would bump into each other at the coffee pot, and I would ask a question, and he would tell me it was impossible. And the next time I'd see him, he'd say, “I thought about it. It's actually not impossible,” and I'd say, “Oh, but that method won't work.” After weeks of drinking at the same coffee pot, we eventually got to a point where it was clear we could apply the technology to embryos and do a microscopic MRI instead of a macroscopic MRI. The conventional wisdom at the time was that as you made the resolution smaller and smaller, the time it takes to get the image goes up astronomically. The back of the envelope calculation tells you that—to go from clinical instrument down to something that would let you see single cells, it would require about 10 times as long for the image. So, say if I loaded you into the MRI, it would take a few minutes to take a nice image of your brain. Well, 10 minutes is a really long time. It would be blurred by plate tectonics, that sort of time scale! What Russ came up with in this whole dialogue was a way to go to higher resolution by using a much stronger magnet. Looking back on them, the first images we got were horrible, but in other ways they were spectacular because we were seeing inside of the embryo. We've been pushing hard over the years to get down from the millimeter scale to the micron scale, and we just now have a couple of papers coming out with some of these recent images of the frog embryo. We can literally watch lineages as they divide and watch cells as they interact—it's letting us address all the things we were blind to before. Yes, Russ and I moved together to Caltech, and the first thing we did when we set up our labs here was to buy a very good espresso machine, restaurant-quality, and put it in the conference room. I would say at least three dozen patents have come out of that coffee pot! Yeah, a very, very good investment! Recently we've been using labeling agents that we can see in both the MRI as well as the light microscope to play those technologies off against one another. Yes, so then we can validate the results between the two and jump orders of magnitude between the two. In fact, one of the people in the lab right now, Mike Tyszka, is trying to make a stage microscope to look at embryos by MRI. Normally, in MRI, you mount the embryos inside something that's not convenient for light microscopy. What Mike's trying to do is make something that, from above, we'll be able to look at embryos in a Petri dish with a light microscope, and from below we'll be able to look at them with an MRI attachment. We're also now trying to apply our imaging tools to disease models and to clinical medicine. For example, we'd love to look in and ask things about a tumor, about the way that the cells respond to chemotherapy, or other therapies, not by waiting until the cells die or the lump grows or the lump gets smaller, but to really ask, are you hitting that cell? Are you inflicting damage on that cell? Are you hitting the cancer stem cells as well as the non–stem cells? Similarly, we're now trying to now make microscopes that can look into somebody's eye and image the earliest events in macular degeneration. You want to be able to catch the process before somebody's lost their vision. So, we think that these imaging tools could allow us to understand the disease progression and to identify the earliest antecedents that would provide an alarm sign. Then, we could do proactive therapy instead of trying to save somebody's eyes once they've already lost a good bit of their vision. There's a variety of other people in the lab trying to make similar leaps to clinical applications, as translation is something we really believe in.

In the early 1980s, methodology was established to allow for culture of pluripotent embryonic stem cells (ESCs) isolated from the inner cell mass of mouse blastocysts (for review see ). Gene targeting in ESCs was soon thereafter achieved through homologous recombination, followed by generation of mouse strains from such manipulated ESCs. Occasionally, gene targeting results in early embryonic lethality, which precludes understanding of the contribution of genes to subsequent developmental processes including de novo blood vessel formation (vasculogenesis) and formation of new blood vessels from preexisting vessels (angiogenesis). Blood vessels are essential for the delivery of nutrients and oxygen to tissues, as well as for removal of waste products. All blood vessels share a number of basic features, although the detailed gene expression pattern, morphology, and function vary between different vascular beds (e.g., arteries, veins, and capillaries). The inside of blood vessels is lined with endothelium, a thin layer of endothelial cells (ECs), which separates the blood from tissues. The outside of the endothelium is covered with a specialized layer of connective tissue (the basement membrane) followed by a layer of mural cells (pericytes and vascular smooth muscle cells). Angiogenesis is a tightly controlled process where EC proliferation and migration is regulated by secreted factors as well as by surrounding cells and matrix. There are currently considerable efforts invested into the development of drugs aimed to control blood vessel growth in conditions such as ischemia and cancer, characterized by deficient or excessive vessel growth, respectively. To this end, it is essential to create easily accessible models by which vessel development can be both manipulated and studied at high resolution. The isolation of EC lines and the establishment of conditions required for their maintenance in cell culture represents a milestone in the vascular biology field (). However, such cultures do not provide a proper microenvironment, involving three-dimensional (3D) interactions between ECs and adjacent supporting cells and matrix that are known to be absolutely vital in regulation of vascular processes. In contrast, cultures of human and murine ESCs possess the capacity to differentiate into most if not all major cell lineages (), creating an environment with parallel development of several cell types. Thus, in differentiating ESCs assembled into embryoid bodies (EBs), vascular development occurs in a context of continuous interactions with adjacent non-ECs. The first indication that EC development and subsequent vascular morphogenesis in differentiating ESC cultures proceed in an in vivo–like fashion was provided by . Formation of EBs can be controlled through aggregation of ESCs in hanging drops (), after the removal of feeder cells and leukemia inhibitory factor that otherwise are used to keep the ESCs pluripotent. The hanging drop culture proceeds for a few days to allow EB growth and differentiation, followed by seeding into a two-dimensional (2D) culture (), or into a 3D collagen gel (). For more information on EB culture procedures, see (detailed protocols will be made available upon request to the authors). At d 3 of differentiation, the onset of vasculogenesis is demonstrated by the presence of a precursor common for endothelial and hematopoietic cells, the hemangioblast. The hemangioblast, which expresses T cell acute leukemia 1/stem cell leukemia (TAL/SCL), vascular endothelial growth factor receptor (VEGFR)-2, and brachyury, has also been detected in human EBs (; ). Subsequently, hemangioblasts will be committed to either the hematopoietic or the EC lineage. The EC precursors, the angioblasts, undergo sequential maturation to eventually express a set of markers characteristic for mature ECs such as VEGFR-2, CD31, vascular endothelial (VE)– cadherin, Tie-1, and Tie-2. Angioblast development in EBs thus closely mimics the in vivo maturation process (). The primary vascular plexus in the EB is remodeled from d 6 and onwards, by sprouting angiogenesis. This process is regulated by growth factors, which may be produced endogenously or added in as exogenous factors. Also without growth factor treatment but in the presence of 15% serum, vascular development is evident by the presence of blood islands that differentiate to form small networks of ECs located in the center of the EB. Addition of growth factors stimulates further expansion of the endothelium. There is a distinct morphology of the vascular plexus formed in 2D EB cultures dependent on the growth factor present in the culture; VEGF isoforms VEGF-A121 and VEGF-A165, fibroblast growth factor-2, and platelet-derived growth factor (PDGF)-BB each enhance vessel formation in a distinct pattern. Typically, VEGF-A165 stimulates the formation of a peripheral capillary plexus in 2D EB cultures () (). Invasive angiogenesis in 3D collagen gels is preferentially induced by VEGF-A165 and manifested around d 8 by the formation of EC sprouts protruding from the central core of the EB. The stalk cells are guided by tip cells with numerous filopodia, a process with striking similarities to vascular development in zebra fish and the retina (; and ) (; ). Subsequently, the sprouts branch and occasional tip cells fuse with adjacent vessels to form networks. The EC sprouts are surrounded by perivascular cells that share features such as morphology (i.e., close apposition to the endothelial cells) and protein expression pattern (expressing nerve-glia2 [NG2] and/or α-smooth muscle actin [αSMA]) with pericytes seen in vivo (). Furthermore, the vessels are enclosed by a vascular basement membrane whose detailed composition, dynamics, and function in the EBs remain to be described. Lumen formation is detectable at about d 10 of EB differentiation, and occasionally a mature lumen is evident at d 12 (). Subsequently, large lumenized vascular networks become established. It is an interesting possibility that the EB endothelium has the capacity to undergo arterial/venous specification, as endothelial cells formed from ESCs in vitro specifically express either ephrin B2 or EphB4, which are markers for arterial and venous endothelium, respectively (). Inactivation of genes with a vascular function (, , , or ) results in similar phenotypes in the EB model as in vivo, with regard to temporal effects on development and consequences for EC morphology () (). For example, deletion of VEGF-A, one of the main VEGFR-2 ligands, results in an arrest in vascular development and remodeling in vivo as well as in vitro (). The EB model is particularly suitable in this context because it allows rapid and easy testing of the unique contribution of different VEGF-A isoforms to vascular development. Accordingly, treatment of VEGF-A–deficient EBs with purified VEGF-A165 rescued EC morphogenesis (), in agreement with the fact that mice expressing only VEGF-A165 display normal vascular development. In certain cases, data generated using in vivo models have been extended by studies performed in different ESC-based culture models. ESCs to be required for vascular morphogenesis but not essential for early endothelial and hematopoietic cell commitment (). Furthermore, a number of gene deletions have resulted in developmental arrest before the onset of vasculogenesis, for example due to defects in implantation. However, by generation of ESCs from recombinant blastocysts, critical stages incompatible with in vivo growth may be studied in the EB model (see for a comparison of vascular phenotypes in gene-targeted embryos and EBs). A severe developmental phenotype (lethal before gastrulation) is caused by simultaneous deletion of the enzymes -deacetylase/-sulfotransferase 1 and 2 (NDST1/2), central to the synthesis of heparan sulfate (HS) (). Proteins modified by attachment of HS, so-called proteoglycans, are essential co-receptors for many tyrosine kinase receptors, including VEGFR-2. Deletion of the NDST1/2 enzymes severely hampers ESC differentiation with a close to complete loss of vascular development (). and ESCs (). stem cells (expressing VEGFR-2) were complemented by normal HS produced by pericytes derived form the ESCs (). This exemplifies the versatility of the EB model, which allows combinations of knock-out ESCs to study the requirement for genes in subpopulations of cells during cell specification and development, to unravel new mechanisms in cell communication. #text As described above, the flexibility of the EB model is remarkable due to the fact that ESCs can be derived from all types of genetically engineered mice. Thus, ESC-based models may in the future be used widely to complement and occasionally even replace animal experimentation, especially when very early embryonic lethality becomes a severe limitation (see ). ESCs from transgenic mice and gene-targeted mutant mice can be evaluated independently, or in combination, by generating chimeric cultures of two or more different ESC types (). Inducible gene expression and deletion systems, such as the Tetracyclin-On and -Off expression systems, and the site-specific DNA recombinase system Cre/loxP, in combination with labeling of cells by expression of reporter genes (e.g., β-galactosidase or fluorescent proteins) will allow clonal analysis and lineage tracing similar to what can be done in animal models (; ). Several ESCs with fluorescent reporters under the control of EC-specific promoters have already been created, constituting powerful tools for live imaging at single-cell or even subcellular resolution by confocal or two-photon excitation microscopy (). Generation of homozygous gene deletions and recombinant animals is expensive, time consuming, and laborious. Gene silencing can instead be achieved by RNA-interference, where RNAi can be delivered to both murine and human ESCs through lentiviral transduction (). Because RNAi delivery can be traced and enriched through selection for reporter gene expression, a very high degree of silencing can be attained. Moreover, previously reported problems with clonal selection and loss of expression during differentiation of ES cells appear to be circumvented when using lentivirus as a strategy for introduction of RNAi (see for further discussion). The fact that stem cells under proper conditions have the capacity to differentiate into both blood and lymphatic ECs may be explored for therapeutic purposes (). Importantly, both mouse and human ESCs can be used to generate functional ECs contributing to formation of stable vessels that connect to the host circulation (; ). The use of human ESCs for therapeutic purposes obviously presents a moral dilemma. Possibly, retrieval of ESCs from other locations than the fetus, such as umbilical cord blood or amniotic sources may present a feasible alternative in the future (). Furthermore, it is an interesting possibility that pathological conditions characterized by impaired blood and lymph vessel function may be treated by administration of adult stem cells or progenitors isolated from the patient's own bone marrow. For a recent review on the contribution of circulating stem cells to angiogenesis, see . Despite striking similarities between mouse and human development, numerous therapies developed in mice (e.g., to treat diseases such as cancer) have failed when tested in humans. A contributing factor to such failures may be genetic differences between the species. Because most experimentation on humans is prohibited for ethical reasons, preclinical testing has relied solely on animal experimentation. In the future, application of human ESCs may constitute an additional step in the development and testing of drugs, with regard to toxicity and teratogenic effects. Furthermore, experimentation with human ESCs offers means to study human embryonic development. Already, homologous recombination and introduction of RNAi have been demonstrated in human ESCs, paving the way for new insights in human biology ().

Centromeres are specialized chromosome domains that are responsible for chromosome segregation during meiosis and mitosis. They assemble around repetitive DNA sequences in a complex structure that has yet to be fully elucidated. A simplistic view involves the division of this domain into two areas: the central core region or centromeric chromatin () and the flanking heterochromatic regions, which are called pericentromeres. The protein composition of the central core region varies between interphase and mitosis. In this model, constitutive proteins are permanently associated with the centromere even during interphase, whereas facultative proteins are recruited only during mitosis to assemble the kinetochore, which is the site of microtubule attachment. As such, the central core region serves as the assembly platform for the kinetochore. A specific feature of the chromatin structure of the core centromere is that it contains interspersed blocks of nucleosomes that contain histone H3 and a histone H3 variant called centromeric protein (CENP) A in human cells (). In addition to histones, six constitutive proteins named CENP-A, -B, -C, -H, -I, and hMis12 are known as the major components of the interphase centromeric chromatin. However, another set of 11 proteins associated with the CENP-A–containing nucleosomes or with the CENP-H–I complex has recently been described (; ). Cajal bodies (CBs) are nuclear domains that were discovered in 1903 by the Spanish physiologist Santiago Ramón y Cajal (). These bodies are concentrates of several proteins and small nuclear ribonucleoproteins (). Among these proteins, coilin was described in the early 1990s as the major component of CBs (), although its precise biological activity remains elusive. Orthologues of human coilin are known in many vertebrates, including the mouse (), (), (), (), and (Liu, J.L., and J.G. Gall, personal communication). Coilin is not strictly essential for mouse embryonic development, although a substantial reduction of viability has been observed in inbred homozygous embryos (). Coilin contains nuclear and nucleolar localization domains, an arginine-glycine (RG)–rich box, and an autointeraction domain that facilitates CB formation (). The formation of CBs depends, at least in part, on the autointeraction domain and on posttranslational modifications of coilin. Indeed, hyperphosphorylation considerably reduces the coilin autointeraction, which leads to CB disassembly during mitosis (; ). The biological function of coilin within CBs remains mysterious, and its additional diffuse staining in the nucleoplasm has been proposed to be the mark of still unrevealed CB-independent activity (). Herpes simplex virus type 1 (HSV-1) infection of cultured cells induces the destabilization of centromeres during interphase, preventing the assembly of the kinetochore and the binding of microtubules during mitosis (). The factor responsible for this centromere destabilization is the viral protein infected cell protein 0 (ICP0). ICP0 is a RING finger nuclear protein with characterized E3 ubiquitin ligase activity (for review see ). As soon as it enters the nucleus, ICP0 temporarily localizes to centromeres and induces the proteasomal degradation of CENP-A, -B, and -C (; ; ). Thus, ICP0-induced degradation of essential constitutive CENPs during interphase is likely to modify the structure of the central core region extensively, thereby preventing the assembly of the kinetochore. As a consequence, cells that express ICP0 just before entering mitosis are stalled in early mitosis and eventually suffer premature cell division without chromosomal segregation, leading to aneuploidy (). Although the biological significance of ICP0-induced centromere destabilization is unclear, from the cellular viewpoint, ICP0 is of exceptional interest as a unique tool for studying centromere structure and the putative cell mechanisms implicated in centromere architectural maintenance. Indeed, although it is known that the cell uses kinetochore surveillance mechanisms during mitosis, such as the mitotic checkpoints (for review see ), it remains unknown whether the cell is able to detect centromeric structural defects in interphase. In this study, using ICP0 as a tool, we reveal a previously unreported and unexpected cell response to human and mouse centromeres that have sustained structural modifications during interphase. This response is characterized by the accumulation at damaged centromeres of two CB proteins, coilin and fibrillarin, and one CB-associated gemini of CB (gems) nuclear domain protein, the survival motor neuron (SMN) protein. We show that this response does not implicate the entire CBs and/or gems. In addition, we confirm the physical association between coilin and centromere higher order type I α-satellite (α-SAT) DNA. Using siRNA against CENPs, we demonstrate that depletion of CENP-B leads to the accumulation of coilin at centromeres. This confirms the existence of a cell response that is triggered by interphase centromere instability, for which we propose the term interphase centromere damage response (iCDR). After our work on the ICP0-induced degradation of CENPs and the resulting destabilization of centromere structure, we decided to check whether and how cells were able to respond to centromere instability during interphase. Interphase HeLa cells were infected with HSV-1 wild-type (wt) virus and analyzed by immunostaining at 2 and 4 h after infection. As expected from our previous work, ICP0 transiently colocalized with centromeres at 2 h after infection but not at 4 h after infection (i.e., after degradation of the CENPs; compare ICP0 patterns in ; insets; ). At 4 h after infection (), ICP0 accumulated in large, visible foci inside the nucleoli as described previously (). We found that at early stages of infection, the nuclear pattern of coilin underwent profound changes. shows noninfected cells in which coilin was localized in CBs. Infection by HSV-1 wt for 2 h did not substantially affect the coilin distribution (), whereas at 4 h after infection, coilin adopted a multidotted pattern in >90% of the infected cells (). These coilin foci were more abundant and smaller than those that corresponded to CBs (0.2–0.5 μm for coilin foci vs. 1–1.2 μm for CBs in noninfected cells; compare coilin patterns in ). Upon close examination, it was noticed that the nuclear distribution of these coilin dots looked very much like the pattern of immunostained centromeres. Therefore, we stained the latter together with coilin in ICP0-expressing cells; strikingly, most of the coilin foci indeed colocalized with centromeres (, yellow foci in the merged and magnified images). As the centromeric localizations of ICP0 and coilin were temporally separated (2 vs. 4 h after infection), the accumulation of coilin at centromeres is unlikely to be a consequence of an interaction with ICP0. Collectively, these data suggested that coilin accumulates at centromeres as a consequence of ICP0 activity on centromeres. To verify this notion, cells were infected with either the vFXE virus, which expresses the nonfunctional ICP0 RING finger mutant FXE, or the ICP0-null mutant dl1403, whose infection is detectable by ICP4 viral protein staining. We found that infection with these viruses did not induce coilin redistribution (). ICP0 has E3 ubiquitin ligase activity that is associated with its RING finger domain. This activity is responsible for the induction of degradation via the proteasome of CENPs, which can lead to the destabilization of interphase centromeres and to defects in kinetochore assembly (). To verify the implication of proteasome activity in ICP0-induced coilin accumulation at centromeres, cells were infected with the HSV-1 wt virus in the presence of the proteasomal inhibitor MG132 or DMSO alone (unpublished data). In the presence of MG132, HSV-1 wt no longer induced the accumulation of coilin at centromeres (). These data suggest that there is a strong correlation between ICP0- and proteasomal-induced protein degradation (and probably CENP degradation) and the accumulation of coilin at centromeres. Finally, to exclude any effect of the infection, we verified that transfected cells that expressed ICP0 () but not those that expressed FXE (not depicted) showed a centromeric accumulation of coilin similar to HSV-1 wt–infected cells. As expected, 100% of the ICP0-expressing cells showed centromeric coilin. Note that during the course of these experiments, coilin colocalization with centromeres was never seen in mitotic cells. In conclusion, the aforementioned results demonstrate the existence of a cell response that is triggered by the ICP0-induced structural damage of interphase centromeres (i.e., the iCDR) and that coilin is implicated in the iCDR. These observations strongly suggest a role for coilin in a mechanism that is dedicated to the detection and/or repair of unstable interphase centromeric structures. To determine whether coilin physically interacts with the central core region, we performed chromatin immunoprecipitation (ChIP) assays with HSV-1 wt–infected cells () and looked for the association of coilin with centromeric, chromatin-specific higher order type I α-SAT DNA. In each case, we compared the ChIP of noninfected cells with that of cells at 4 h after infection (i.e., the time point at which coilin was found to have accumulated at centromeres; ). The method and results were validated using an anti–CENP-A antibody as a control. Indeed, as expected, we observed a major decrease in the amount of CENP-A associated with type I α-SAT DNA as a consequence of ICP0-induced CENP-A degradation. We then tested an anticoilin antibody in parallel with the anti–CENP-A and anti-myc antibodies, with the latter being used as a nonspecific control (). In each experiment, a single batch of chromatin was split into three aliquots, one for each antibody, and the three ChIPs were performed simultaneously. The results obtained (, left) show that more coilin is associated with α-SAT DNA at 4 h after infection compared with noninfected cells, whereas less CENP-A is present at 4 h after infection. In these experiments, the amount of α-SAT DNA associated with the anti-myc control antibody did not notably change. Considering that the 4-h postinfection CENP-A was undetectable at centromeres by immunofluorescence (IF) and almost completely degraded, as shown by Western blotting (WB; ), a twofold decrease in the amount of α-SAT DNA retrieved (P < 0.05) with CENP-A can be regarded as representative of a major effect. In this context, the 1.7-fold increase in α-SAT DNA retrieved with coilin (P < 0.05) is suggestive of a significant increase in coilin levels at centromeres, which fits with our aforementioned results (). The DNA of the housekeeping gene () was measured to control for binding of the anticoilin and anti–CENP-A antibodies to unrelated DNA, and, as expected, no enrichment of coilin was noted (, right). From these data, we conclude that ICP0-induced centromere destabilization results in a physical interaction between coilin and centromeric DNA. A major issue in the biology of nuclear domains is whether the entire domain is involved in a process or only some components thereof. Thus, there was a need to determine whether components of CBs other than coilin also accumulate at damaged centromeres and, if so, whether whole CB domains are associated with centromeres in ICP0-expressing cells. In addition to coilin, CBs concentrate several small noncoding RNAs as well as an array of proteins, all of which are more or less implicated in transcription (). Apart from fibrillarin, no other tested CB-associated protein exhibits centromeric accumulation in ICP0-expressing cells (, i and ii; ; and ). Fibrillarin is one of the most abundant proteins in the fibrillar regions of the nucleolus. It is conserved from yeast to humans and is essential for early development in the mouse, being a catalyst of preribosomal RNA methylation (; ). Because CBs also contain RNAs, immuno-RNA FISH assays were performed to analyze the pattern of the major CB-associated U2 small nuclear RNA (snRNA) in ICP0-expressing cells. No centromeric accumulation of U2 was detected (, i and ii). In addition, antibodies raised against both the 5′-terminal caps of the snRNAs and the snRNA-binding Sm proteins did not show any centromeric signals in ICP0-expressing cells (; and ). Collectively, these data indicate that only some components of the CBs, such as coilin and fibrillarin, accumulate at damaged centromeres. In the cell nucleus, gems were originally defined as CB-associated domains on the basis of IF staining with antibodies against the SMN protein (). This protein is the product of the gene, whose loss of function mutations are responsible for the severe inherited disorder spinal muscular atrophy, one of the major genetic causes of infant mortality (). We decided to check the patterns of gem- associated proteins in ICP0-expressing cells. Three gem proteins, SMN, gemin 2, and gemin 3, were tested and showed foci that colocalized with CBs in control cells (, iii; arrows for SMN; and ; gemin 2 and 3). In ICP0-expressing cells, SMN but not gemin 2 or 3 was detected in numerous small foci that, similar to coilin and fibrillarin, colocalized with centromeres (, iv; arrowheads; ; and ). As was the case for coilin, 100% of the ICP0-expressing cells showed centromeric fibrillarin and SMN. The aforementioned experiments suggest that neither coilin, fibrillarin, nor SMN is degraded in ICP0-expressing cells. To confirm this point, cells were infected with HSV-1 wt in the presence or absence of the proteasome inhibitor MG132, vFXE, or dl1403 viruses at a multiplicity of infection of 10 (100% of the cells infected; ). Coilin, fibrillarin, and SMN did not sustain ICP0-induced degradation. We observed the iCDR in additional human cell lines of various origins (e.g., breast carcinoma [T47D], colon carcinoma [SW480, HCT116, and HT29], and skin cancer [HaCaT]) as well as in primary keratinocyte cells. To determine whether this response is conserved in other species, the centromeric accumulations of coilin, fibrillarin, and SMN were analyzed in mouse NIH3T3 cells, which express ICP0. In mouse nuclei, chromosomes are distributed in clusters, and pericentromeric heterochromatin forms chromocenters that can be visualized by DAPI staining. The DNA sequences of the central core and pericentromere regions are based on two different types of DNA repeat, called minor and major satellites, respectively. These two domains are spatially separated and are clearly distinguishable by in situ hybridization (; ; ). We performed immuno-DNA FISH assays on ICP0- expressing cells to detect ICP0 (), minor satellite DNA, and coilin or fibrillarin (the anti-SMN mAb does not work on mouse cells). In the absence of ICP0, coilin and fibrillarin showed the same patterns in the mouse as in human cells (i.e., present in CBs and/or nucleoli; ). In ICP0-expressing cells, coilin (, i–iii) and fibrillarin (, i–iii) showed a multidotted pattern that was similar to the pattern in human cells. These dots clearly colocalized with the minor satellite signals and, thus, with the central core region of the centromere. These data show that the iCDR is conserved in mammalian cells, at least between humans and mice, and is not an artifact of a single cell line. A recent study has described the effect of UV-C irradiation on CB fragmentation (). Consequently, coilin showed a change of pattern and increased interaction with PA28γ (proteasome activator subunit γ), a protein that is implicated in some aspects of proteasome activity in vitro (). This interaction resulted in the colocalization of coilin and PA28γ in UV-C–treated cells. Because UV light irradiation induces DNA breaks, the putative participation of coilin in a mechanism that is designed to resolve such damage to DNA has to be considered. Therefore, we investigated whether the response to ICP0-induced damaged centromeres implicated protein complexes involved in DNA break repair. PA28γ did not colocalize with coilin at the centromeres of ICP0-expressing cells ( and ). Likewise, several proteins involved in single- or double-strand DNA break repair pathways (including nucleotide excision, base excision, and double-strand break repair) did not relocalize at the damaged centromeres in ICP0-expressing cells (; and ). Given that irradiation of HeLa cells with γ rays (from 2 to 10 Gy) does not provoke CB fragmentation (unpublished data) and that ICP0 is not known to provoke DNA breaks, these data do not support the accumulation of centromeric coilin, fibrillarin, and SMN as part of a putative DNA damage response–associated mechanism with activity in resolving DNA breaks. The aforementioned results demonstrate that ICP0 provokes the accumulation of at least three proteins at damaged centromeres. However, one could argue that the accumulations of coilin, fibrillarin, and SMN at centromeres are caused by an as of yet unknown activity of ICP0 and are not a direct result of CENP degradation. Therefore, we induced centromere destabilization independently of ICP0 using siRNAs that target the mRNAs of CENP-A, -B, and -C, the three known CENPs that are degraded in an ICP0-dependent manner. We verified by IF () and WB () the effects of the siRNAs on the stability of the targeted proteins. HeLa cells were transfected with single-type siRNAs or a mixture of two or three siRNAs, and the cells were then immunostained to detect decreases in the amounts of the targeted proteins from centromeres and centromeric accumulations of coilin, fibrillarin, and SMN. Cells with multidotted coilin, fibrillarin, or SMN, which were representative of the accumulations of these proteins at centromeres, were counted in several experiments that were performed independently. A scrambled sequence siRNA never had any effect on the coilin, fibrillarin, or SMN patterns. Similarly, none of the three CENP siRNAs was able to induce multidotted fibrillarin or SMN in a manner similar to ICP0 even when used in combination (unpublished data). Interestingly, coilin showed a multidotted pattern in a small proportion (∼5%) of cells that were treated with the CENP-A or -C siRNA and in a large proportion (∼30%) of cells that were treated with the CENP-B siRNA (). We confirmed by IF that the multidotted coilin colocalized with centromeres (). To show a more representative view of the colocalization, the merged images were processed, as in (c and g) with the colocalization module of the LSM 510 software, so that all of the colocalized pixels appeared black on a white background (, right column). CENP-A colocalization with huACA staining was used as a positive control (, i). A large proportion of coilin colocalized with centromeres in ICP0-expressing cells (, iii) as well as in CENP-B siRNA-treated cells (, iv). Much fewer black spots were visible in the few CENP-A (or CENP-C; not depicted) siRNA-treated cells that showed multidotted coilin (, v), and almost no black spots were seen in the scrambled sequence siRNA-treated cells (, ii). In addition, the proportion of centromeres that colocalized with coilin was determined (, right column; numbers in parentheses). Several images were scanned in each experiment () to determine a mean colocalization coefficient between centromeres and coilin. The coefficient for CENP-A colocalization with centromeres was arbitrarily fixed at 1. The colocalization coefficient in untreated cells or cells transfected with the scrambled sequence siRNA was very low, with a value of 0.046. Cells that were treated with the CENP-A (or CENP-C) siRNA showed a fivefold increase in the colocalization coefficient (0.28). ICP0-expressing cells and CENP-B siRNA-treated cells gave the highest coefficient (0.45) for coilin/centromere colocalization, with a 10-fold increase compared with the normal situation. The specificity of the accumulation of coilin at centromeres that lacked CENP-B and, to a lesser extent, CENP-A and -C renders an off-target effect of the siRNAs highly unlikely. However, we tested another siRNA against CENP-B and obtained similar results (unpublished data). These results confirm that the iCDR is triggered as a direct consequence of the loss of some constitutive CENPs and, thus, by the instability of the centromere structure. In this study, we took advantage of the effect of the ICP0 protein on the destabilization of centromere structures to investigate whether cells survive and adapt to such dramatic modifications at domains of major importance for their viability. We reveal a novel cellular response, named iCDR, suggesting the existence of mechanisms dedicated to dealing with structural damage to centromeres during interphase (i.e., before the onset of mitosis). We also describe a role for coilin in the iCDR, which implicates two other proteins, fibrillarin and SMN. In addition, the fact that several human cell lines and mouse cells manifest a similar response suggests a general mechanism that is likely to be conserved, at least in mammals. Importantly, our data from cells knocked down for CENP proteins confirm the existence of a cellular response that is triggered directly by structural modifications to centromeres. Although efficient for coilin, the single and combined siRNAs were, unlike ICP0, ineffective in stimulating fibrillarin and SMN. Therefore, the centromeric accumulations of fibrillarin and SMN are probably induced by more severe damage than that arising from the absence of CENP-A, -B, or -C or by the absence of other CENPs. Interphase centromeres are complex structures organized into multisubunit protein domains that are associated partly with the CENP-H–I complex and partly with the centromere-specific CENP-A–containing nucleosomes in the distal (CENP-A nucleosome distal) and proximal (CENP-A nucleosome- associated complex) layers (; ). In light of our results, these data are informative in two respects. First, to engineer the collapse of the entire centromeric structure using CENP-directed siRNAs, it is necessary to affect more than one protein. Second, it is anticipated that ICP0 induces the proteasomal degradation of more CENPs than the individual CENP-A, -B, and -C proteins. Therefore, a simple explanation for the differential centromeric accumulations of coilin, fibrillarin, and SMN seen in this study may be the levels of damage caused to the different layers of CENPs. Coilin accumulation at damaged centromeres is induced by CENP-B depletion, which suggests that the absence of CENP-B, unlike the absence of CENP-A and -C, is sufficient to trigger the response. CENP-B is a DNA-binding protein that has been implicated by in vitro studies in the positioning of nucleosomes (; ). It is unclear whether there is an absolute requirement for CENP-B for the preservation of centromere structure and function, as knockout mice for CENP-B are viable (; ; ) and CENP-B is essential for the de novo formation of centromeres, control of the epigenetic state of centromeric chromatin, and assembly of CENP-A (Masumoto, H., personal communication; ). mice studies (; ; ). The fact is that in the particular context of mouse cells, centromeres are stable and functional even if the gene is missing. mice, the centromere structures could have acquired a CENP-B–independent equilibrium that is epigenetically transmissible. This situation is not quite the same as the one described in our present study, in which we hypothesize the rapid disruption of the equilibrium of the centromere structure by the rapid degradation of CENPs. In this regard, we investigated the coilin and fibrillarin distributions in untreated (not expressing ICP0 and not transfected with CENP siRNAs) mouse embryonic fibroblast cells derived from CENP-B knockout mice and did not detect any accumulations of these proteins at centromeres (unpublished data). We know that the accumulation of coilin and fibrillarin at damaged centromeres occurs in ICP0-expressing mouse NIH3T3 cells (this study) and normal mouse embryonic fibroblast cells (unpublished data). mouse embryonic fibroblast cells suggests that the centromeric coilin, fibrillarin, and SMN proteins do not act as part of a putative structural complex that replaces the missing proteins but rather as a response of the cell to dramatic modifications of the centromere structure at a given time point. Centromeres possess a very specific organization of their chromatin compared with noncentromeric chromatin, and, thus, they are likely to concentrate unique cellular processes. Our data showing that ICP0-expressing cells lack centromeric accumulations of proteins implicated in DNA break repair mechanisms do not favor the hypothesis of DNA lesions as a direct consequence of the accumulations of centromeric coilin, fibrillarin, and SMN. Therefore, it is likely that the modification of centromere structure itself is responsible for triggering the iCDR. Interestingly, two out of the three known CENP targets of ICP0, CENP-A and -B, are established chromatin-related proteins. This raises the question as to whether the recruitment of centromeric coilin, fibrillarin, and SMN is initiated by the abnormal protein content of the centromere and/or by the abnormal chromatin structure. On the basis of the data in the literature, it is difficult to come up with a convincing explanation for the interactions of coilin, fibrillarin, and SMN with damaged centromeres. First, this association was unexpected. Second, there is a general lack of information concerning the architecture of the central core centromere, particularly during interphase. However, coilin, fibrillarin, SMN, and centromeres may be linked by RNAs, especially small noncoding RNAs. Indeed, coilin, fibrillarin, and SMN are components of nuclear bodies, CBs, and gems, whose best-characterized functions remain the maturation of small noncoding RNAs that are implicated in the processing of larger transcripts. The association of small RNAs through the RNA interference mechanism with the epigenetic modifications of centromeric heterochromatin is now well documented, particularly in the yeast (for reviews see ; ). Importantly, this RNAi-dependent heterochromatinization is directly linked to transcriptional activity in the pericentromeric heterochromatin. Although the existence of a similar mechanism in higher eukaryotes has not yet been demonstrated, it is clear that centromeric heterochromatin–associated transcriptional activity is conserved at least in humans, maize, rice, and chickens (; ; ; ). Moreover, recent studies have shown that transcription and small RNAs can participate in the architecture and function of centromeres in murine cells (; ). From these data, it is tempting to speculate that the presence of coilin, fibrillarin, and SMN at damaged centromeres reflects the involvement of RNAs in maintaining the centromeric chromatin structure. Future studies should provide new information that is relevant to this hypothesis. At the molecular level, we anticipate a close link between the iCDR and the need to reform a functional centromere structure that has been accidentally damaged during interphase. This response could trigger a mechanism that eventually results in the reformation of a fully functional prekinetochore. This inevitably raises questions as to the consequences of centromere instability for general chromosomal instabilities that result in aneuploidy and cancer development (). Several types of genetic alteration are responsible for chromosomal instabilities, including those that affect kinetochore functions (). Although there is a clear need for correct kinetochore structures to prevent chromosomal instabilities, not much is known concerning the capacities of interphase centromeres to serve as platforms for kinetochore nucleation. If interphase centromeres are unable to build functional kinetochores as a result of structural problems, it is likely that the mitotic spindle checkpoint will be weakened and its activity bypassed, forcing mitosis and provoking aneuploidy (for review see ). This is precisely what has been described in a recent study of late embryonic development in –null mutants (). Therefore, it is reasonable to propose the existence of a cellular response that acts as a sensor mechanism to signal the emergence of structural problems to interphase centromeres. In retrospect, it is not surprising that cells are sensitive to defects at interphase centromeres considering the major roles of these genetic loci. To date, it has been unknown whether cells have developed mechanisms during interphase to check the functionalities of these structures. Our present results clearly show that this is probably the case, and they raise questions as to the existence of pathways that are committed to sensing, signaling, and repairing centromeres before the cell enters mitosis. In this framework, future studies will need to address the importance and functions of proteins, such as coilin, fibrillarin, SMN, and other proteins, in these types of pathways. HeLa and NIH3T3 cell lines were cultivated in BHK-21 and DME, respectively, which were supplemented with 10% FBS, -glutamine (1% vol/vol), 10 U/ml penicillin, and 100 μg/ml streptomycin. The pci110 plasmid, which expresses ICP0, has been described previously (). The wt strain HSV-1 syn+ (17+) is the parental strain (referred to as HSV-1 wt in this study). The dl1403 virus, which was deleted of ICP0 (), the vFXE virus, which expresses a nonfunctional ICP0 isoform mutated in its RING finger domain, and the standard infection procedures have been described previously (). Cells were seeded at 1.5 × 10 cells per well for transfection and at 2.5 × 10 cells per well for infection in 24-well plates that contained round coverslips. 24 h later, the cells were transfected with the pci110 plasmid according to the manufacturer's recommendations (Effectene Transfection Reagent; QIAGEN) or infected. Cells were treated for IF as described previously (). With the exception of , in which a CCD camera (CoolSNAP HQ; Roper Scientific) was used for the analysis (Metaview software; Molecular Devices), all of the samples were examined under a confocal microscope (LSM 510 Meta; Carl Zeiss MicroImaging, Inc.). The data from the channels were collected separately to avoid channel overlap, with fourfold averaging at a resolution of 512 × 512 pixels using optical slices of 0.8–1.0-μm thickness. A microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) was used at either 63× (NA 1.25) or 100× (NA 1.3) magnification by oil-immersion objective lenses (Carl Zeiss MicroImaging, Inc.). Datasets were processed using LSM 510 software (Carl Zeiss MicroImaging, Inc.) and exported in preparation for printing using Photoshop (Adobe). italic xref #text xref #text Cells were seeded at 5 × 10 cells per 100-mm Petri dish. The following day, the cells were infected with the HSV-1 virus for 4 h and were subjected to the ChIP assay (Upstate Biotechnology). The samples were incubated with IgG Fab fragments (US Biological) before adding the test antibodies to decrease the nonspecific binding of alphoid DNA repeats to IgG. DNA that was immunoprecipitated with the different antibodies was quantified by PCR using specific primers. The PCR reaction was performed in a LightCycler (Roche Diagnostics) using 1 μl DNA, 0.5 μM of each primer, and 10 μl SYBR green (QIAGEN) in a final volume of 20 μl. The data were analyzed using the LightCycler software (Roche Diagnostics). The α-SAT and GAPDH designations indicate oligonucleotides that are specific for the centromeric type I α-SAT DNA and the gene that encodes the GAPDH enzyme, respectively. The α-SAT oligonucleotides were designed based on the consensus α-SAT sequence (), which was derived from the alignment of >500 type I alphoid sequences. Moreover, a BLAST search using this consensus type I α-SAT showed hits for cosmids from several chromosomes. Therefore, these oligonucleotides amplify DNA fragments from several different chromosomes, and the ChIP data are likely to reflect the association of proteins with several chromosome centromeres. The primers used were as follows: for α-SAT, forward 5′-AATCTGCAAGTGGATATTT-3′ and reverse 5′-CTACAAAAAGAGTGTTTCAAAAC-3′; for GAPDH, forward 5′-CACGTAGCTCAGGCCTCAAGA-3′ and reverse 5′-AGGCTGCGGGCTCAATTTAT-3′. The data shown are the ratios between the values obtained for the immunoprecipitated chromatin and the input DNA and are the mean values from six independent experiments. HeLa cells were seeded at 4 × 10 cells per well in a 24-well plate that contained round coverslips for IF or at 1 × 10 cells per 60-mm Petri dish for WB. 1 d later, the cells were transfected with 300 ng (for IF) or 7 μg (for WB) siRNA that targets the mRNA of CENP-A, -B, or -C and were incubated for 48 h. A second round of siRNA transfection was performed as described above, and cells were incubated for an additional 48 h before the cells were harvested for IF or WB experiments. For WB, 40-μg aliquots of total protein were loaded per well in an SDS-polyacrylamide gel before electrophoresis, transfer, and detection as described previously (). The siRNA sequences used were as follows: for , 5′-GGUUGGCUAAAGGAGAUCCTT-3′; for , 5′-CUACACCGCCAACUCCAAGTT-3′; and for , 5′-GAAGCCUCUCUACAGUUUGTT-3′. The following antibodies were used at the indicated dilutions for the IF, WB, or ChIP experiments: mAbs anti-p62 (TFIIH at 1:400 for IF), [3–19] anti–CENP-A (1 μg/ml for WB and 2 μg per sample for ChIP; Abcam), [5E6C1] anti–CENP-B (1 μg/ml for WB), anticoilin (1:400 for IF, 1 μg/ml for WB, and 2 μg per sample for ChIP; Sigma-Aldrich), [25–4] anti–DNA-PKcs (1:200 for IF; NeoMarkers); [72B9] antifibrillarin (1:500 for IF), [38F3] antifibrillarin (1:500 for WB; Abcam), anti–gemin 2 (1:200 for IF; Abcam), [12H12] anti–gemin 3 (1:200 for IF; Abcam), anti–HCF-1 (1:400 for IF), anti-ICP0 (1:1,000 for IF and 1:10,000 for WB), [8.F.137B] anti-ICP4 (1:100 for IF; US Biological), [9E10] anti-myc (2 μg per sample for ChIP; Abcam), [D0-7] anti-p53 (1:500 for IF; DakoCytomation), [5E10] anti–promyelocytic leukaemia (PML; 1:50 for IF), [51RAD01] anti-Rad51 (1:200 for IF; NeoMarkers), [9H8] anti-RPA32 (1:200 for IF; Abcam), Y12 anti-Sm (1:500 for IF; NeoMarkers), anti-SMN (1:400 for IF; 1 μg/ml for WB; BD Biosciences), and anti–2,2,7-trimethylguanosine (TMG at 1:300 for IF; Oncogene Research Products). In addition, the following rabbit polyclonal antibodies were used: antiactin (1 μg/ml for WB; Sigma-Aldrich), 1343 anti-BLM (1:500 for IF), anti–CENP-A (1:200 for IF; Upstate Biotechnology), anti–CENP-B (1:2,000 for IF), r554 anti–CENP-C (1:1,000 for IF and WB), rp80 anticoilin (1:400 for IF), anti-FLASH (1:4,000 for IF), anti-γH2AX (phospho-S139 at 1:500 for IF; Abcam), R190 anti-ICP0 (1:200 for IF), anti-PA28γ (N terminus; 1:200 for IF; Zymed Laboratories), and BL647 antiphospho-RPA32 (S4/S8 at 1:1,000 for IF; Bethyl Laboratories). The human autoimmune serum huACA against CENPs was also used (1:3,000 for IF). For IFs, the secondary antibodies used were as follows: goat anti–rabbit, anti–mouse, and anti–human antibodies coupled to AlexaFluor488, -555, or -647 (1:200; Invitrogen).

The Pax gene family defines an evolutionary conserved group of transcription factors that play critical roles during organogenesis and tissue homeostasis (; ). Nine Pax proteins have been described in mammals, where the presence of the paired box DNA binding domain is a common feature. The family is further subgrouped by the presence of an octapeptide motif and the presence, absence, or truncation of a homeodomain region. Pax3 and Pax7 are two closely related family members (; ; ; ; ) that are involved in the specification and maintenance of skeletal muscle progenitors. Genetic analyses in mice showed that Pax3 is critical for delamination and migration of muscle precursors from the somites to the limbs (; ; ). mice have no gross defects in muscle formation. However, in the absence of Pax7, adult skeletal muscles are completely devoid of satellite cells (; ), which are thought to represent the stem cell compartment responsible for postnatal muscle growth and regeneration. Accordingly, Pax7-null mice exhibit reduced muscle growth, marked muscle wasting, and an extreme deficit in muscle regeneration after acute injury (; ). Despite these differences, both Pax3 and Pax7 appear to mark a population of muscle progenitors (Pax3/Pax7 cells) in the dermomyotome of embryonic somites (; ; ; ). Pax3/Pax7 cells proliferate and persist throughout embryonic and fetal development and are proposed to be the cellular origin for satellite cells. Pax3 expression is down-regulated in satellite cells before birth and appears to be confined to a subpopulation of satellite cells in specific muscle groups (; ). Thus, cumulative evidence supports distinct roles for Pax3 and Pax7 during myogenesis and a critical requirement for Pax7 in satellite cell specification, survival, and potentially, self-renewal (; ; ; ; ; ). In adult muscle, quiescent satellite cells express Pax7, whereas expression of Myf5 and MyoD is low or nondetectable (; ; ). Pax7 persists at lower levels in recently activated, proliferating satellite cells and is rapidly down-regulated in cells that commit to terminal differentiation (; ). In culture, Pax7 appears to be up-regulated and persists in a small population of myogenic cells that down-regulate MyoD expression. This subpopulation remains undifferentiated and mitotically inactive, resembling a quiescent satellite cell (; ). We have previously shown that Pax7 overexpression recapitulates these events in proliferating myogenic cells (). Moreover, ectopic expression of Pax7 can efficiently repress the MyoD-dependent conversion of mesenchymal cells to the muscle lineage (). Although this is evidence for a functional relationship between Pax7 and the MyoD family of transcription factors, the exact nature of this relationship is controversial (; ; ; ; ). Here, we attempted to delineate the molecular mechanisms involved in Pax7-mediated repression of MyoD function and myogenic progression. Our data indicate that Pax7 blocks myogenesis independently of its transcriptional activity, by a mechanism involving regulation of MyoD protein stability. Similarly, myogenin, but not MyoD, appears to regulate Pax7 function by affecting Pax7 levels. These results provide evidence supporting the existence of a reciprocal inhibition between Pax7 and the muscle regulatory factors (MRFs). Our data suggest that this mechanism may function to regulate the decision of an activated satellite cell to proliferate, commit to terminal differentiation, or reacquire a quiescent state. During myogenic differentiation, Pax7 up-regulation is observed in cells that remain undifferentiated and down-regulate MyoD expression (; ), which is reminiscent of the reserve cell phenotype (). Furthermore, overexpression of Pax7 down-regulates MyoD in satellite cells and myogenic cells lines, preventing terminal differentiation and cell cycle progression (). Pax7 also inhibits MyoD-induced myogenic conversion of C3H10T1/2 cells (), suggesting that Pax7-mediated inhibition of muscle differentiation occurs before induction of myogenin expression. To determine whether these effects are regulated at the transcriptional level, we asked if Pax7 differentially affects MyoD and myogenin transcriptional activity. We first assessed the ability of MyoD and myogenin to activate transcription from a luciferase reporter driven by the proximal regulatory region of the gene (; ), in the presence or the absence of Pax7. Ectopically expressed MyoD activates the reporter gene >5,000-fold during the myogenic conversion of C3H10T1/2 cells, whereas ectopically expressed myogenin activates the reporter gene >700 fold (). Cotransfection of Pax7 represses MyoD transcriptional activity up to 90% in a dose-dependent manner (). However, myogenin activity was substantially less affected by Pax7 coexpression (approximately threefold repression at the highest Pax7 dose) than MyoD (). These data suggest that Pax7-dependent repression of myogenesis is specific for MyoD. We hypothesized that inhibition of MyoD function could arise via competition of Pax7 and MyoD for binding to common DNA targets. Thus, MyoD transcriptional activity on a noncanonical regulatory element should be insensitive to Pax7 repression. We tested this possibility by changing the DNA binding specificity of MyoD using a Gal4-MyoD fusion protein and determining the activation of a reporter gene (). Surprisingly, Pax7 was able to repress the activity of the fusion protein (). The inhibition of the Gal4-MyoD activity was quantitatively equivalent to that observed for wild-type MyoD (). This effect is specific for MyoD because a constitutive activator (Gal4-VP16) shows a greatly reduced sensitivity to cotransfection of Pax7 (), suggesting that the ability of Pax7 to repress MyoD transcriptional activity is unlikely to reflect a competitive binding to a common DNA target. This is further supported by the inability of Pax7 to either bind directly to a MyoD target sequence (MCK-Ebox) or disrupt the binding of MyoD, E47, or MyoD-E47 dimers to DNA in electrophoretic mobility shift assays (EMSAs; ). Consequently, we envision at least two mechanisms whereby Pax7 could inhibit MyoD activity: regulating transcription of additional genes required for MyoD function or a nontranscriptional mechanism, such as competition for a common interaction partner. To determine the contribution of Pax7 transcriptional activity to the inhibition of myogenesis, we performed deletion analysis of domains required for this function in Pax7 and tested the ability of the mutant proteins to repress MyoD activity during myogenic conversion of C3H10T1/2 cells. A series of Pax7-deletion mutants were generated (see Materials and methods) containing a myc-tag epitope followed by an NLS inserted at the N terminus of each mutant construct (, top). A prior set of mutants lacking the exogenous NLS exhibited cytoplasmic mislocalization and high variability in protein expression, suggesting major differences in protein stability (unpublished data). The mutant proteins used in subsequent assays (myc-NLS) were expressed at relatively similar levels (, bottom), with the exception of the ΔC mutant, which showed higher levels of protein expression at equivalent amounts of transfected expression vector (, bottom). This difference appears to be related to enhanced protein stability compared with other mutant products (unpublished data). The ability of each Pax7-deletion mutant to repress myogenic conversion of C3H10T1/2 cells induced by ectopic expression of MyoD was then evaluated. Pax7 mutants lacking either the paired-box or the transactivation domains repressed MyoD activity (, left), resembling the effect of the full-length Pax7. These findings correlated with a severe reduction in both myotube formation and expression of myosin heavy chain (MyHC), a marker of terminal differentiation (, right). In contrast, deletion of the homeodomain region abolished the effect of Pax7 on MyoD activity (, left) and failed to block myogenic differentiation (, right). Expression of a deletion mutant containing only the homeodomain and transactivation domain is sufficient to repress myogenic conversion of C3H10T1/2 cells, preventing terminal differentiation (). Interestingly, this mutant appeared more potent than the full-length Pax7 protein (). Ectopic expression of the N terminus plus the paired-box domain or the transactivation domain alone had no considerable effect on MyoD activity (). Together, these data suggest a critical role for the Pax7 homeodomain in repressing myogenesis and inhibiting MyoD function. This effect appears specific for MyoD, as neither wild-type Pax7 protein nor the deletion mutants had substantial effects on myogenin transcriptional activity (). To further determine whether inhibition of MyoD activity requires Pax7- dependent transcription, we analyzed the transcriptional activity of Pax7 and Pax7 mutants on a reporter gene driven by a regulatory sequence derived from the gene (; ) containing both paired-box and homeodomain binding sites (). Unexpectedly, we detected only weak Pax7-dependent activation of the reporter gene under conditions that repressed MyoD activity (, left). However, the Pax-dependent reporter gene can be activated by full-length Pax7 under proliferation conditions (, right). As expected, deletion of either the paired-box or the transactivation domain abolished Pax7 transcriptional activity (). Interestingly, deletion of the homeodomain region, required for repression of MyoD activity, increased Pax7 transcriptional activity (; both under proliferation and differentiation conditions). This is in agreement with previous studies showing a cis-acting transcription repression activity for this domain in Pax7 (). These observations indicate that the ability to repress MyoD activity does not correlate with active Pax7-dependent transcription, suggesting that MyoD protein could be regulated by Pax7 protein interactions. Prompted by these observations, we asked if MyoD protein levels were affected by ectopic expression of Pax7. Western blot analyses of myogenic-converted C3H10T1/2 cell lysates revealed that inhibition of MyoD transcriptional activity and terminal differentiation correlated with changes in the levels of MyoD protein (, top). Full-length Pax7 (FL) or a Pax7 mutant that represses myogenesis (ΔN) reduced MyoD protein levels upon cotransfection, whereas a Pax7 mutant that does not repress myogenesis (ΔHD) had no effect on MyoD levels (, top). These changes appear specific for MyoD, as the levels of Pax7 protein were consistent with the amount of expression plasmid added to the cells (, bottom). Thus, MyoD protein stability appears specifically affected by coexpression with Pax7 and Pax7 mutants that inhibit myogenic differentiation. MyoD is subject to regulation through proteasome- dependent degradation (; ; ; ; ; ; ); thus, we asked if this pathway was involved in down-regulating MyoD protein upon Pax7 coexpression. Loss of MyoD protein after the switch to differentiation media in C3H10T1/2 cells expressing both MyoD and Pax7 can be detected clearly by 24 h (, lane 3). Treatment with the proteasome inhibitor MG132 prevented loss of MyoD protein and rescues MyoD to control levels (, lanes 2 and 1, respectively). Interestingly, MyoD stability is affected by Pax7 coexpression in C3H10T1/2 only upon a switch to differentiation conditions, as we detected no difference in MyoD levels in the presence or absence of Pax7 when cultures were maintained in proliferation media (, lanes 4–7). This observation indicates that MyoD degradation does not occur via nonspecific effects derived from Pax7 overexpression. Most important, proteasome-mediated protein degradation also contributes to MyoD down-regulation induced by Pax7 overexpression in adult primary myoblasts (), as MG132 treatment partially rescues MyoD expression under these conditions (). The inability of MG132 to fully rescue MyoD protein levels in adult myoblasts could be due to a decrease in transcription of the endogenous gene caused by down-regulation of MyoD protein. We expected that rescuing MyoD protein levels would rescue its transcriptional activity. Interestingly, MyoD function was not restored upon proteasome inhibition, as MG132 treatment did not rescue MyoD-dependent activation of the reporter in the presence of Pax7 (), even when robust nuclear coexpression of both transcription factors was observed under these conditions (). This finding suggests that additional events are involved in Pax7-dependent regulation of MyoD activity. Initial events in myoblast differentiation include permanent withdrawal from the cell cycle and induction of myogenin followed by induction of muscle-specific genes. Along with others, we have shown that myogenin and Pax7 expression is mutually exclusive, whereas Pax7 is retained (and up-regulated) only in a small population of cells that escape differentiation and down-regulate MyoD expression (; ). In light of our new observations, we asked if up-regulation of myogenin controls Pax7 protein levels. Western blot analysis of C3H10T1/2 cell lysates cotransfected with myogenin and Pax7 revealed a reduction in Pax7 protein when compared with Pax7 levels upon cotransfection with MyoD (, left; compare lanes 3 and 4). Interestingly, myogenin is also considerably reduced upon Pax7 coexpression (, left), suggesting a reciprocal effect on relative protein levels. As observed previously for MyoD, myogenin reduction under these conditions involves proteasome-dependent protein degradation, as treatment with MG132 blocks myogenin loss (, right). Interestingly, MG132 treatment also blocks Pax7 reduction when myogenin is coexpressed (, right). Although the levels of Pax7 and myogenin appear to be reciprocally affected, Pax7 and myogenin are not coexpressed in adult myoblasts (in mice and humans), indicating that these observations may reflect complex population dynamics inherent in an asynchronous population of cells undergoing terminal differentiation. To definitively determine whether Pax7 and myogenin are coexpressed during the early stages of muscle differentiation, we used the MM14 satellite cell line, where cells can be synchronized at M/G1 by mitotic shake-off (; ; ). When induced to differentiate, synchronized MM14 cells express muscle-specific genes within 6–12 h and begin fusion into multinucleated myotubes by 12–15 h, providing a useful assay for cell cycle–specific events associated with terminal differentiation (; ; ). Synchronized MM14 cells were allowed to adhere for 8–10 h in the presence of growth medium and then cultured in differentiation medium for various periods of time (). We observed that Pax7 expression persists in a large fraction of the cell population until 12 h after differentiation induction (, left). As expected, myogenin protein was detectable by 8 h after induction of differentiation, reaching a maximum at 21 h (, middle). Between 8 and 12 h of differentiation, Pax7 and myogenin proteins were largely coexpressed within the same cell population, as 85.5 ± 1.2% (8 h) and 82.2 ± 5.2% (12 h) of the myogenin cells showed robust expression of both markers, indicating that myogenin protein accumulates in Pax7 cells (, right). Coexpression of Pax7 and myogenin is transient because 9 h later (21 h in differentiation medium) the percentage of myogenin cells reached a maximum, whereas the percentage of Pax7 cells dropped to a minimum (, middle and left, respectively). At this time point, expression of both Pax7 and myogenin becomes mutually exclusive, as the percentage of Pax7/myogenin cells decreases to 7 ± 1.9%. By 30 h, percentages of myogenin and Pax7 cells have not changed substantially (>80 and <17%, respectively), but we could no longer detect cells that were positive for both Pax7 and myogenin. The change in the Pax7/myogenin ratio during myogenic differentiation suggests that accumulation of myogenin protein could down-regulate Pax7. Indeed, we observed a reduction in Pax7 protein levels upon ectopic myogenin expression in MM14 myoblasts, even under proliferation conditions (; nondetectable Pax7 in >96% of transfected cells). If myogenin expression is responsible for reducing Pax7 protein, forced loss of myogenin under differentiation conditions should result in the persistence of Pax7 expression. To test this idea, we attempted to knock down myogenin through RNAi. Because siRNA transfection in MM14 myoblasts is inefficient and thus cannot provide a quantitative assessment for the extent of myogenin reduction, we initially tested the efficacy of myogenin-specific siRNAs in C3H10T1/2 cells ectopically expressing MyoD. As determined by Western blot analysis, maximum myogenin knockdown (>120-fold) was obtained at all doses tested (). This effect appears to be specific because MyoD expression was not affected under the same conditions (, right) and myogenin protein remained unaffected in the presence of a nonspecific control siRNA (). RNAi-mediated down-regulation of myogenin prevented the loss of Pax7 protein (high Pax7 signal in ≥60% of total transfected cells) in MM14 myoblasts (). As shown previously, control siRNA had no significant effect on myogenin () or Pax7 protein (not depicted). These data support the hypothesis that Pax7 levels are negatively regulated by myogenin in cells undergoing commitment to terminal differentiation. We then asked whether the rapid loss of Pax7 during commitment to differentiation in MM14 cells involved proteasome activity. Mitotically synchronized MM14 myoblasts were induced to differentiate for 15 h and treated with DMSO (control) or the proteasome inhibitor MG132 for additional 6 h (). At this time point (21 h after differentiation induction), >85% of the control cells expressed myogenin, whereas ∼15% of the cells expressed Pax7 in a mutually exclusive pattern (). After MG132 treatment, the percentage of Pax7 cells increased to ∼60%, whereas the percentage of myogenin cells remained at ∼80% (). Under these conditions, myogenin and Pax7 were coexpressed in ∼50% of the cells analyzed (). Together, these results indicate that proteasome-dependent degradation appears to play an important role in the loss of Pax7 during myoblast commitment to terminal differentiation, correlating with the expression and accumulation of myogenin. Our findings suggest a reciprocal regulation between Pax7/MyoD and Pax7/myogenin during the progression of cell differentiation. We asked whether these observations reflected interactions at the protein level by attempting copurification of Pax7–MyoD complexes or Pax7–myogenin complexes from nuclear extracts. Preliminary data indicated that putative Pax7–MyoD (and Pax7–myogenin) interaction was transient and/or unstable in adult primary myoblasts cultures and in MM14 cells (unpublished data). Thus, we asked whether these complexes could be detected in C3H10T1/2 cells coexpressing myc-tagged Pax7 and MyoD. Under control differentiation conditions, little if any detectable MyoD coimmunoprecipitated with Pax7 (, lane 2), yet MyoD was readily detectable in immunoprecipitates from MG132-treated cells (, lane 3). We could not detect any significant copurification of MyoD and Pax7 under proliferation conditions (unpublished data). We were unable to detect any specific Pax7–myogenin interactions using the same copurification strategy as for MyoD and Pax7 complexes (unpublished data). This could be explained by the strong effect on protein stability observed when both myogenin and Pax7 are coexpressed and thus may reflect a transient interaction disrupted during isolation. Our previous results () and the apparently weak Pax7–MyoD physical interaction, suggest that Pax7 and MyoD coexist in protein complexes through indirect interactions. This idea is further supported by the observation that these proteins do not interact directly during in vitro coimmunoprecipitation assays (). Similarly, we cannot detect a direct interaction between Pax7 and myogenin (). Together, these data suggest that upon external stimuli, Pax7 and members of the MRF family (i.e., MyoD and myogenin) can interact with common elements in a protein complex, leading to functional inhibition and changes in protein stability perhaps by altering interactions within the protein complexes. #text C3H10T1/2 cells were cultured in DME and 10% fetal bovine serum at 37°C and 5% CO. For myogenic conversion assays, cultures were induced to differentiate in DME and 2% fetal bovine serum for 48 h or as specified. MM14 cells were cultured in F12-C, 15% horse serum, and 500 pM FGF-2 at 37°C and 5% CO. Differentiation was induced by culture in F12-C and 10% horse serum. When specified, cells were treated with 20–25 μM MG132 (Calbiochem) for 6–8 h before harvesting or fixation. Pax7 deletions were constructed via PCR mutagenesis using pcDNA-Pax7d vector () as a template. Appropriate restriction sites were included at the 5′-end of forward and reverse primers (). Pax7 (and mutant) cDNAs were subcloned into pcDNA3-myc-NLS expression vector (BamHI and XhoI sites; a gift from J. Lykke-Andersen and G. Singh, University of Colorado, Boulder, CO). In frame cloning introduces a single copy of a myc-tag epitope followed by the SV40 T-antigen NLS, to the 5′-end of each cloned cDNA. Myogenic conversion of C3H10T1/2 cells was induced by transfecting (Superfect; QIAGEN) 1 μg/well (12-well plate) of the pRSV-MyoD vector. Differentiation was induced 24 h after transfection for 24 or 48 h as indicated. When required, pcDNA-Pax7 vector was cotransfected along with pRSV-MyoD or pEMS-myogenin vectors at the indicated molar ratios. pcDNA3 was used as control DNA. To evaluate MyoD transcriptional activity, 1 μg of the reporter gene was transfected in the absence or the presence of 0.4 μg pRSV-MyoD and in the absence or presence of pcDNA-Pax7 (at the specified molar ratios), in triplicate for each condition. 0.025 μg of the CMV-LacZ expression vector was cotransfected as a marker for transfection efficiency, and pcDNA3 was used as control DNA. After differentiation induction, whole cell lysates were collected and luciferase and β-galactosidase activities were determined using the Dual-Light System (Applied Biosystems) as reported previously (). Total protein content was estimated (micro BCA; Pierce Chemical Co.) for subsequent analyses. Where indicated, the fold difference between maximum activation (reporter plus MyoD or myogenin expression vector) and the activation in different experimental conditions was represented as fold repression. C3H10T1/2 cells were cotransfected with reporter gene and either Gal4-MyoD or Gal4-VP16 fusion proteins in the presence or the absence of pCDNA3-Pax7 at the indicated molar ratios. Pax7 and Pax7-deletion mutants were tested for transcriptional activation in C3H10T1/2 cells as described above by cotransfection with the reporter gene (provided by F. Barr, University of Pennsylvania, Philadelphia, PA), in the presence or absence of MyoD. For in vivo coimmunoprecipitation experiments, C3H10T1/2 cells were transiently transfected with 1:1 molar ratio (MyoD/myc-Pax7), as described previously. When indicated, cells were incubated with 20 μM MG132 for 6 h before harvest. Cells were washed twice and harvested in ice-cold PBS using a cell scraper. Cell pellet was recovered by centrifugation and resuspended in 1 ml buffer A (10 mM Hepes, pH 7.6, 1.5 mM MgCl, 10 mM KCl, and 0.5 mM DTT). After a 10-min incubation in ice, cell pellet was recovered, resuspended in 400 μl buffer A, and disrupted in ice using a Dounce tissue grinder. Cell nuclei were recovered by centrifugation and resuspended in 200 μl buffer B (20 mM Hepes, pH 7.6, 0.5 mM EDTA, 100 mM KCl, 10% glycerol, 2 mM DTT, 3 mM CaCl, 1.5 mM MgCl, 0.25 mM NaVO, 1 mM NaF, 50 mM β-glycerophosphate, and protease inhibitor cocktail). Nuclear fraction was treated with nuclease S7 (Roche; 6 mU/μg of total DNA) for 10 min at 37°C. Nuclease activity was stopped by addition of EDTA (20 mM final concentration), and nuclear fraction was incubated for 2 h at 4°C with gentle rotation. Extracts were recovered by centrifugation. For immunoprecipitation, total protein was equalized (∼200 μl at 1 mg/ml), precleared with 20 μl of agarose–protein G (50% slurry; Pierce Chemical Co.), and incubated in the presence or absence of anti–myc tag antibody (clone 9B11 at a dilution of 1:1,000; Cell Signaling) at 4°C overnight. Immunocomplexes were captured by incubation with agarose–protein G for 3 h at 4°C, washed five times for 5 min each in buffer B, and eluted by resuspending beads in 50 μl 2× SDS-PAGE loading buffer and boiling for 5 min. For in vitro coimmunoprecipitation experiments, S-labeled proteins were obtained by coupled transcription and translation in rabbit reticulocyte lysate (Promega). Protein interaction and immunopurification (using equivalent protein concentration estimated by autoradiography) was performed as described by using anti–myc tag antibody. Proteins were visualized by SDS-PAGE and autoradiography (Storm 860 Scanner [Molecular Dynamics]; control software version 5.03). Gel mobility shift assays were performed from rabbit reticulocyte translated proteins () or purified proteins () as required. Approximately equal amounts of each factor were added to the binding reactions (estimated by S-methionine incorporation in a translation reaction performed in parallel). pEMS-myogenin (1.5 μg/well; 6-well plate) was used to ectopically express myogenin in MM14 cells and adult primary myoblasts (Lipofectamine 2000; Invitrogen). Cells were fixed and subjected to immunofluorescence staining 24 h after transfection. For myogenin expression knockdown, 200 nM pool siRNA duplexes (Dharmacon) were transfected in MM14 cells (Transmessenger; QIAGEN). RISC-free siRNA (Dharmacon) was used as a negative control. Cells were fixed 48–72 h after transfection. Specific and control siRNA duplexes were provided by Y. Fedorov (Dharmacon, Lafayette, CO). Whole C3H10T1/2 cell extracts were obtained by disruption in modified RIPA lysis buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% IGEPAL, 1 mM NaFl, 1 mM NaVo, and 1× Complete anti-protease cocktail [Roche]), and incubating for 10 min at 4°C. Lysates were cleared by centrifugation. 30–50 μg total protein were loaded onto 10% SDS-PAGE gels and transferred onto polyvinylidene difluoride membranes (Millipore). Primary antibodies and dilutions used were as follows: mouse monoclonal anti-MyoD1 (clone 5.8A; Vector Laboratories) at 1:100; mouse monoclonal anti-Pax7 (Developmental Studies Hybridoma Bank) at 1:10 (cell culture supernatant); mouse monoclonal anti-myogenin (F5D; Developmental Studies Hybridoma Bank) at 1:10 (cell culture supernatant); mouse monoclonal anti–α-tubulin (DM1A; Sigma-Aldrich) at 1:100; mouse monoclonal anti–myc tag (9B11; Cell Signaling) at 1:1,000. Anti-mouse HRP-conjugated secondary antibodies (Promega) were used at 1:5,000, and HRP activity was visualized using the ECL Plus Western Blotting Detection System (GE Healthcare). When required, x-ray films were scanned (Powerlook 1120 scanner; UMAX), digitalized (VueScan 7.6.8; Hamrick Software), and analyzed (ImageJ; NIH) for figure preparation. Cells were fixed in 4% paraformaldehyde for 20 min. Primary antibodies and dilutions used were as follows: mouse monoclonal anti-Pax7 (Developmental Studies Hybridoma Bank) at 1:5 (cell culture supernatant); rabbit polyclonal anti-MyoD (Santa Cruz Biotechnology, Inc.) at 1:30; rabbit polyclonal anti-myogenin (Santa Cruz Biotechnology, Inc.) at 1:30; mouse monoclonal anti-MyHC (MF20; Developmental Studies Hybridoma Bank) at 1:5 (cell culture supernatant). Secondary antibodies conjugated to Alexa 594 or Alexa 488 were obtained from Invitrogen. Vectashield (Vector Laboratories) was used as mounting media. Micrographs were taken from an epifluorescence microscope (Eclipse E800 [Nikon] using 20×/0.50 and 40×/0.75 objectives [Nikon]) at RT, using Slidebook v3.0 acquisition software (Intelligent Imaging Innovations, Inc.) coupled to a digital camera (Sensicam; Cooke). Digital deconvolution for single plane images (no neighbors) was applied (when required) to acquired images (Slidebook v3.0). For figure preparation, images were exported into Photoshop (Adobe). If necessary, the brightness and contrast were adjusted to the entire image, the image was cropped, and individual color channels were extracted (when required) without color correction adjustments or γ adjustments. Final figures were prepared in PowerPoint (Microsoft) and Illustrator (Adobe).

The small GTPase Ras (H-, K-, and N-Ras) activated by extracellular signals exerts its cellular or physiological functions through binding to a variety of target proteins (; ). Raf is one of the best characterized target proteins of Ras. The Raf family consists of three members: Raf-1 (C-Raf), A-Raf, and B-Raf. They share the three conserved regions CR1, 2, and 3. CR1 contains the Ras-binding and cysteine-rich domains (CRDs), both of which participate in binding to active Ras and membrane recruitment, whereas CR3 encompasses the Ser/Thr protein kinase domain (; ; ). The binding of active Ras to Raf triggers activation of the extracellular-regulated kinase (ERK)–MAPK pathway (Raf–MAPK and ERK kinase [MEK]–ERK), which participates in a variety of cellular, physiological, and pathological responses (; ; ). For instance, the growth factor–induced activation of Ras leads to ERK pathway–mediated cell proliferation and morphogenesis during animal development. Oncogenic Ras causes constitutive activation of the ERK pathway, resulting in cellular transformation and tumorigenesis. The Ras–ERK pathway also regulates differentiation in a variety of cell types (). This pathway is essential for NGF-induced neuronal differentiation in PC12 cells (). In contrast, this pathway, which is activated by serum mitogens or growth factors as well as by oncogenic Ras, negatively regulates skeletal muscle cell differentiation (; ). Skeletal muscle cell differentiation necessitates irreversible cell cycle arrest in G phase, the expression of a battery of muscle-specific genes, and cell fusion to form multinucleated myotubes (; ). The tumor suppressor protein Rb is indispensable for cell cycle arrest (; ; ), whereas MyoD family and MEF2 family transcription factors bring about the myogenic program by inducing muscle-specific gene expression (; ). The negative regulation of skeletal muscle cell differentiation by the Ras–ERK pathway is ascribable to interference with the expression or functions of MyoD and MEF2 families (; ; ; ) and with the activation of Rb (; ; ). Thus, it is reasonable that the deprivation of serum mitogens or growth factors leads to inactivation of the Ras–ERK pathway and, consequently, results in skeletal muscle cell differentiation. However, detailed mechanisms of this inactivation of the Ras–ERK pathway have not been fully elucidated. The activity of the Ras–ERK pathway is tightly controlled through the coordinated action of both positive and negative regulators that function at various levels of the pathway and at different time points of the responses. Recently, several suppressor proteins of the Ras–ERK pathway have been identified. These include Sprouty (Spry)/Spred proteins (; ; for review see ), RKIP (Raf kinase inhibitor protein; ; ), and IMP (; for review see ). Both Spry and RKIP bind to Raf to inhibit the ERK pathway by preventing the binding of MEK. Spry also intercepts growth factor receptor–Ras–ERK signaling by interacting with distinct proteins, depending on the signaling. In contrast, IMP uncouples signal transduction from Raf to MEK by inactivating the scaffold protein KSR (kinase suppressor of Ras). However, it has been poorly understood whether regulatory proteins that directly bind to Ras to prevent the ERK pathway are present. M-Ras, a member of the Ras family small GTPases, induces the reorganization of actin cytoskeleton, cellular transformation, inhibition of myogenic differentiation, and neuronal differentiation in PC12 cells (; ; ; ). To elucidate the molecular mechanisms of these cellular functions of M-Ras, we have identified proteins that bind to a constitutively activated mutant of M-Ras. One of them was a novel splicing isoform of A-Raf, designated as DA-Raf1, which contains the Ras-binding domain (RBD) but lacks the kinase domain. DA-Raf1 bound to both active Ras and M-Ras and interfered with the ERK pathway. Its expression level was prominently elevated, and it positively regulated myogenic differentiation by inducing cell cycle arrest, muscle-specific protein expression, and myotube formation. Thus, DA-Raf1 is the first identified intrinsic dominant-negative antagonist of the Ras–ERK pathway. To elucidate the molecular mechanisms of the cellular functions of M-Ras, we have screened a mouse brain cDNA library by a yeast two-hybrid system and cloned cDNAs of proteins that bind to a constitutively activated mutant of M-Ras(G22V) (Gly22 is substituted by Val). These proteins include B-Raf, RalGDS, Rgl2, AF-6, and a C terminus–truncated isoform of A-Raf consisting of N-terminal 186 amino acids (). We designate this truncated protein DA-Raf1 (deleted A-Raf). DA-Raf1 bound not only to M-Ras(G22V) but also to H-Ras (both wild type and a constitutively active G12V mutant), whereas it did not bind to their dominant-negative mutants M-Ras(S27N) and H-Ras(S17N), as determined by a two-hybrid binding assay (). A pull-down assay also showed that GST-tagged DA-Raf1 bound to the constitutively active mutants of M-Ras and H-Ras as well as v–K-Ras but not to their dominant-negative mutants (). Moreover, immunoprecipitation of myc-tagged H-Ras(G12V) or M-Ras(G22V) exogenously expressed in COS-1 cells with the anti-myc mAb myc1-9E10 resulted in the coprecipitation of HA-tagged DA-Raf1 (), indicating their interaction in vivo. Immunoprecipitation of H-Ras in differentiating C2C12 skeletal muscle cell lysates and in mouse brain extracts brought about the coprecipitation of endogenous DA-Raf1 (see ), also implying the binding of endogenous Ras and DA-Raf1. Immunofluorescent staining in combination with myc epitope tagging located both H-Ras(G12V) and M-Ras(G22V) to membrane ruffles and H-Ras(G12V) to vesicles such as macropinosomes as well () as in C2C12 myoblasts (Fig. S1 A, available at ). HA–DA-Raf1 was colocalized with the constitutively active H-Ras and M-Ras to these membrane-associated structures (Fig. S1 A). On the other hand, in serum-starved C2C12 cells expressing wild-type myc–H-Ras or M-Ras, HA–DA-Raf1 was diffusely distributed throughout the cytoplasm and coincided with neither wild-type H-Ras nor M-Ras (). When the cells were stimulated with FBS or FGF2, however, HA–DA-Raf1 was colocalized with the activated H-Ras and M-Ras to the plasma membrane or vesicle membrane–associated structures (). In contrast, if myc–H-Ras(S17N) or M-Ras(S27N) was expressed, the membrane ruffles or vesicles were not formed even by stimulation with FBS or FGF2 (Fig. S1 B). HA–DA-Raf1 was diffusely distributed throughout the cytoplasm and coexisted with neither H-Ras(S17N) nor M-Ras(S27N) in these stimulated cells (Fig. S1 B). These results support the interaction of DA-Raf1 with active Ras and M-Ras in vivo. All known Raf family proteins (Raf-1, A-Raf, and B-Raf) have three conserved regions: CR1, 2, and 3 (; ; ; ). CR1 contains the RBD and CRD, both of which participate in binding to active Ras and membrane recruitment, whereas CR3 represents the kinase domain. The amino acid sequence of DA-Raf1 deduced from its nucleotide sequence indicated that DA-Raf1 exactly corresponds to the N-terminal 186 amino acids of A-Raf, which contains CR1 but lacks CR2 and 3 (). Comparison of the nucleotide sequence of mouse cDNA with that of the genomic sequence of the mouse gene revealed that mRNA is generated by the alternative splicing of pre-mRNA (). cDNA retains intron 6–7 linked to exon 6 and, consequently, gives rise to the termination codon TAG starting from the second nucleotide of intron 6–7 (). Poly(A) tail is added 12 bases downstream of the putative poly(A) addition signal AATAAA. To confirm that is actually expressed in vivo, we conducted RT-PCR and Northern blotting. Quantitative RT-PCR using the sense primer (nt 62–79), which is common to both and , and the antisense primer (nt 620–639), which is specific for (), amplified major 578-bp and minor 695-bp DNA bands from mRNAs of various mouse tissues (). Nucleotide sequences of these DNAs indicated that the former band represented expected , whereas the latter band corresponded to the DNA containing intron 5–6 (117 nt) between exons 5 and 6. The retention of intron 5–6 generates the termination codon TGA starting from the second nucleotide of intron 5–6 (). We refer to this splicing isoform as DA-Raf2. encodes a protein consisting of the N-terminal 153 amino acids of A-Raf, which also includes RBD and CRD (). Examination of the genomic sequences of rat and human genes revealed that both of the genes have exon–intron junctional arrangements and sequences very similar to those of the mouse gene. Consequently, mRNAs encoding DA-Raf1 and 2 seem to be generated by the alternative splicing of rat and human pre-mRNAs as well (Fig. S2, A and B; available at ). Indeed, we have cloned rat and human cDNAs by RT-PCR from rat PC12 cells and human HeLa cells, respectively. Quantitative RT-PCR showed that both and were ubiquitously expressed in a variety of tissues examined and that they were prominently expressed in the brain and heart, among others (). In contrast, had more limited tissue distribution. It was most highly expressed in the liver among the tissues examined, as has been previously reported (), whereas it was scarcely expressed in skeletal muscle (). was also highly expressed in the brain, whereas was widely expressed except in the brain. Next, we conducted Northern blotting with probe 1, which can hybridize to , , and mRNAs in common, and conducted blotting with probe 2, which specifically recognizes mRNA (). mRNA (1.65 kb) and mRNA (1.35 kb) were abundantly present in the brain and heart and moderately in other tissues, which is consistent with the results of RT-PCR analysis. On the other hand, mRNA (2.4 kb) was expressed in the liver at a moderate level and had a more restricted tissue distribution (). Immunoblotting analysis with an anti–DA-Raf pAb showed that DA-Raf1 protein was abundantly present in the brain and moderately in the heart, skeletal muscle, uterus, and spleen (). DA-Raf2 was also detected at least in the brain. In contrast, A-Raf was predominantly present in the liver and spleen, as corroborated by immunoblotting with an anti–A-Raf pAb (). Collectively, mRNA and protein are actually present in mouse tissues with a wider tissue distribution than A-Raf. The structure of the DA-Raf1 protein, which contains both RBD and CRD but lacks the kinase domain, suggests that it serves as an intrinsic dominant-negative antagonist of the Ras–ERK pathway and possibly as an antagonist of other Ras-activated signaling pathways such as Ras–phosphatidylinositol 3-kinase (PI3K)–Akt signaling. To assess this possibility, we transfected NRT9 cells (the mouse NIH3T3 fibroblasts stably transformed with the oncogenic v- gene) with -tagged cDNA and selected eight stable transfectants (NRT/DR1–8). The expression of exogenous myc–DA-Raf1 in these cell lines was detected by immunoblotting with the anti-myc mAb (). The levels of ERK activation and Akt activation were estimated by immunoblotting with an antiphospho-ERK1/2 mAb and an antiphospho-Akt pAb, respectively. NRT9 and mock-transfected NRT9 cells had much higher activating phosphorylation levels of ERK1/2 than did normal NIH3T3 cells, whereas all of the NRT/DR cell lines showed ERK1/2 phosphorylation levels comparable with that in normal NIH3T3 cells (). In contrast, activating phosphorylation levels of Akt in NRT9 and mock-transfected NRT9 cells were almost equivalent to that in normal NIH3T3 cells. This is consistent with the results that K-Ras efficiently activates Raf but not PI3K (; ). The Akt phosphorylation levels were substantially unchanged in any of the NRT/DR cell lines (). Transiently expressed myc–DA-Raf1 in C2C12 myoblasts also reduced the ERK1/2 phosphorylation level but not the Akt phosphorylation level (Fig. S3 A, available at ). Consequently, the exogenously expressed DA-Raf1 is likely to antagonize the Ras–ERK pathway but not Ras–PI3K–Akt signaling. Activation of the Raf-mediated ERK pathway by oncogenic Ras induces transformation in mouse fibroblasts (; ; ). Because the activation of ERK was abrogated by DA-Raf1 expression, we examined whether the v–K-Ras–induced transformed phenotype in NRT9 cells was suppressed by DA-Raf1. Normal NIH3T3 cells contained well-developed actin stress fibers and showed an extended flattened cell shape. Although NRT9 cells exhibited transformed morphology (i.e., the loss of stress fibers and consequent round cell shape), all of the NRT/DR cell lines restored the normal flattened cell shape with stress fibers (). Next, we analyzed the cell proliferation rate under a serum starvation condition in 0.5% FBS by counting the cell number. NRT9 and mock-transfected NRT9 cells showed growth rates several fold higher than that of normal NIH3T3 cells, whereas all of the NRT/DR cell lines exhibited growth rates similar to or lower than that of NIH3T3 cells (). When NRT9 and mock-transfected NRT9 cells were cultured for 7 d after reaching confluence in the growth medium, they formed innumerable foci, which merged with each other because of the loss of contact inhibition of transformed cells (Fig. S3 B). In contrast, all of the NRT/DR cell lines as well as normal NIH3T3 cells scarcely formed foci. We further analyzed the anchorage-independent growth potential of these cells in soft agar. Although ∼10–15% of NRT9 and mock-transfected NRT9 cells formed colonies, all of the NRT/DR cell lines formed colonies at very low efficiencies (Fig. S3 C). Next, we injected 1 × 10 cells subcutaneously into dorsal flanks of athymic nude mice. All mice (three out of three) injected with NRT9 or mock-transfected NRT9 cells formed tumors 6–7 mm in diameter within 4 wk after the injection (Fig. S3 D). However, no mice injected with normal NIH3T3 cells or any of the NRT/DR cell lines formed tumors even 8 wk after the injection (Fig. S3 D). These results indicate that the exogenously expressed DA-Raf1 can suppress all of the transformed phenotypes, including tumorigenicity induced by oncogenic Ras. The Ras–ERK pathway negatively regulates skeletal muscle cell differentiation by suppressing the expression or functions of the muscle-specific transcription factors MyoD and MEF2 (; ; ; ). This suppression results in the aborted expression of myogenin, which induces muscle differentiation. Thus, to assess the involvement of DA-Raf1 in the regulation of skeletal muscle cell differentiation, we first analyzed the expression of during the differentiation of C2C12 skeletal muscle cells by quantitative RT-PCR and immunoblotting. Undifferentiated myoblasts terminally differentiate to form multinucleated myotubes by cell fusion by shift from the mitogen-rich growth medium containing 10% FBS to the mitogen-poor differentiation medium containing 5% horse serum. Well-differentiated myotubes are developed by 96 h under this culture condition. mRNA was present at a low level in myoblasts but was markedly induced about 10-fold during differentiation (). mRNA was almost absent in myoblasts but highly induced during differentiation in accordance with the expression (). Immunoblotting showed that DA-Raf1 was almost absent in myoblasts but prominently induced during differentiation (). Similarly, myogenin and late muscle-specific proteins, myosin heavy chain (MyHC) and troponin T (TnT), were also induced (). Although the band derived from mRNA was detected by RT-PCR, DA-Raf2 protein was hardly detected by immunoblotting even in differentiated C2C12 cells, which is different from in the brain. The increased rate of DA-Raf1 protein during differentiation was much higher than that of mRNA. In contrast, the amounts of Raf-1, B-Raf, and A-Raf were almost constant during differentiation (). These results suggest the possibility that DA-Raf1 positively regulates skeletal muscle cell differentiation. In proliferating myoblasts, the activating phosphorylation of MEK and ERK was easily detected by immunoblotting with an antiphospho-MEK1/2 pAb and the antiphospho-ERK1/2 mAb, respectively (). In contrast, phospho-Akt was marginally present in myoblasts. The phosphorylation levels of both MEK and ERK were reduced, whereas that of Akt was elevated during differentiation (). These results are consistent with previous studies (; ). Immunoprecipitated B-Raf from myoblasts efficiently phosphorylated MEK, but the kinase activity was markedly diminished during differentiation (). In contrast, immunoprecipitated Raf1 and A-Raf scarcely phosphorylated MEK regardless of the differentiation stage ( and not depicted). These results are compatible with the notion that B-Raf is the main activator of MEK in most cell types (; ). DA-Raf1 was not coimmunoprecipitated with H-Ras in myoblasts (). As differentiation proceeded, however, DA-Raf1 was coimmunoprecipitated with H-Ras at the level similar to that in the brain (). In contrast, B-Raf was coimmunoprecipitated with H-Ras in myoblasts, and the amounts of precipitated B-Raf decreased as differentiation proceeded (). Collectively, these results imply that the decrease in MEK–ERK activity during C2C12 cell differentiation is ascribed to the retardation of B-Raf activity and that the association of DA-Raf1 with Ras is likely to be responsible for the decline in Ras–B-Raf interaction and the consequent reduction in B-Raf–MEK–ERK activity during the differentiation. Because DA-Raf1 seems to negatively regulate the B-Raf– MEK–ERK pathway during C2C12 cell differentiation, it may be involved in the induction of differentiation. Skeletal muscle cell differentiation entails cell cycle arrest in G phase together with muscle-specific gene expression and myotube formation. To examine whether DA-Raf1 is implicated in the induction of cell cycle arrest, C2C12 cells were transfected with –, and proliferating cells in S phase were detected by the incorporation of BrdU in the nuclei. expression markedly reduced the ratio of BrdU-incorporating cells (). Next, we addressed whether DA-Raf1 participated in the induction of myogenin and late muscle-specific protein expression. The ratio of myogenin-expressing cells was facilitated 1.5–2-fold by the transfection of during 48–72 h in the differentiation medium (). The transfection of also elevated the ratio of MyHC-expressing cells () and TnT-expressing cells () two- to several fold during differentiation. The effects of transfection to induce myogenin and the late muscle-specific proteins were more obvious around 48–72 h after the transfection than at later time points probably because of the up-regulation of endogenous DA-Raf1 protein in control cells and breakdown of the transfected plasmids expressing in the process of time. Therefore, highly expressed DA-Raf1 can bring on the myogenic differentiation program by inducing the cell cycle arrest and expression of myogenin and late muscle-specific proteins. We further examined the role of endogenous DA-Raf1 by RNAi through expression of the two siRNAs, M1 and M2, which target the sequences in the 3′ untranslated regions of mouse / mRNAs. Expression of these siRNAs markedly reduced the amount of endogenous DA-Raf1 protein during the differentiation of C2C12 cells (). In contrast, expression of the other two siRNAs, R1 and R2, targeting rat / mRNA sequences, which are distinct from the mouse counterparts, did not affect the amount of DA-Raf1 protein in C2C12 cells (). C2C12 cells cultured in the differentiation medium for 48 h became quiescent, and only ∼2% of the mock-transfected control cells incorporated BrdU for a further 8 h. However, expression of the mouse-specific siRNAs M1 and M2 resulted in BrdU incorporation in ∼8–12% of the cells for 8 h after the 48-h culture in the differentiation medium (). On the other hand, expression of the rat-specific R1 and R2 did not facilitate the BrdU incorporation (). These results indicate that intrinsic DA-Raf1 is required for the cell cycle arrest that is indispensable for skeletal myocyte differentiation. When the mouse-specific siRNA M1 was expressed, the ratio of myogenin-expressing cells strikingly decreased during 24–96 h in the differentiation medium (). The expression of another siRNA, M2, was less effective than that of M1 but significantly reduced the ratio of myogenin expression. In contrast, the expression of siRNAs R1 and R2 did not reduce the ratio. Both the expression of M1 and that of M2 almost completely suppressed myotube formation by cell fusion, and, consequently, myocytes with the knockdown of DA-Raf were mononucleated (). The mean number of nuclei in individual C2C12 cells expressing R1 or R2 was not so high (2.5–3 at 96 h). This is probably because myotubes generated by cell fusion of a small number of myoblasts are more easily detected than those generated by cell fusion of a large number of myoblasts as a result of limited transfection efficiency. Altogether, these results imply that not only exogenously expressed DA-Raf1 but also intrinsic DA-Raf1 play crucial roles in the induction of myocyte differentiation represented by cell cycle arrest, by the expression of myogenin and a set of late muscle-specific proteins, and by cell fusion–mediated myotube formation. Transfection of in C2C12 cells considerably reduced the ratio of the cells containing active phospho-ERK as detected by immunofluorescent staining (). In contrast, expression of the siRNA M1 or M2 facilitated the ratio of the cells containing phospho-ERK at 72 and 96 h in the differentiation medium (). These results corroborate the notion that the function of DA-Raf1 to induce myocyte differentiation is mediated by suppression of the ERK pathway. There are crucial differences among the canonical members of the Raf family (Raf-1, A-Raf, and B-Raf) in the regulatory mechanisms, target proteins, activities, and biological functions (; ; ). Furthermore, alternative splicing generates multiple splicing variants of B-Raf with different activities (; ). Nevertheless, all of the known Raf family proteins, including these splicing variants, share the three conserved regions CR1, 2, and 3. In marked contrast, DA-Raf1/2 that we have reported here almost exclusively comprises CR1, which encompasses Ras-binding and membrane-targeting RBD and CRD, and lacks CR2 and 3 representing the kinase domain. As expected from its structure, DA-Raf1 bound to active Ras and M-Ras and interfered with the ERK pathway. The binding to active Ras and M-Ras in vivo was confirmed by the findings that DA-Raf1 was translocated to the cellular membranes to colocalize with Ras and M-Ras activated by mitogenic signals. Induction of endogenous DA-Raf1 during C2C12 cell differentiation was accompanied by reduction in the activating phosphorylation levels of MEK and ERK but not that of Akt. In addition, the exogenous expression of in v-–transformed fibroblasts and in C2C12 myoblasts interfered with the Ras–ERK pathway but not with Ras–PI3K–Akt signaling. This is probably because the domains in Ras involved in the binding of Raf and PI3K p110 are partially overlapping but distinct (; ). Thus, DA-Raf1 is likely to compete with the binding of Raf proteins but not with that of PI3K p110. Consequently, DA-Raf1 may exclusively antagonize the Ras–ERK pathway but not the signaling pathways of other target proteins. Several suppressor proteins of the Ras–ERK pathway have been identified, including Spry/Spred, RKIP, and IMP (; ; for reviews see ; ). These proteins antagonize the Ras–ERK pathway by interacting with a variety of components of the growth factor receptor–Ras–ERK pathway, but none of them directly bind to Ras. Thus, the mechanism of DA-Raf1 to interfere with the Ras–ERK pathway is unique and distinct from those of the other suppressor proteins. Exogenously expressed DA-Raf1 suppressed all of the transformed phenotypes, including the tumorigenic potential of the v-–transformed NIH3T3 fibroblasts in nude mice. Accordingly, we can expect that DA-Raf1 might serve as a tumor suppressor protein if the gene has loss of function mutations or if the expression levels are noticeably reduced in tumors of certain tissues. Even if DA-Raf is not a bona fide tumor suppressor protein, it might be applicable for the treatment of Ras-induced tumors, which comprise 15–30% of human cancers (; ), by promoting its expression or alternative splicing efficiency to generate DA-Raf1. The binding of DA-Raf1 to Ras/M-Ras and suppression of the ERK pathway in the cells highly expressing DA-Raf1 imply that DA-Raf can serve as a dominant-negative antagonist of the Ras–ERK pathway. DA-Raf1 seems to bind to Ras/M-Ras faster and more easily than the typical Raf family members because of its small size and its structure that is comprised almost exclusively of the RBD without regulatory sequences. The binding of DA-Raf1 to Ras/M-Ras may not be regulated by a variety of regulatory mechanisms for the typical Raf proteins, including phosphorylation and the binding of 14-3-3 proteins, scaffolding proteins such as KSR, and the suppressor proteins Spry and RKIP (; ; ; ). Thus, DA-Raf1 might accomplish the dominant-negative functions in vivo even if its amount is lower than those of the typical Raf proteins. A-Raf is specifically located to mitochondria (), whereas DA-Raf1 was diffusely present throughout the cytoplasm and translocated to the plasma membrane and vesicle membranes by the mitogenic signals. This suggests that DA-Raf1 exerts the dominant-negative functions against Raf proteins close to these membranes rather than against mitochondrion-associated A-Raf. Among the Raf family members, B-Raf is the main protein that couples Ras to MEK (; ), and this was true for C2C12 cells. In contrast, most of the signaling mechanisms and cellular functions of A-Raf remain enigmatic. Considering that is more highly expressed than , particularly in the brain and myotubes, DA-Raf1 is the primary product of the gene by alternative splicing, at least in these tissues and cells. Furthermore, the binding of endogenous DA-Raf1 to Ras lead to the decline in Ras–B-Raf interaction and B-Raf activity during C2C12 cell differentiation. Consequently, DA-Raf1 is likely to serve as the intrinsic dominant-negative antagonist of the Ras–B-Raf–MEK–ERK pathway. Because the amount of DA-Raf1 was prominently elevated during C2C12 cell differentiation, this elevated amount of DA-Raf1 seems to be required to achieve the positive regulation of differentiation. Indeed, DA-Raf1 exogenously expressed in C2C12 cells facilitated myogenic differentiation by inducing myogenin and a set of late muscle-specific protein expression as well as the cell cycle arrest. Moreover, the knockdown of endogenous DA-Raf interfered with myogenic differentiation by suppressing cell cycle arrest, myogenin expression, and myotube formation by cell fusion. Therefore, DA-Raf1 is essential for skeletal myocyte differentiation. This essential role of DA-Raf1 is probably through suppression of the Ras–ERK pathway because endogenous DA-Raf1 bound to Ras, and the B-Raf–MEK–ERK pathway was suppressed during differentiation. Furthermore, overexpressed DA-Raf1 abrogated ERK activation, whereas the knockdown of DA-Raf facilitated ERK activation. The Ras–ERK pathway negatively regulates muscle-specific gene expression by interfering with the expression of the MyoD family and MEF2 family transcription factors (; ; ; ). The Ras–ERK pathway also counteracts the cell fusion and activation of Rb (; ; ). Rb is required for myogenic differentiation by inducing not only the irreversible cell cycle arrest (; ; ) but also the MEF2-dependent expression of late muscle-specific proteins (). The Ras–ERK pathway causes the inactivation of Rb by transactivating cyclin D1 (), which participates in inactivating the phosphorylation of Rb by forming the complex with Cdk4/6. Insulin-like growth factors activate PI3K–Akt signaling that induces skeletal muscle myogenesis and hypertrophy (; ). Thus, suppression of the Ras–ERK pathway by DA-Raf1 without affecting Ras–PI3K–Akt signaling is favorable for skeletal muscle cell differentiation. Culturing skeletal muscle myoblasts under mitogen-poor differentiation conditions leads to inactivation of the Ras–ERK pathway and, consequently, results in myocyte differentiation. However, the mechanisms of this inactivation of the Ras–ERK pathway have not been fully understood. The prominent induction of DA-Raf1 and its binding to Ras under the differentiation condition and suppression of the Ras–ERK pathway by DA-Raf1 explain the mechanisms of this inactivation of the Ras–ERK pathway. The knockdown of DA-Raf by RNAi interfered with ERK activation and with myocyte differentiation under the differentiation condition. This implies that DA-Raf1 is indeed required for adequate inactivation of the Ras–ERK pathway for myocyte differentiation even under the mitogen-poor condition. We have unraveled the essential role of DA-Raf1 in skeletal muscle cell differentiation. Likewise, scrutiny of the cellular and physiological functions of DA-Raf in other cell types or tissues, in which DA-Raf is highly expressed or highly induced under particular conditions, may lead to the elucidation of unexpected roles of this novel isoform of the Raf family. Mouse brain cDNA library constructed in pACT2 vector (CLONTECH Laboratories, Inc.) was screened by a yeast two-hybrid system with the constitutively active rat (G22V) cDNA () ligated to the DNA-binding domain in pGBD9 vector (CLONTECH Laboratories, Inc.). The yeast strain Y190 was transformed with the pGBT9/Mras(G22V) bait plasmid and then with the pACT2/cDNA library prey plasmids. Double transformants were selected on plates of the minimal synthetic dropout medium. Large colonies were further selected by β-galactosidase colony- lift filter assay. The nucleotide sequence of the obtained 11 positive cDNA clones was determined with a DNA sequencing system (4200G; LI-COR). One of the cDNAs encoded mouse DA-Raf1. Rat and human cDNAs were cloned from PC12 and HeLa cell mRNAs, respectively, by RT-PCR using Omniscript reverse transcriptase and ProofStart DNA polymerase (QIAGEN). Primers used were sense primer 5′-GACAACATGGAACCACG-3′ and antisense primer 5′-TGAAACCTGGAGTGACCAGG-3′ for rat and sense primer 5′-AAGGCTCCATGGAGCCACCA-3′ and antisense primer 5′-GTACCAGATCCTGTTCTAGGC-3′ for human . The sequences of mouse, rat, and human cDNAs have been deposited to the GenBank/ EMBL/DDBJ database with the accession nos. , , and , respectively. RNAs were prepared from a variety of mouse tissues and from cultured cells as described previously (; ). Quantitative RT-PCR and Northern blotting were conducted as described previously (; ). Mouse cDNA was inserted into pQE32 vector (QIAGEN) in frame with the 6× His tag. The recombinant His-tagged DA-Raf1 was expressed in the strain XL1-Blue and affinity purified with Talon Metal Affinity Resin (CLONTECH Laboratories, Inc.). The protein was separated by SDS-PAGE and recovered from the gel. A New Zealand white rabbit was immunized with the purified His–DA-Raf1 emulsified with Freund's complete or incomplete adjuvant (Difco Laboratories). The anti–DA-Raf pAb was affinity purified through formyl-cellulofine (Seikagaku Corp.) coupled with the antigen. Mouse tissues were extracted with radioimmunoprecipitation assay (RIPA) lysis buffer, their protein concentrations were determined by Bradford method, and the extracts were treated with SDS sample buffer. Cultured cells were directly lysed with the SDS sample buffer. These samples were subjected to SDS-PAGE and immunoblotting. Antibodies used for the immunoblotting were the anti–DA-Raf pAb; pAbs to Raf-1, B-Raf, A-Raf (Santa Cruz Biotechnology, Inc.), MEK, phospho-MEK (Ser217/221), p44/42 MAPK, Akt, and phospho-Akt (Ser473; Cell Signaling Technology); and mAbs to myogenin (F5D), MyHC (A4.1025), β-tubulin (E7; Developmental Studies Hybridoma Bank), TnT (NT302; ), and phospho-p44/42 MAPK (Thr202/Tyr204; E10; Cell Signaling Technology). The immunological reaction was detected with Western Lightning Chemiluminescence Reagent Plus (PerkinElmer) and analyzed with a luminoimage analyzer (LAS-1000mini; Fuji). cDNA was inserted into pGEX-6P2 vector (GE Healthcare) in frame with the GST tag. The recombinant GST–DA-Raf1 was expressed in XL1-Blue and affinity purified with glutathione–Sepharose 4B (GE Healthcare). The cDNAs encoding the constitutively active and dominant-negative mutants of M-Ras and H-Ras as well as the v- gene were fused in frame to the N-terminal myc tag in pEF-BOS/myc vector. These recombinant plasmids were transfected to COS-1 cells with FuGENE 6 Transfection Reagent (Roche). The cells were lysed, and the GST pull-down assay was conducted as described previously (). Binding proteins were detected by immunoblotting with the anti-myc mAb myc1-9E10 (American Type Culture Collection). COS-1 cells were cotransfected with the recombinant plasmids pEF-BOS/HA-DAraf1 and pEF-BOS/myc-Hras(G12V) or pEF-BOS/myc-Mras(G22V). 24 h after the transfection, the cells were lysed with 800 μl of lysis buffer (10 mM Tris-HCl, pH 8.0, 150 mM NaCl, 5 mM MgCl, 1% NP-40, 1 mM DTT, 0.1 mM PMSF, 2 μg/ml leupeptin, and 1 μg/ml pepstatin) and centrifuged at 16,000 for 15 min. C2C12 cells cultured in the differentiation medium were lysed with the lysis buffer. Mouse brain was extracted with RIPA lysis buffer. The supernatant solutions were preabsorbed with protein G–Sepharose 4 Fast Flow (GE Healthcare). Protein G–Sepharose was mixed with the mAb myc1-9E10 or mAb to H-Ras (238; Santa Cruz Biotechnology, Inc.) for 1 h and then mixed with the preabsorbed supernatant solution for 1 h. The beads were washed five times with the lysis buffer. Proteins bound to the beads were eluted with 9 M urea and detected by immunoblotting with the pAbs to HA tag (Medical and Biological Laboratories), DA-Raf, and B-Raf or with the anti–pan-Ras mAb RAS10 (Calbiochem). C2C12 cells cultured in the differentiation medium were lysed with RIPA lysis buffer. B-Raf, Raf-1, and A-Raf were immunoprecipitated with each specific pAb prebound to protein A–Sepharose 4 Fast Flow (GE Healthcare). The Sepharose beads were washed five times with RIPA lysis buffer and suspended in 20 μl of the assay buffer (20 mM MOPS, pH 7.2, 75 mM MgCl, 25 mM EGTA, 25 mM glycerol 2-phosphate, 1 mM sodium orthovanadate, and 1 mM DTT) containing 0.5 μg of recombinant GST–MEK1 (Upstate Cell Signaling Solutions). They were mixed with 20 μl of the assay buffer containing 0.74 MBq γ-[P]ATP (166.5 TBq/mmol; MP Biomedicals) and incubated for 30 min at 30°C. The supernatant was mixed with 40 μl of 2× SDS sample buffer and subjected to SDS-PAGE. Phosphorylation of MEK1 was detected by autoradiography and analyzed with a bioimaging analyzer (BAS-1500 Mac; Fuji). Mouse C2C12 myoblasts were cotransfected with pEF-BOS/HA-DAraf1 and pEF-BOS/myc-Hras or pEF-BOS/myc-Mras by using LipofectAMINE 2000 (Invitrogen). 24 h after the transfection, the cells were cultured under a serum-free condition for 48 h and stimulated with 10% FBS or 50 ng/ml recombinant human FGF2 (Wako) for 30 min. They were fixed with 4% PFA in PBS and permeabilized with 0.2% Triton X-100 in PBS. The cells were then incubated with the anti-HA tag pAb, the mAb myc1-9E10, and with AlexaFluor546-conjugated goat anti–rabbit IgG and AlexaFluor488–goat anti–mouse IgG (Invitrogen). Fluorescent images were acquired at room temperature using a confocal laser-scanning microscope (LSM 410; Carl Zeiss MicroImaging, Inc.) with a 63× NA 1.40 plan-Apochromat objective lens and were assembled with Photoshop 9 (Adobe). In each plate, photographs were trimmed, and each fluorochrome was adjusted identically for brightness and contrast to represent the observed images. The NRT9 cell line was established by stably transfecting the mouse NIH3T3 fibroblasts with the v- gene in pLNCX vector (CLONTECH Laboratories, Inc.) followed by selecting with G418. To establish NRT9 cell clones stably transfected with , an XbaI fragment of in pEF-BOS/myc-DAraf1 was subcloned in pMIKHygB vector. The stable transfectants (NRT/DR cells) were selected with 500 μg/ml hygromycin B. The exogenously expressed myc–DA-Raf1 was detected by immunoblotting with the mAb myc1-9E10. The amounts of ERK1/2, Akt, and their activating phosphorylation levels were analyzed by immunoblotting with the p44/42 MAPK pAb, Akt pAb, phospho-p44/42 MAPK (Thr202/Tyr204) mAb E10, and phospho-Akt (Ser473) pAb, respectively. Cell morphology and actin cytoskeleton were analyzed by staining the cells with rhodamine–phalloidin (Invitrogen). Focus formation assay and colony formation assay in soft agar were performed as described previously (). To detect tumorigenicity, 1 × 10 cells were suspended in 100 μl of sterile PBS and subcutaneously injected into dorsal flanks of nude mice. All mouse protocols were approved by the guidelines of Chiba University. To induce terminal differentiation in C2C12 cells, undifferentiated myoblasts maintained in the growth medium (DME containing 10% FBS) were shifted to the differentiation medium (DME containing 5% horse serum) as described previously (). The undifferentiated myoblasts were transfected with pEF-BOS/HA-DAraf1 in the growth medium and were left under the differentiation condition. DA-Raf1 and muscle-specific proteins (myogenin, MyHC, and TnT) were detected by staining with the anti-HA tag pAb and the mAbs to myogenin, MyHC, and TnT followed by AlexaFluor488–goat anti–rabbit IgG and AlexaFluor546–goat anti–mouse IgG. Fluorescent images were acquired at room temperature using a microscope (Axioskop; Carl Zeiss MicroImaging, Inc.) with a 40× NA 1.30 plan-Neofluar objective lens, CCD camera (CoolSNAP; Roper Scientific), and Openlab 3.0.3 software (Improvision). The photographs were processed with Photoshop 9 (Adobe) as described in the Epitope-tagging immunofluorescent localization section. RNAi of DA-Raf was conducted by expressing siRNAs with pSilencer 2.1-U6 neo vector (Ambion) as described previously (). The target sequences of mouse siRNA M1 and M2 were 5′-AAAGAAGGATTGGAGGACCCT-3′ (nt 595–615 from the initiation codon) and 5′-AAAGCAAGATCTGATACAGAG-3′ (nt 818–838), respectively. The target sequences of rat siRNA R1 and R2 were 5′-AAGTATTGAATAAGGGCATGG-3′ (nt 670–690; 90.5% homologous to the corresponding mouse sequence) and 5′-AACAAGATCTGATGTAGAGTG-3′ (nt 820–840; 85% homologous to the corresponding mouse sequence), respectively. C2C12 myoblasts were cotransfected with these siRNA-expressing pSilencer vectors and one tenth of the amount of pEGFP-C1 vector (CLONTECH Laboratories, Inc.) to monitor the cells expressing siRNAs. They were shifted to the differentiation medium 24 h after the transfection and were cultured for a further 96 h. To assess the interfering effect of these siRNAs, pSilencer 5.1-U6 Retro vector (Ambion) expressing these siRNAs was constructed, and the retroviruses were produced. C2C12 cells were infected with these retroviruses, and infected cells were selected with puromycin. Lysates of the selected cells cultured in the differentiation medium were subjected to immunoblotting with the anti–DA-Raf pAb. Proliferating or cell cycle–arrested cells were determined by the incorporation of BrdU. C2C12 myoblasts transfected with pEF-BOS/HA-DAraf1 were cultured for 24 h in the growth medium. Then, they were incubated for 2 h in the same medium supplemented with 10 μM BrdU and 1 μM 5-fluoro 2′-deoxyuridine. C2C12 cells transfected with the pSilencer plasmids encoding siRNAs together with one tenth of the amount of pEGFP-C1 vector were cultured for 24 h in the growth medium. They were shifted to the differentiation medium and cultured for 48 h and for 8 h in the same medium supplemented with BrdU/5-fluoro 2′-deoxyuridine. The BrdU-incorporating nuclei were detected by staining with an anti-BrdU mAb (GE Healthcare) supplemented with 10 μg/ml DNase I followed by AlexaFluor546–goat anti–mouse IgG. Fig. S1 shows the colocalization of DA-Raf1 with constitutively active H-Ras and M-Ras but not with dominant-negative H-Ras and M-Ras. Fig. S2 indicates the nucleotide sequences and corresponding amino acid sequences of rat and human DA-Raf1/2. Fig. S3 shows the suppression of ERK phosphorylation without affecting Akt phosphorylation in C2C12 myoblasts and suppression of the v–K-Ras–induced transformed phenotype in NRT9 cells by DA-Raf1. Online supplemental material is available at .

The ubiquitous cytoskeletal 8–12-nm intermediate filaments (IFs) are made of cell type–specific molecular components that are encoded by several multigene families encompassing at least 71 functional genes in human (; ; ). The largest subfamilies are the type I and type II keratins in epithelial cells, which are obligatory heteropolymers contributing equally to mature keratin filaments (KFs) by forming stable double-stranded coiled-coil heterodimers (). KFs provide mechanical stability and overall resilience for epithelial tissues (; ). They are organized in different ways in the various epithelial cell types, generating thick bundles in epidermal keratinocytes, apically restricted and densely woven mats in enterocytes, subplasmalemmal enrichments in hepatocytes, or finely dispersed three-dimensional networks in several cultured epithelial cell types. These alternative arrangements in combination with the diverse cell shapes that are required in living tissues suggest that the KF cytoskeleton is highly dynamic. Two types of regulation are being considered: differential association of KFs with scaffolding proteins and keratin modification (; ). A scaffolding function is apparently provided by cell adhesion structures, and key molecular players have been identified such as the desmosomal plaque proteins desmoplakin/plakophilin/plakoglobin (; ; ; ) and the hemidesmosomal components plectin and bullous pemphigoid antigen 1 (; ). The multifunctional cytoskeletal cross-linker plectin may also participate in attachment to other cytoskeletal elements and the nucleus (; ; ). In addition, keratin bundling is favored by proteins such as filaggrin (). The importance of protein modification for keratin organization has been widely recognized and phosphorylation is considered to be the major contributing factor (). Because altered phosphorylation is often accompanied by structural changes, it is generally assumed that a cause-and-effect relationship exists between both. In accordance, increased keratin phosphorylation is observed during mitosis and in various stress paradigms, i.e., in situations of considerable keratin reorganization (; ; ). It was further suggested that keratin phosphorylation is the result of antagonistic kinase and phosphatase activities that are regulated in a cell type–specific manner (). Yet, a direct temporal and spatial correlation between specific enzymatic activity, altered target phosphorylation sites in keratin polypeptides and consecutive keratin reorganization, has not been established so far in the context of a living cell. To examine direct linkages between kinase/phosphatase activities, keratin modifications, and KF organization, we therefore established epithelial cell culture systems in which we are able to monitor in real time the rapid and reversible orthovanadate (OV)-induced KF network disassembly into keratin granules by live-cell fluorescence microscopy (). Although overall keratin phosphorylation did not change substantially under these conditions (), keratin reorganization could be prevented by preincubation with a specific p38 MAPK inhibitor (). Because p38 is known to phosphorylate keratins (; ; ), we decided to analyze the relationship between its activity, modification of keratin target sites, and keratin arrangement in more detail. We have recently shown that rapid and reversible restructuring of the keratin cytoskeleton occurs in the presence of OV, a well known, yet rather unspecific tyrosine phosphatase inhibitor that also effects other enzymes such as cellular ATPases (; ). This reorganization can be effectively prevented by ambient light, and to a lesser degree, by preincubation with the specific p38-inhibitor SB203580 (). The latter observation suggested that signaling via the p38-MAPK pathway is involved in the regulation of KF organization. To further pursue this idea, we examined the distribution of activated p38 by immunofluorescence microscopy of OV-treated vulva carcinoma-derived AK13-1 cells producing fluorescent HK13-EGFP. Shortly after addition of the drug, a remarkable redistribution of phosphorylated p38 (p38) from a diffuse cytoplasmic pool lacking colocalization with the keratin system to a marked granular pattern occurred, coinciding with the appearance of keratin granules (). At intermediate stages of KF disassembly remnant, normal-appearing KFs were negative for p38, whereas thick KF bundles were weakly positive and newly formed granules were most strongly stained with p38 antibodies (). The same pattern of codistribution was noted using either polyclonal or monoclonal antibodies (compare with Fig. S1, A and B; available at ). Furthermore, cotransfection of fluorescent K18 and p38 resulted in colocalization of both proteins in prominent aggregates of living epithelial cells (Fig. S1, C and D). On the other hand, antibodies directed against the other phosphorylated MAPKs ERK and JNK did not react with OV-induced keratin granules (not depicted). It is known that p38 phosphorylates specific keratin residues (; ; ). Using an antibody directed against keratin 8 (K8)-S73, the major site in the K8 head domain that is phosphorylated by p38 (; ), we could confirm previous results demonstrating that this epitope is virtually absent in normal-appearing interphase KF networks (; ). Soon after OV addition, however, K8-S73 was readily detected on newly formed keratin granules (). In contrast, keratin phosphoepitope K8-S431 was present in both intact KF networks and keratin granules (Fig. S2, available at ). The same constitutive phosphorylation in untreated and OV-treated cells was also noted for K18-S33 (not depicted). The consistently observed keratin aggregation in cells overexpressing p38-GFP (Fig. S1 C) suggested that p38 activation determines keratin organization. To further examine this idea, A431 cells were transfected with the constitutively active p38 upstream regulators MKK3 and MKK6 () either alone or in combination. These cells were identified by immunofluorescence microscopy detecting the attached Flag-tag (), or by direct fluorescence microscopy of a linked CFP moiety (). Transfected cells presented keratin granules that were positive for p38 () and contained K8-S73 epitopes (). Many dead cells were noted 24 h after transfection, probably a consequence of p38-induced apoptosis, which was also noted in p38-GFP-producing cells. As an alternative to the slow-acting genetic p38 activation, pharmacological means were used to facilitate short-term induction and to prevent complex downstream effects of p38 action. Already 3 min after addition of the p38 activator anisomycin (), abundant keratin granule formation was observed (see for 5-min time point) that was accompanied by p38 recruitment () and appearance of K8-S73 (not depicted). Collectively, these observations show that p38 activation leads to keratin reorganization and keratin modification. Conversely, A431 cells were treated with specific p38 inhibitors. In addition to the previously used p38 inhibitor SB203580 (), we tested SB202190 that also preferentially interferes with the α and β isoforms of p38 (). This treatment did not disrupt the KF network over a wide concentration range, although KFs appeared to coalesce and concentrate gradually in the central cytoplasm over time. When, in addition, cells were incubated with OV, keratin granule formation was efficiently prevented (compare ). To down-regulate p38 synthesis genetically, expression of p38 isoforms was first determined by RT-PCR. α, γ, and δ isoforms could be amplified from AK13-1 cells but not p38-β. Therefore, plasmids were constructed encoding α-, δ-, and δ/γ-specific p38 shRNAs together with fluorescent indicator proteins. Transfected AK13-1 cells exhibited considerable reorganization of the keratin cytoskeleton in each instance (Fig. S3, A and B; available at ). A substantial depletion of KFs was seen in most parts of the cytoplasm, sparing only desmosome-anchored filaments. Most material coalesced in a juxtanuclear position. It still contained filaments that were compacted, but did not aggregate into granules. When these cells were treated with OV, the remaining filaments did not form granules as in neighboring nontransfected cells (). Collectively, the data suggested that the p38 K8-S73 target residue contributes to keratin granule formation. We therefore decided to compare the KF network-forming properties of the phosphorylation-incompetent K8-S73A mutant and the K8-S73D mutant mimicking constitutive phosphorylation. When introduced together with human K18 chimera HK18-YFP into A431 cells, only 31.33 ± 3.87% of K8-S73A–producing cells contained keratin granules ( = 560; four experiments) whereas 59.9 ± 1.59 of K8-S73D–producing cells presented abundant granules ( = 604; four experiments). To abolish the mitigating effects of endogenous wild-type keratins, K8 constructs were transfected together with HK18-YFP chimeras into SW13 cells that lack cytoplasmic IFs (compare ). In each instance, however, a normal-appearing KF network was formed (for similar results in NIH-3T3 cells, see also ), although an increase in the soluble pool of cells producing K8-S73D cannot be excluded (Fig. S3, C and D). Turnover of these K8 mutant-containing networks and motility of KF precursors were analyzed by time-lapse fluorescence microscopy and FRAP. No differences were noted in comparison to cells producing only wild-type keratins (unpublished data). In addition, motility of cells transfected with mutant K8 constructs was indistinguishable from cells synthesizing wild-type keratins. These results demonstrate that K8-S73 alone is not sufficient to mediate KF network rearrangements, although it appears to contribute, in combination with other factors, to keratin rearrangement in a cell context–dependent fashion. To examine the extent of phosphorylation of p38 and keratins upon keratin granule formation, biochemical analyses were performed of cells treated with OV. A rapid and considerable rise of p38 was readily detectable in immunoblots of total cell lysates in response to OV (). Furthermore, reaction of cytoskeletal fractions with antibodies directed against K8-73 revealed a similarly rapid and coincident increase (), whereas no changes were observed for other keratin phosphoepitopes (). To examine interactions between keratins and p38, coimmunoprecipitation experiments were performed. Using different detergents including NP-40 and empigen BB (), we were able to detect p38 in anti-keratin precipitates from colon carcinoma-derived HT29 cells whose level was, however, not increased upon OV treatment in these cells or in AK13-1 cells (; unpublished data). Either we were not able to solubilize the newly formed keratin granules efficiently (see also ), and/or existing bonds were disrupted during cell fractionation. To investigate whether p38 recruitment and simultaneous increase of site-specific keratin phosphorylation apply also to other situations when keratin granules are formed, AK13-1 cells were subjected to various types of stress. A 5-min incubation at 60°C induced keratin granules that were most prominent in peripheral regions and colocalized with p38 (). Hypotonic stress that was applied by incubation in 150 mM urea resulted in reorganization of the KF system into clumped material that also stained for p38 (). Conversely, hypertonic stress also induced disassembly of the KF network into granular material. p38 antibodies reacted again specifically with the granular material, but not with the remaining thin filaments (). The K8-S73 epitope was also detected in the granular material in each situation (not depicted). These observations support the notion that p38 recruitment is a general mechanism that is associated with KF phosphorylation and reorganization. Considerable keratin reorganization takes place during mitosis, and it was reported that A431-cells almost completely disassemble their network into soluble material and rapidly moving keratin granules during early prophase (, ). When we stained dividing AK13-1 cells with p38 antibodies, an almost complete colocalization with forming keratin granules was noted during metaphase (). In very early prophase, KFs disintegrated into p38-positive granules (). Occasionally, cells were seen with intermediate phenotypes, i.e., with dense filament bundles or granules emanating from thin filaments, both of which may correspond to intermediate stages of either assembly or disassembly (). Interestingly, peripheral parts of KF networks were sometimes labeled by p38 antibodies in areas of cells that did not attach directly to neighboring cells but contained lamellipodial-like extensions (not depicted). These areas were recently identified as regions of high KF turnover (; ). K8-S73 appearance was noted in each instance (not depicted). Given that p38 is recruited to keratin granules that are formed in very different situations, we decided to examine the composition of granules containing mutant keratins. We used MCF7-derived cell line MT5K14-26, producing mutant EYFP-K14 fluorescent chimeras (). The abundant peripheral keratin granules were strongly stained by p38 antibodies, whereas the residual perinuclear KFs were not (). In comparison, cell line MT5K14-25 synthesizing wild-type fluorescent K14 chimera EYFP-K14 presented only diffuse p38 fluorescence (). Coimmunoprecipitation experiments, however, did not reveal an increased association of keratins with p38 in the mutant cells, possibly due to the inability to solubilize the p38-positive granular material or due to disruption of the association during immunoprecipitate preparation (). Quantification of the p38 level in MT5K14-25 and MT5K14-26 showed that total p38 was the same in both, whereas p38 was twofold increased in EYFP-K14 cells (), reminiscent of the reported increase of JNKs in keratinocytes expressing other keratin mutants (). Furthermore, endogenous keratins colocalized with the mutant polypeptides and K8-S73 epitopes were seen in perinuclear filaments and most prominently in keratin granules (). In contrast, this epitope was only expressed in mitotic cells of line MT5K14-25 (). Similar to A431-derived cells, all different keratin organizational forms were positive for K8-S431 in both MCF7-derived cell lines (). shows that p38 inhibitors do not result in immediate KF network disassembly, although long-term down-regulation resulted in network depletion (Fig. S3). To find out whether dynamic aspects of KF organization are altered in these conditions, time-lapse fluorescence microscopy was performed. A typical sequence is shown in and Video 1 (available at ). Addition of the p38 inhibitor SB202190 led to an increased concentration of the fluorescent KF network toward the central cytoplasm. This altered arrangement was, however, not caused by cell retraction because the periphery remained in place and continued to exhibit high ruffling activity with multiple dynamic filopodial extensions. Remarkably, the peripheral cytoplasmic area did not contain KF precursors that are usually generated in this region (). Stress fibers were noted in close proximity to the periphery of the retracted keratin network (, arrowheads). To find out whether a similar inhibition of KF precursor formation occurs also in cells producing mutant keratins, MT5K14-26 cells were treated with SB202190 ( and Video 2; available at ). KF precursor formation ceased immediately after drug application. At the same time, ruffling activity of the peripheral cytoplasm continued. Upon washout of the drug, keratin particle formation resumed in the peripheral cytoplasm. Despite these strong effects of p38 inhibition on keratin particle formation, cells retained keratin granules even after extended periods of SB202190 treatment (unpublished data). Time-lapse fluorescence analysis helped to solve this apparent paradox, revealing that keratin particles became stabilized upon p38 inhibition (Videos 2 and 3). The rapid dissolution observed in untreated MT5K14-26 cells (see also ) was almost completely abolished. In sum, our observations highlight the importance of p38 activity for KF precursor formation and KF network turnover. The current study identified p38 as a major regulator of KF network formation by revealing a tight temporal and spatial correlation between activation of p38, recruitment of p38 to KFs, keratin phosphorylation at specific p38 target sites, and ensuing disassembly of KFs into granules. This sequence of events was observed during physiological situations of KF reorganization, most notably in dividing cells, in cells subjected to stress and, quite remarkably, in cells producing mutant keratins (summary of colocalization results in Fig. S4, available at ). Furthermore, experimental up-regulation of p38 activity led to keratin granule formation, whereas its down-regulation prevented it. The speed and reversibility of the observed p38-dependent processes make them highly suitable to accomplish transient network attenuations in various in vivo situations that require finely tuned cell shape changes. Indeed, p38 is present in epithelial cells and responds rapidly (i.e., within a few minutes) to various types of stress. These include physiologically relevant mechanical pressure (), osmotic shock (; ), and UV irradiation (). Moreover, p38 is induced in keratinocytes upon wounding (). Accordingly, it has been observed that keratinocyte outgrowth from human skin explants and keratinocyte migration are dependent on p38 (; ; ; ). Furthermore, p38 is activated by proinflammatory cytokines in A431 cells () and is increased in psoriatic skin (). The migrating and dynamic keratinocytes require increased flexibility of their cytoskeleton that may in part be provided by p38-mediated keratin network alterations. In support, p38 staining was frequently observed in lamellipodia in our cell systems. The relevance of p38 activity in epithelial physiology is further underscored by the recent observation that pemphigus vulgaris IgGs that bind to the extracellular portion of the desmosomal cadherin desmoglein 3 induce “retraction” of the KF system via p38 (). Collectively, overwhelming evidence exists demonstrating a prominent role of p38 in short-term regulation of epithelial plasticity that should be distinguished from long-term effects on keratinocyte differentiation and apoptosis (, ; ). Furthermore, functions of p38 signaling are apparently not restricted to keratins, but are also of relevance for vimentin () and neurofilaments (). On the other hand, other stress-activated protein kinases may be involved in IF organization, although activated JNKs and ERKs were not found in association with keratin granules in our cell systems. Yet, in other cells, K8 has been identified as a binding partner of JNKs that also phorphorylate K8-S73 in vitro () and are elevated in cells producing mutant keratins (). Similarly, altered phosphorylation, presumably mediated by ERK1/2, has been reported for K8-S431 upon EGF stimulation and in response to osmotic stress (; ). Our results in combination with many other publications (compare ; ; ) strongly suggest that keratin phosphorylation is the primary mechanism by which the keratin network is reorganized. K8-S73 has received particular attention because it presents an on/off behavior during mitosis, in various stress situations including shear stress, and during apoptosis (; ; ). Furthermore, the sequence motif surrounding K8-S73 is conserved among several type II keratins as LLS/TPL where the corresponding threonine residue is also phosphorylated by p38 in an on/off fashion, leading to increased keratin solubilization, filament reorganization, and collapse during mitosis and UV- or anisomycin-induced apoptosis, as well as in psoriatic skin and squamous cell carcinoma (). Phosphorylation of sites in the head domain has been shown to be essential for the assembly of different IF types (; ; ; ; ). The increase in negative charge by phosphorylation is believed to prevent interactions of the head domain with the negatively charged rod thereby keeping the head in an “open” configuration. Presumably, this configuration is part of opening up the filament structure during disassembly into granules and may also be needed during intermediate assembly steps. The observed p38-dependent and head domain–specific phosphorylation of K8-S73 before KF disassembly, as well as the inhibition of both KF precursor formation and mutant keratin granule disassembly by the p38 inhibitor SB202190, strongly support this notion (Videos 1–3). Yet, further experiments are needed to find out whether network disintegration into granules is due to keratin disassembly or simply a “clumping” of filaments, both of which may be determined by phosphorylation. Further support for the importance of head domain phosphorylation was provided for vimentin, in which case S55A mutants were shown to prevent network disassembly during mitosis (). Similarly, light chain neurofilament S55D mutants interfered with proper neurofilament assembly in cultured cells and transgenic mice (, ). On the other hand, K8-S73 is not alone sufficient for KF network disruption (Fig. S3; ), indicating that additional p38 target sites in K8 and/or other keratins are necessary. Constitutive differences in overall keratin phosphorylation could well explain the different reactivities of KF networks in different cell types during mitosis and in various stress situations (compare ), and, even more, the observed lack of keratin reorganization in vivo, e.g., in K8-S73-containing hepatocytes (). It has been proposed, therefore, that multiple events of phosphorylation and dephosphorylation cooperate in KF organization (). Cooperation of several phosphorylation sites for IF formation has also been documented for GFAP in transgenic mice (), and the importance of cross talk between head and tail domain phosphorylation for neurofilament assembly in specific cellular topologies has been described (). Finally, we cannot exclude that p38 activity affects, in addition to keratins, factors which in turn regulate KF properties (; ; ). The strong and highly specific staining of cytoplasmic granules containing mutant keratins with antibodies against p38 and K8-S73 was not expected, and we were even more surprised to be able to almost instantaneously prevent keratin granule formation by pharmacological p38 inhibition. Interestingly, hyperphosphorylated keratin granules are present in toxic liver disease in the form of cytoplasmic Mallory bodies (; ; ; ), whose formation also relies on p38 activity (). p38 activity is likely also relevant for other IF aggregates that occur in many different diseases, including cardiac myopathy, glial Alexander disease, and several neurodegenerative diseases (; ; ). Notably, neurofilament aggregates that are formed in motoneurons of patients suffering from amyotrophic lateral sclerosis contain p38 together with phosphorylated NF-M and NF-H (; ). A similar colocalization was also noted in a transgenic mouse model of amyotrophic lateral sclerosis (; ). In addition, mimicking-increased IF phosphorylation by expression of the NF-L S55D mutants led to prominent neuropathology with neurofilament inclusion bodies in neuronal perikarya and swollen axons in transgenic mice (). While this investigation focused on the consequences of p38 recruitment for structural and dynamic properties of the keratin cytoskeleton, several publications have provided evidence that this interaction bears also important consequences for cell physiology. In particular, it has been suggested that keratins act as a phosphate “sponge” for stress-activated kinases based on observations in transgenic mice overexpressing K8-S73A and presenting increased susceptibility to liver injury and apoptosis (). Our data, however, extend this model by demonstrating that activated p38 is not simply bound to the IF cytoskeleton, but also induces considerable organizational alterations and thereby affects cell shape, flexibility, and most likely other basic cellular functions (). cDNAs coding for HK8-CFP and HK18-YFP have been described previously (; ), and a cDNA for HK18-RFP was obtained from Anne Kölsch (this institute). A cDNA in the EcoRI site of Bluescript coding for keratin mutant K8-S73A was provided by Dr. Omary (Stanford University, Palo Alto, CA; ). The ∼600- bp HindIII fragment encompassing the mutated part of K8 was excised and exchanged for the corresponding wild-type fragment in HK8-ECFP–encoding plasmid that was described recently (). In addition, a cDNA coding for K8-S73D mutant in a mammalian expression vector was also given to us by Dr. Omary (). A p38-GFP cDNA was given to us by Dr. Bradham (Duke University, Durham, NC; ). Flag-tagged cDNAs coding for constitutively active MKK3 (in pRc/RSV) and MKK6 (in pCDNA3) were provided by Dr. Davis (University of Massachusetts Medical School, Worcester, MA; ). The HindIII/SpeI fragment coding for MKK3 was further subcloned into the corresponding sites of modified plasmid pTER () containing additional CMV promoter-driven fragments coding for either ECFP (pTER-ECFP) or mRFP (pTER-mRFP; see ). In the case of MKK6, the MKK6-encoding plasmid and both pTER derivatives were cleaved with XbaI, blunt-ended, and cut with HindIII before ligation. RT-PCR using the Enhanced Avian Reverse Transcriptase kit (Sigma- Aldrich) was performed for amplification of RNAs coding for specific p38 isoforms. The oligonucleotides used to amplify the α, β, γ, and δ isoforms are listed in Fig. S5. The following cell lines were propagated as described previously: vulva carcinoma-derived A431 cells of clones E and AK13-1 (), colon adenocarcinoma-derived HT29 cells (ATCC HTB 38), spontaneously immortalized mammary epithelial EpH4 cells (compare ), and mammary adenocarcinoma-derived MCF7 cells of lines MT5K14-25 producing EYFP-K14 and MT5K14-26 synthesizing EYFP-K14 (). Foreign DNA was transfected into subconfluent cells by using the Lipofectamine 2000 reagent following the instructions provided by the manufacturer (Invitrogen; ). OV was obtained from Sigma-Aldrich and a 1M stock solution was prepared in ddHO. The dissolved drug was added to subconfluent cultured cells in the dark at final concentrations between 10 and 30 mM for 5–10 min. To specifically inhibit p38α and β activity, cells were treated with SB202190 (Sigma-Aldrich) at final concentrations ranging from 50 to 100 μM. To induce p38 activity pharmacologically, cells were incubated with anisomycin (Sigma-Aldrich) at 30 μM. In hyperosmotic stress assays, cells at 70–80% confluence were incubated in medium containing 200 mM sorbitol for 5–25 min at 37°C before fixation. Hypoosmotic stress conditions were attained by incubation in medium supplemented with 150 mM urea for 5–15 min at 37°C. Cells recovered subsequently in normal medium for 5–20 min before further processing. For heat stress, subconfluent cells were placed in a 60°C incubator for 5–10 min and were then fixed. In most instances cells were fixed by incubation for 5 min in −20°C cold methanole followed by a short 10-s treatment with −20° cold acetone. After air drying, cells were ready for antibody incubation. To detect soluble fluorescent proteins it was necessary to fix cells for 10 min at 4°C in 3% formaldehyde freshly prepared in PBS. A short 1-min treatment with 0.01% digitonin in PBS followed at room temperature. Alternatively, cells were treated with −20°C cold methanole for 10 min. After a subsequent 10-min incubation in 4°C PBS, cells were treated with 5% bovine serum albumin for 15 min at room temperature. Further antibody incubations followed in the same way as for methanol/acetone-fixed cells (). The following antibodies were used: polyclonal rabbit antibodies directed against total p38, dual phosphorylated p38 (recognizing T180/Y182), total JNK, JNK, total ERK1/2, ERK1/2, and against the Flag epitope DYDDDK were obtained from New England Biolabs, Inc.; murine monoclonal antibodies against dual phosphorylated p38 (recognizing T180/Y182) were from New England Biolabs Inc.; and monoclonal antibodies against K8-S73 (LJ4), K8-S431 (5B3), K18-S33 (IB4), and total K8/K18 (L2A1) were provided by Dr. Omary (; ); secondary antibodies were ordered from Dianova and Rockland. Images were recorded with an inverse fluorescence microscope (IX-70; Olympus) and an attached slow scan camera (model IMAGO, Till Photonics; ; ). In some instances a confocal laser scanning microscope was used (SP5; Leica). Pictures were edited with Adobe Photoshop CS software to prepare figures. Recording of phase-contrast images and fluorescence patterns on an inverse fluorescence microscope were performed as described previously (; ). Total cell lysates were prepared by adding 200–500 μl buffer (62.5 mM Tris-HCl, 2% [wt/vol] SDS, 10% glycerol, 50 mM DTT, and 0.01% [wt/vol] bromophenol blue) per 100 mm Petri dish. Solubilized cells were scraped off, sonicated briefly, and heated to 95°C for 5 min before SDS-PAGE. Cytoskeletal fractions were prepared by standard procedure (compare ). SDS-PAGE and immunoblotting was done as described previously (). In some instances, membranes were stripped by incubation in buffer containing 62.5 mM Tris, 2% (wt/vol) SDS, and 100 mM mercaptoethanol for 30 min at 55°C. For immunoprecipitation, cells were washed twice with PBS supplemented with 5 mM EDTA, scraped off, and solubilized in ice-cold buffer containing 1% NP-40, 5 mM EDTA, and 0.1 mM PMSF together with protease inhibitors (1 tablet of protease inhibitor cocktail “cOmplete” from Roche per 50 ml) by incubation at 4°C in a shaker for 2 h. Particles were centrifuged down at 18,000 for 20 min at 4°C. Keratin antibody L2A1 was added to the supernatant. After incubation for 1 h, preequilibrated protein A–Sepharose CL-4B (GE Healthcare) was added and incubation at 4°C continued for another 2 h under constant agitation. Three brief wash steps in buffer containing 0.1% NP-40, 5 mM EDTA, and 0.1 mM PMSF followed and the remaining material was suspended in 62.5 mM Tris-HCl, 2% (wt/vol) SDS, 10% glycerol, and 0.01% (wt/vol) bromophenol blue, heated to 95°C for 2 min and subjected to SDS-PAGE. The images shown in Fig. S1 (A and B) demonstrate that monoclonal antibodies directed against p38 present the same colocalization with keratin granules as other polyclonal antibodies (see Fig. 1). Similarly, fluorescent K18 and p38 chimeras colocalize in prominent cytoplasmic aggregates (Fig. S1, C and D). The fluorescence micrographs provided in Fig. S2 show that phosphorylation of K8-S431 is not affected by OV; those in Fig. S3 demonstrate that the keratin cytoskeleton is reorganized in response to p38 down-regulation and that K8-S73D mutation does not affect overall network formation. Fig. S4 summarizes colocalization results for keratins and specific keratin phosphoepitopes or phosphorylated p38, JNKs, and ERKs during various situations of pronounced KF network alterations. Fig. S5 lists the oligonucleotides used for cloning. Videos 1 and 2 corresponding to Fig. 10, A and B, respectively, reveal the inhibitory effects of pharmacological p38 inactivation on KF precursor formation. Video 3 further shows that p38 inhibition prevents mutant keratin granule turnover. Online supplemental material is available at .

Chemotaxis is a pivotal response of many cell types to external spatial cues. It plays important roles in diverse functions, such as finding nutrients in prokaryotes, forming multicellular structures in protozoa, tracking bacterial infections in neutrophils, and organizing the embryonic cells in metazoa (; ; ). In cells, extracellular cAMP functions as a chemoattractant that is detected by specific G protein–coupled surface receptors. Chemotaxis is achieved by coupling gradient sensing to basic cell movement. Two important questions on chemotaxis are What is the compass detecting the cAMP gradient? and How is this signal transduced to localized pseudopod formation? Pseudopod extension at the leading edge is mediated by the formation of new actin filaments, whereas acto-myosin filaments in the rear of the cell inhibit pseudopod formation and retract the uropod. In , myosin filament formation is regulated by cGMP (), whereas in mammalian cells it is regulated by Rho-kinases (). We are beginning to understand the signals that regulate actin polymerization at the front of the cell (; ; ). Phosphatidylinositol-trisphosphate (PIP) is a very strong candidate to mediate directional sensing in neutrophils and . PIP is formed at the side of the cell closest to the source of chemoattractant (; ; ; ; ). Furthermore, PIP is a very strong inducer of pseudopod extensions, as demonstrated in ()–null mutants with elevated PIP levels, and subsequently more pseudopods (). Unexpectedly, inhibition of PI3-kinase (PI3K) has only moderate effects on chemotaxis in (; ; ; ) and mammalian cells (; , ), demonstrating that PI3K signaling is dispensable for chemotaxis. What are the signaling pathways that mediate chemotaxis in -null cells? It has been argued that the optimal second messenger mediating directional sensing will have a lifetime of ∼5 s and a diffusion rate constant of ∼1 μm/s (). Molecules like cAMP, cGMP, H, K, or IP do not meet these criteria. PIP, however, perfectly fits in this biophysical profile: very low basal levels, rapid transient accumulation after cAMP stimulation with a half-life of ∼5 s, and a diffusion rate constant of ∼0.5 μm/s (). Possible alternatives for PIP in -null cells are the lipid products of PLC, PLA2, and PLD, or combinations of these enzymes. Ca may play a role in chemotaxis as well, because in cells its diffusion is rather slow (). In this study, we investigated the role of several potential second-messenger systems in chemotaxis. The results show that inhibition of PI3K and PLA2 strongly reduces chemotaxis. Inhibition of PLC or intracellular Ca signaling has little direct effect on chemotaxis. However, chemotaxis in -null cells appears to be completely dependent on PLA2 activity, whereas chemotaxis in cells lacking the IP receptor depends only on PI3K activity, suggesting that PLC and intracellular Ca are not mediators of chemotaxis but regulate the activity of PI3K and PLA2, respectively. The chemoattractant cAMP stimulates at least eight second-messenger systems in , adenylyl cyclases, guanylyl cyclases, proton production, K fluxes, Ca uptake and release from internal stores, PLC, PI3K, and PLA2. Null mutants of most second-messenger systems have been made; none of these mutants exhibit a severe chemotactic defect (; ). Apparently, multiple pathways mediate chemotaxis. To identify the signaling pathways that collectively mediate chemotaxis, we measured chemotaxis in several mutant cells treated with mixtures of drugs to inhibit one, multiple, or all messenger systems. To limit the number of combinations to be tested (all possible combinations of 8 pathways is 8! or 40,320 conditions), as explained in the Introduction, we identified potential second messengers that have a diffusion rate constant between 0.3 and 3 μm/s and accumulate transiently with a lifetime of 3–10 s after stimulation with the chemoattractant cAMP. Potential second messengers are phospholipids such as PIP and phosphatidylinositol-bisphosphate (PIP), fatty acids, and cytosolic Ca. Therefore, we investigated four signaling pathways, PI3K, PLC, PLA2, and cytosolic Ca. For each second-messenger system, we aimed at collecting two independent datasets, obtained either with a mutant defective in that second-messenger system or with a drug inhibiting the enzyme activity, respectively. We have tested the following conditions: -null or wild-type cells treated with the PI3K inhibitor LY294002 at a concentration of 50 μM, -null or wild-type cells treated with the PLC inhibitor U73122 at a concentration of 10 μM, the PLA2 inhibitors quinacrine at 20 μM and p-bromophenacyl bromide (BPB) at 2 μM, and a combination of mutants and drugs to affect cytosolic Ca, notably mutant cells lacking the IP receptor in combination with 10 mM EGTA to block Ca uptake. The concentrations of drugs used were obtained either from published dose response curves (U73122 [] and LY294002 []) or from dose response curves presented in for BPB and quinacrine. Half-maximal inhibition of chemotaxis to 50 nM cAMP was observed at 0.5 μM BPB or 5 μM quinacrine and >80% inhibition at a concentration of 2 μM BPB or 50 μM quinacrine. Chemotaxis to 1,000 nM cAMP is only slightly inhibited, even at the highest concentrations used, indicating that BPB and quinacrine are not harmful to the cells. In addition, cells treated with BPB exhibit a normal PIP response, as shown by the translocation of PHcrac-GFP to the leading edge, suggesting that inhibition of the PLA2 pathway does not interfere with the PI3K pathway (see ). All single and multiple combinations inhibiting four signaling pathways yielded ∼70 experimental conditions, to be tested at two cAMP concentrations with at least three replicates (Table S1, available at ). We have used the small population assay () to screen these 500 conditions. In this assay, small droplets containing wild-type or mutant cells are deposited on hydrophobic agar. Cells can freely move within the boundary of the droplet but cannot move out of the droplet. Therefore, any directional movement of the cells leads to the accumulation of cells at the boundary of the small population, which is rapidly scored. In the shallow absolute gradient induced by 50 nM cAMP (dC/dx = 100 pM/μm), a weak chemotaxis response is observed in ∼50–70% of the populations, whereas 1,000 nM cAMP induces a steep absolute gradient (dC/dx = 2,000 pM/μm) and a strong chemotaxis response in 90–100% of the populations. In videos that were analyzed for detailed cell movement, chemotaxis by 50 nM cAMP had the characteristics of a biased random walk with a chemotaxis index of ∼0.5, whereas the response toward 1,000 nM cAMP exhibits the characteristics of directional movement with a chemotaxis index of ∼0.85. With this assay it is possible to simultaneously test, in 1 d, 30 different conditions (mutant cells or drugs) at two cAMP concentrations, each with 12 populations that are observed at least four times. This large dataset is presented as Table S1. In and , the results are discussed in a logical and reduced format. p fig The strain AX3 was used as wild-type control in all experiments. The mutants strains used are the -null strain 1.19 (), the -null strain GMP1 (), and the -null strain HM1038, lacking the IP receptor (). Cells were grown in shaking culture in HG5 medium (contains, per liter, 14.3 g oxoid peptone, 7.15 g bacto yeast extract, 1.36 g NaHPO × 12 HO, 0.49 g KHPO, and 10.0 g glucose) at a density between 5 × 10 and 6 × 10 cells/ml. Cells were harvested by centrifugation for 3 min at 300 , washed in PB (10 mM KHPO/NaHPO, pH 6.5), and starved in PB in 6-well plates (Nunc) for 5 h. Cells were then resuspended in PB, centrifuged, washed once in PB, and resuspended in PB at a density of 6 × 10 cells/ml. Chemotaxis measured with the small population assay () was performed in the wells of a 6-well plate with 1 ml agar nonnutrient hydrophobic agar (11 mM KHPO, 2.8 mM NaHPO, and 7 g/liter hydrophobic agar) containing the indicated concentration of the drugs. Droplets of ∼0.1 μl of 5 h–starved cells (6 × 10 cells/ml) were placed on the agar. After 30 min, chemotaxis toward cAMP was tested by placing a second 0.1 μl droplet, with the indicated concentration of cAMP, next to the droplet of cells. The distribution of the cells in the droplet was observed about every 10 min for 90 min. Chemotaxis of cells within a droplet was scored positive when the cell density at the cAMP side was at least twice as high as the opposite side of the droplet (). The maximal chemotactic response is faster for 50 nM cAMP (20–40 min) than for 1,000 nM cAMP (40–60 min). Also, some mutants respond faster (-null) or slower (-null) than wild-type cells. Recorded is the fraction of droplets scored positive, averaged over three successive observations at and around the moment of the maximal response. The data presented are the means and SEMs of at least three independent measurements on different days. Chemotaxis was also measured with micropipettes containing 100 μM cAMP using an inverted light microscope (CK40; Olympus) with a 20× NA 0.4 objective (LWD A240; Olympus) equipped with a charge-coupled device camera (TK-C1381; JVC). The field of observation is 358 × 269 μm. Images were captured every 10 s for 30 min on a PC using VirtualDub software and Indeo Video 5.10 (Ligos) compression. The chemotaxis index, defined as the ratio of the cell displacement in the direction of the gradient and its total traveled distance, was determined for ∼25 cells in a video as follows. First, the position of the centroid of a cell was determined with ImageJ (rsb.info.nih.gov/ij) for frames at 60-s intervals, yielding a series of coordinates for that cell. Using these coordinates, the chemotaxis index of each 60-s step was calculated and averaged, yielding the chemotaxis index for that cell in the video. The data shown are the mean and SEM of the chemotaxis indices from at least three independent experiments with ∼25 cells per experiment. The same experimental setup with micropipettes was used for analyses of cells expressing the PIP sensor PHcrac-GFP. Confocal images were recorded with a confocal laser-scanning microscope (LSM 510 META-NLO; Carl Zeiss MicroImaging, Inc.) equipped with a plan-apochromatic 63× NA 1.4 objective (Carl Zeiss MicroImaging, Inc.). For excitation of the fluorochrome GFP (S65T variant) a 488-nm argon/krypton laser was used, and the fluorescence was filtered through BP500-530 and IR LP560 and detected by a photomultiplier tube. The field of observation was 206 × 206 μm. In the small population assay, the applied cAMP diffuses in the agar, leading to a transient cAMP gradient at the cells. The maximal cAMP concentration and the absolute spatial gradient are described as where is the radius of the cAMP droplet (150 μm), is the distance between cell population and cAMP source (250 μm), and is the applied cAMP concentration in the droplet. The absolute cAMP gradients at 50 and 1,000 nM cAMP are ∼100 and 2,000 pM/μm, respectively. When a pipette filled with cAMP is inserted in a field of cells, cAMP will diffuse continuously from the pipette, leading within 1 min to a stable spatial gradient. The concentration and the spatial gradient are dependent on the distance () from the pipette according to where is the cAMP concentration in the pipette and α is a proportionality constant that depends on the geometry of the pipette and the applied pressure. The formation of the cAMP gradient was deduced by measuring the release of the fluorescent dye Lucifer yellow (mol wt = 457 D) from the pipette with the confocal fluorescent microscope and calibrated using the fluorescence intensity of diluted Lucifer yellow added homogeneously to the bath. The experiments yield α = 0.05 and demonstrate that the equations are accurate descriptions of the cAMP gradient at a distance >15 μm from the pipette; at shorter distances, more complex equations are required. At 100, 50, and 20 μm from the pipette, the absolute cAMP gradient is 500, 2,000, and 12,500 pM/μm, respectively. Table S1 presents the chemotaxis data obtained with the small population assay of ∼70 conditions with different combinations of cell strains and inhibitors. Video 1 shows chemotaxis of wild-type AX3 cells toward a pipette with cAMP in buffer. The PLA2 inhibitor BPB is added at 4 min, and the PI3K inhibitor LY294002 is added at 12 min. Video 2 shows the localization of PHcrac-GFP (detecting PIP) at the leading edge of wild-type cells chemotaxing toward cAMP. Video 3 shows -null cell movement toward a pipette with cAMP. Online supplemental material is available at .

Neuritogenesis is an essential event in making the complex architecture of neuronal networks. Initially, the original round shape of neurons is broken down to make buds. Thereafter, some protrusions are selected and stabilized into neurites. These morphological changes are accompanied by the cytoskeletal reorganization of actin and microtubules (). Rho-family GTPases (RhoA, Rac1, and Cdc42) play central roles in cytoskeletal regulation involved in a range of cellular functions (; ). Also, they are committed to neuronal morphogenesis, including axon growth and guidance, dendritic elaboration, and the formation of synapses (; ). Our previous studies using fluorescence resonance energy transfer (FRET)–based probes have shown that Rac1 and Cdc42 are locally and repetitively activated at protruding sites during neurite outgrowth in NGF-stimulated PC12 cells and in sensory neurons (; ), suggesting that their local activation is required for neurite outgrowth. Phosphatidylinositol 3-kinase (PI3-kinase) has been shown to be required for NGF-induced neurite outgrowth of PC12 cells () and sufficient to induce neurite extension in PC12 (; ) and SH-SY5Y cells (). Although PI3-kinase is a multifunctional signaling molecule having various effectors (), a close linkage between PI3-kinase and Rac1/Cdc42 has been reported in a wide range of morphological responses to external stimuli (; ). Moreover, a local positive feedback loop composed of Rac1/Cdc42 and PI3-kinase has been hypothesized to be responsible for the symmetry breaking and persistent activation required for morphogenesis (; ). We have previously shown that in PC12 cells, NGF/TrkA signaling drives the cycling of a positive feedback loop composed of PI3-kinase, Vav2/Vav3, and Rac1/Cdc42 at neurite tips (). On the other hand, it is argued that negative regulators define the spatiotemporal window of active signaling. Thus, the local and repetitive activation of PI3-kinase and Rac1/Cdc42 in NGF-treated PC12 cells implies the involvement of negative regulators in this signaling pathway. However, the dynamic and interdependent properties of this signaling network make it difficult to analyze these negative regulators intuitively. We have therefore used in silico simulation of the kinetic reaction (; ) as a tool to help generate hypotheses and to validate the reliability of experimental results. In this study, we demonstrate that two phosphatidylinositol trisphosphate (PIP) phosphatases, i.e., Src homology 2 domain–containing inositol phosphatase 2 (SHIP2) and phosphate and tensin homologue (PTEN), act as key molecules of negative regulation in the NGF–PIP–Rac1/Cdc42 signaling network. Depletion of SHIP2 and PTEN by RNA interference markedly potentiated NGF-induced Rac1/Cdc42 activation and PIP accumulation and substantially increased the number and the length of neurites in PC12 cells. Moreover, our computational and experimental studies showed the presence of an NGF-dependent negative feedback from Rac1 to SHIP2. This negative feedback loop has previously been unexpected. Finally, direct activation of PI3K and SHIP2 by Rac1 was indicated by acute Rac1 activation using the inducible translocation technique. We propose that SHIP2 and PTEN work coordinately with positive regulators to form the initial protrusive pattern and, after initial neuritogenesis, punctuate the PIP accumulation to maintain proper neurite outgrowth in neuronal cells. We examined the effect of shutting down TrkA activity on the regulation of PIP, Rac1, and Cdc42 in NGF-treated PC12 cells. As a control, shows the spatiotemporal change of PIP level and Rac1/Cdc42 activity after NGF stimulation. Immediately after NGF addition (<5 min), PI3-kinase, Rac1, and Cdc42 were transiently activated in broad areas at the cell periphery, where the lamellipodial extension was concurrently observed. Thereafter, the local accumulation of PIP and active Rac1 and Cdc42 was observed at the protruding tips (>10 min). Next, we inhibited TrkA phosphorylation with 10 nM K252a at 10 min after NGF addition (Fig. S1, available at ). K252a treatment immediately induced a transient decrease in Rac1 activity to below the basal level and the suppression of cell protrusion (, red line). Rac1 activity and the morphological repression returned to the basal levels within 20 min. A similar change was observed for PIP level and Cdc42 activity (, red lines). Meanwhile, without NGF pretreatment, K252a did not induce any changes in the PIP level or Rac1/Cdc42 activity (Fig. S1 D). Therefore, the K252a-induced super-suppression in NGF-treated cells indicated the presence of an NGF-dependent negative regulation of PIP, Rac1, and Cdc42. We similarly examined the effect of LY294002, a specific PI3-kinase inhibitor. 20 μM LY294002 treatment at 10 min after NGF addition decreased the PIP concentration and Rac1/Cdc42 activity to below the basal level for >20 min (, green lines). The addition of LY294002 without NGF pretreatment also suppressed the PIP concentration and Rac1/Cdc42 activity to below the basal level (Fig. S1 E). The kinetics of the LY294002-induced down-regulation of Rac1/Cdc42 activity in unstimulated cells (Fig. S1 E) was comparable to that of NGF-stimulated cells (). This observation excludes GTPase-activating proteins (GAPs) for Rac1 and Cdc42 from the potential NGF-dependent negative regulator. To obtain a clue for identifying the negative regulators of PIP, Rac1, and Cdc42, we examined the change in phosphatidylinositol bisphosphate (PIP) level after NGF stimulation. The basic structure of the FRET probe for PIP named Pippi-PIP () was identical to that of the PIP reporter Pippi-PIP/fllip-pm () except that the pleckstrin homology (PH) domain of TAPP1, which binds specifically to PIP (), was substituted for the PH domain of general receptor of phosphoinositides (GRP). Upon NGF stimulation of PC12 cells, PIP was rapidly produced and then gradually decreased (, A [top] and B [blue line]). The spatiotemporal pattern of PIP production was different from that of PIP; the increase in PIP showed a wide distribution throughout the cells and was sustained above the basal level for >30 min. K252a or LY294002 treatment at 10 min after NGF stimulation gradually decreased PIP to or below the basal level, respectively (, A [middle and bottom] and B [red and green lines]). A PI-5-phosphatase generates PIP from PIP. If this 5-phosphatase activity is NGF independent, the time course of PIP level after K252a treatment should be similar to that of PIP level. However, this was not the case. Thus, the 5-phosphatase is a candidate for NGF-dependent negative regulator of PIP. We attempted to construct a simple kinetic model ( and Fig. S2, available at ) that could reproduce the behaviors of PIP, Rac1, Cdc42, and PIP in NGF-treated PC12 cells. At first, we determined the concentration of TrkA, p85 (a regulatory subunit of PI3-kinase), Vav2, and Rac1 in PC12 cells (Fig. S2, G and H; and ). Then, we constructed the kinetic model by assuming that the putative 3-phosphatase activity on PIP was constant with and without NGF, that TrkA activated both PI3-kinase and a putative 5-phosphatase for PIP upon NGF stimulation (based on the results in and ), and that the kinetics of the 5-phosphatase for PIP (activation and inactivation rates) were slower than that of PI3-kinase (, blue lines). We simulated this minimal model numerically in silico and demonstrated that the model approximately reproduced the results of our experiments (). This implies that NGF-induced PI-5-phosphatase activity inhibited an excess production of PIP by NGF stimulation and that the constant activity of PI-3-phosphatase determined the basal level of PIP. To validate our kinetic model, we used targeted depletion of candidate phosphatases for PIP by RNA interference. Candidates for such PI-5- and PI-3-phosphatases include SHIP2 and PTEN, respectively, which are expressed in a wide range of tissues, including neuronal cells. We used a short hairpin RNA (shRNA) expression vector with the gene to allow selection of shRNA-expressing cells with puromycin. In SHIP2 or PTEN shRNA-expressing cells, 80–90% of endogenous SHIP2 or PTEN protein was depleted, respectively (). In (B–D), K252a was added to the control or knockdown cells at 10 min after NGF stimulation. In the control and PTEN-depleted cells, K252a treatment induced a transient decrease in the level of PIP and in the activity of Rac1/Cdc42 to below the basal level (, blue and green lines). However, in the SHIP2-knockdown cells, the PIP level and Rac1/Cdc42 activity decreased only up to the basal level after K252a addition (, red lines). Therefore, SHIP2 was responsible for the NGF-dependent negative regulation of PIP. The data in and Fig. S3 (A–C; available at ) confirmed that SHIP2 and PTEN had a major role in PIP production and degradation after NGF stimulation of PC12 cells, respectively. Next, we examined roles of SHIP2 and PTEN in PIP production and Rac1/Cdc42 activation after NGF stimulation. In the SHIP2 and PTEN double-knockdown cells, marked and prolonged activation of Rac1 and Cdc42 was observed upon NGF stimulation, although there was no substantial difference amongst the control, SHIP2-knockdown, and PTEN-knockdown cells (). Furthermore, the SHIP2 and PTEN double-knockdown cells exhibited a sustained lamellipodial structure and delayed localization of Rac1 activity (). However, we could not detect a further increase in PIP levels, as monitored by Pippi-PIP in the knockdown cells as compared with the control cells (). This seemed strange considering the results on Rac1/Cdc42 activity in double-knockdown cells (); thus, we assumed that the PIP production in these cells exceeded the upper limit of detection by Pippi-PIP. To overcome this difficulty, we developed another method to semiquantitatively measure the amount of PIP using EGFP-tagged GRP (for detail, see Fig. S4, available at ). indicates that the control cells showed a transient translocation of GRP-EGFP by 2 min after the addition of NGF. In contrast, the SHIP2 and PTEN double-knockdown cells demonstrated a massive and sustained translocation of GRP-EGFP upon NGF stimulation (). shows the time course for the amount of PIP measured during NGF treatment. This demonstrates that the depletion of both SHIP2 and PTEN caused the sustained overproduction of PIP (orange line). The SHIP2-knockdown cells (red line) showed no difference from the control cells (blue line), whereas in the PTEN-knockdown cells (green line), a transient PIP overproduction was observed in the early phase after NGF treatment (<10 min). To reproduce the temporal pattern of PIP production in knockdown cells (), we improved the initial model by introducing NGF-dependent positive and negative feedback loops (). In SHIP2- and PTEN-knockdown cells of , the level of PIP was transiently increased upon NGF stimulation and returned to the basal levels during the late phase (>10 min). However, this was not reproduced in our initial model (). To resolve this contradiction, we added a negative feedback loop from Rac1 to SHIP2 to our model (). In this improved model, computer simulations of SHIP2- and PTEN-knockdown cells reproduced a return to the basal level in the late phase (). A transient PIP overproduction in the early phase of NGF stimulation in PTEN-knockdown cells was also reproduced. Furthermore, the depletion of both SHIP2 and PTEN in the improved model showed an overproduction of PIP upon NGF stimulation, as observed in . Next, we attempted to provide evidence of this negative feedback from Rac1 to SHIP2. The production of PIP and PIP in PC12 cells expressing the dominant-negative mutant of Rac1 (Rac1N17) was examined. Our improved model predicted that the expression of Rac1N17 would inhibit the up-regulation of SHIP2 by Rac1, leading to a decrease in PIP production and an increase in PIP production (). We monitored the change of PIP levels in PC12 cells expressing Rac1N17 and demonstrated that PIP production induced by NGF treatment was decreased as compared with the control cells (). Further, NGF-induced PIP production was increased in the Rac1N17-expressing cells (). Interestingly, the local accumulation of PIP was not observed in Rac1N17-expressing cells (), suggesting the involvement of SHIP2 in localizing the PIP production during NGF treatment. We examined the effect of depletion of SHIP2 and PTEN on NGF-induced neurite outgrowth in PC12 cells. In the absence of NGF, depletion of SHIP2 and/or PTEN did not induce any neurites in PC12 cells (Fig. S3 D). Within 60 h of treatment with NGF, the control PC12 cells transfected with an empty pSUPER vector developed neurites (, top left). It is remarkable that depletion of both SHIP2 and PTEN caused a considerable increase in the number and the length of neurites extending from cell bodies (, bottom right). The proportions of neurite-bearing cells showed no clear difference among the control and single- and double-knockdown cells (). However, the depletion of both SHIP2 and PTEN obviously increased the fraction of cells having more than five neurites, although there was no clear difference between the control and single-knockdown cells (). This increase in the neurite number of double-knockdown cells seemed consistent with the sustained overproduction of PIP and the overactivation of Rac1 and Cdc42 during NGF treatment in these cells. We noted that differentiated PC12 cells deficient for SHIP2 had longer neurites as compared with controls (). The length of neurites in double-knockdown cells was even longer than that in the SHIP2-knockdown cells. To further explore the role of SHIP2 and PTEN on the number and the length of neurites, we varied the intervals between the transfection of shRNA vectors and NGF stimulation (Fig. S3, E–H). The decay of SHIP2 and PTEN after the transfection was determined in a preliminary experiment (Fig. S3 D). When pSUPER-SHIP2 and pSUPER-PTEN were transfected on the day of NGF stimulation (KD 0 d), neither the number nor the length of neurites did not increase, whereas the number and the length of neurites increased substantially when the shRNA vectors were transfected 2 or 3 d before NGF stimulation (KD 2 and 3 d; Fig. S3, F–H). Intriguingly, the transfection of the shRNA vectors 1 d before NGF stimulation increased the neurite length without affecting its number (Fig. S3, F–H). These observations indicate that the increase in PIP and the following cytoskeletal changes observed immediately after NGF stimulation play a pivotal role in the determination of the number of neurites in the differentiated PC12 cells. We investigated Rac1-mediated PI3-kinase activation by using the inducible translocation technique to rapidly activate Rac1 (). Rapamycin triggers the heterodimerization of a Lyn N-terminal sequence-tagged FRB (LDR) with an FKBP-fused protein. FKBP-Tiam1 translocated to the plasma membrane after rapamycin addition and activated Rac1 within 2 min (, left). However, we could not detect a rapamycin-triggered production of PIP (, middle) or PIP (, right). These findings clearly demonstrated that Rac1 activation alone is not sufficient for PI3-kinase activation and argue against a positive feedback loop composed of Rac1 and PI3-kinase. Therefore, we next treated the PC12 cells expressing LDR, FKBP-Tiam1, and the FRET probe (Raichu-Rac1, Pippi-PIP, or Pippi-PIP) with rapamycin at 15 min after NGF addition. Again, Rac1 was rapidly activated by rapamycin-triggered Tiam1 translocation within 2 min (, C [top] and D [left]). Of note, PIP was also increased by the rapamycin-triggered activation of Rac1 within 3 min (, middle). This provided direct evidence of an NGF-dependent positive regulation from Rac1 to PI3-kinase. Although rapamycin-triggered Rac1 activation was prolonged for >15 min (, top), rapamycin- triggered PIP production was transient (, middle). Meanwhile, PIP was gradually produced by rapamycin-triggered Rac1 activation (, C [bottom] and D [right]). To explain this observation in our kinetic simulation model, we integrated the positive feedback loop from Rac1 to PI3-kinase and its dependency to NGF by assuming that the production of PIP is partly catalyzed by the complex of Rac1 and phosphorylated PI3-kinase. According to the rapid activation of Rac1 at 15 min after NGF addition in silico (, left, red line), PIP was generated via the positive feedback from Rac1 to PI3-kinase and then degraded by the negative feedback from Rac1 to SHIP2, which gradually produced PIP (, middle and right, red lines). These computational and experimental studies together supported the existence of NGF-dependent positive and negative feedback loops in PC12 cells (). Finally, we investigated the mechanism of determining neurite-budding sites from cell bodies. During the chemotaxis of , the combination of a PIP-dependent positive feedback at the cell front and localization of PTEN at the side and back of the cell has been shown to establish cell polarity (; ). In PC12 cells, Vav2 and Vav3 were translocated to the plasma membrane in response to NGF (). Thus, we examined the localization of SHIP2 and PTEN during the neuritogenesis. Upon NGF stimulation, SHIP2 was rapidly recruited to lamellipodia and protrusions, whereas PTEN did not show any translocation (Fig. S5, A–F, available at ). This observation indicates that the mechanism of initial neurite budding in PC12 cells is different from that of the polarity formation in . On the other hand, the existence of both positive and negative feedback loops in the induction of neurites raises the possibility that the polarity of PC12 cells may be generated by Turing's reaction-diffusion system (); the combination of local positive feedback loop and long-range negative feedback loop generates a spatial pattern autonomously (; ). Therefore, we examined the diffusion rate of a positive regulator, Vav2, and negative regulators, SHIP2 and PTEN, in the NGF-induced protrusions, where these feedbacks were operating. We found that the diffusion of Vav2 was slower than those of SHIP2 and PTEN (; and Fig. S5, G and H), supporting the model that the polarity of PC12 cells was generated by Turing's reaction-diffusion system (). In this study, we have shown that the PI-5-phosphatase SHIP2 constitutes a critical component of a negative feedback loop regulating the PIP level in NGF-stimulated PC12 cells and that the PIP level in PC12 cells is primarily regulated by SHIP2 and PTEN, a constitutively active PI-3-phosphatase. This proposal is supported by the finding that the double knockdown of SHIP2 and PTEN caused a PIP overproduction, a Rac1/Cdc42 overactivation, and an increase in the number and length of neurites in NGF-treated PC12 cells, whereas no marked change was observed in SHIP2 or PTEN single-knockdown cells. Our finding is in line with previous reports showing that the depletion of PTEN by antisense oligonucleotide did not affect the NGF-induced neurite outgrowth () but that overexpression of PTEN blocks NGF-induced neurite outgrowth of PC12 cells (). Meanwhile, SHIP2 has been shown to be phosphorylated and form a complex with Shc after NGF treatment (). This finding agrees with our observation that depletion of SHIP2 impaired PIP generation upon NGF stimulation (). Based on the identification of the SHIP2-mediated negative feedback loop and our previous finding that a positive feedback loop composed of PI3-kinase, Vav2/Vav3, and Rac1/Cdc42 is operated in NGF-treated PC12 cells (), we presumed that the initial budding sites of neurites from cell bodies may be determined by the Turing's reaction-diffusion system. In this system, long-range lateral inhibition (a negative feedback loop that diffuses rapidly), in conjunction with local self-activation (a positive feedback loop that diffuses slowly), autonomously generates spatial patterns (; ). In accordance with this model, we found that the diffusion of Vav2 is slower than that of SHIP2 in the NGF-induced membrane protrusions. Therefore, NGF stimulation of PC12 cells generates the slowly diffusing positive and the rapidly diffusing negative feedback loops regulating the PIP level, which fulfill the conditions of Turing's spatial pattern formation (). We propose that this pattern formation in the PIP level and the following Rac1/Cdc42 activation contributes to the initial neuritogenesis of PC12 cells. A positive feedback loop involving PIP and Rac1 has also been demonstrated in leukocytes (; ) and (); however, because these studies used dominant-negative mutants or inhibitors, direct evidence of the positive feedback loop has not been provided. Here, we directly validated the presence of the positive feedback loop from Rac1 to PI3-kinase using the rapamycin-induced acute activation system. Intriguingly, however, we found that this positive feedback loop only operated in the presence of NGF (). Which model can reconcile these findings? Based on a previous report that the p85 regulatory subunit of PI3-kinase binds to GTP-loaded Rac1 through its RhoGAP homology domain (), we hypothesized that the active PI3-kinase is actually a complex of phosphorylated PI3-kinase and Rac1-GTP. In this way, Rac1-GTP promotes the production of PIP by two related mechanisms; Rac1-GTP increases the active complex and competitively inhibits tyrosine phosphatases. In our model integrating this assumption, we could faithfully reproduce the experimental results in silico (). To consider the mechanism of the negative feedback from Rac1 to SHIP2, the regulation of OCRL1, another PI-5-phosphatase, may provide some clues. After growth factor stimulation, OCRL1 has been shown to bind to GTP-Rac1 through its RhoGAP domain and to be translocated to the plasma membrane (). Although SHIP2 does not bind to GTP-Rac1 directly, SHIP2 binds to several proteins associated with cytoskeletal reorganization, i.e., Cas, Filamin, and Vinexin (; ; ). Because Rac1 stimulates actin assembly, it is possible that these SHIP2 binding proteins are relocated to the actin accumulation sites, accompanied by SHIP2 recruitment, where SHIP2 can then dephosphorylate PIP at these sites. In morphogenesis, similar but different mechanisms exist that coordinate multiple signaling pathways, depending on the cellular context. During the chemotaxis of , the combination of a PIP-dependent positive feedback loop at the cell front and localization of PTEN at the side and back of the cell has been shown to establish cell polarity and promote the forward movement (; ). On the other hand, the present study demonstrates the presence of PIP-dependent positive and negative feedback loops in NGF-stimulated neuritogenesis. Further differences can be pointed out. In neutrophils, the PIP-dependent positive feedback loop is independent of receptor-mediated signals because the delivery of exogenous PIP into these cells induces the formation of cell polarity (). In contrast, an NGF-TrkA signal was required for the positive feedback loop in PC12 cells (). During the elongation process, SHIP2 and PTEN down-regulated neurite protrusion (). Considering the negative feedback from Rac1 to SHIP2, the inhibitory role of SHIP2 should be prominent at the neurite tips, which showed the recurrent activation of Rac1 during neurite elongation (). In fact, the depletion of SHIP2 considerably increased the length of neurites (). The inhibitory role of the PIP phosphatases might be involved in the repeated resetting of the concentration of PIP to the basal levels to allow sensing and adapting to an environment by changing the extending direction, although neurite tips of PC12 cells have lost such guidance ability during transformation. We developed a method to measure the level of PIP semiquantitatively in living cells. FRET efficiency obtained by Pippi-PIP is dependent on both the amount of endogenous PIP and the expression level of the FRET probe. In addition, the detection range of Pippi-PIP is affected by the dissociation constant of the PH domain for PIP, like Ca indicators. In , we could not observe the enhancement of NGF-induced PIP production in the SHIP2 and PTEN double-knockdown cells, suggesting that the level of PIP in these cells exceeded the upper limit of detection of Pippi-PIP. Although many groups use the translocation of the PH domain to measure the level of phosphoinositides (; ; ), the amount of translocation strongly depends on the expression level of the PH domain and cell morphology. Further, Akt phosphorylation is used as an indicator of PIP levels. However, the level of Akt phosphorylation reflects the balance of upstream kinases and phosphatases, and not the level of PIP. Our new method of PIP quantification used in this study basically utilizes the dose dependency of PIP-induced translocation of GRP and obtains the amount of PIP as an asymptotic value. Thereby, this method overcomes the aforementioned problems. In this study, we used in silico simulation of kinetic reactions as a tool to help generate hypotheses and validate the reliability of experimental results. As shown here, combining in vivo analyses with in silico simulation helps us understand the systemic properties of signaling networks. In fact, many groups have been trying to dissect complex signaling pathways using computational simulation (; ; ). Here, we built a simple kinetic model of NGF–PIP–Rac1 signaling. Most of the parameters were estimated arbitrarily except the concentration of all proteins. And no spatial information was taken into account in this model, although many reactions in NGF–PIP–Rac1 signaling occur on the plasma membrane. Nevertheless, this model could reproduce the results of several experiments, i.e., perturbation with inhibitors (K252a and LY294002) and knockdown experiments (SHIP2 and PTEN). Therefore, we emphasize a potential of in silico analysis as a powerful tool in the research of complicated signal transduction pathways. The plasmids encoding the FRET probes, Raichu-Rac1/1011x, Raichu-Cdc42/1054x (), Pippi-PIP/fllip-pm (; ), and Pippi-PIP () have been described previously. The RNA targeting constructs were generated using pSUPER.retro.puro vector (OligoEngine). The 19-nucleotide sequences used to target rat SHIP2 and PTEN mRNAs were 5′-GGCCTACATTGAGTTTGAG-3′ and 5′-GTCAGAGGCGCTATGTATA-3′, respectively. Rat GRP cDNA was obtained by RT-PCR and subcloned into the pCXN2-FLAG vector fused with EGFP at the C terminus (). pCAGGS-ERed-NES encoded Express red (Invitrogen) fused to nuclear export sequence of HIV-1 rev protein (LQLPPLERLTLD). pIRM21-FLAG-Rac1N17 has been described previously (). The cDNAs for LDR (Lyn-targeted FRB), FKBP, and FKBP-Tiam1 () were prepared by PCR and subcloned into pERedNLS and pCAGGS-3HA (). The cDNAs for TrkA, p85, Vav2, SHIP2, PTEN, and Rac1 were subcloned into pCAGGS-EGFP and pCXN2-Flag-EGFP. The cDNA for PA-GFP (), which was provided by T. Nagai (Hokkaido University, Hokkaido, Japan), was used to prepare pCAGGS-Flag-Vav2-PAGFP, pCAGGS-Flag-SHIP2-PAGFP, and pCAGGS-Flag-PTEN-PAGFP. PC12 cells were maintained in RPMI (Invitrogen) supplemented with 10% horse serum and 5% fetal bovine serum. The cells were plated on 35-mm glass-based dishes (Asahi Techno Glass), which were coated with polyethyleneimine (Sigma-Aldrich). NGF and K252a were purchased from Calbiochem. LY294002 and puromycin were obtained from Sigma-Aldrich. Rapamycin was purchased from LC Laboratories. Fluorescent microsphere for calibration of EGFP proteins was a gift from S. Okabe (Tokyo Medical and Dental University, Tokyo, Japan; ). Anti-GFP rabbit serum was prepared in our laboratory (). Anti-Trk (I-20), anti-Vav2 (H-200), and anti-SHIP2 (I-20) polyclonal antibodies were purchased from Santa Cruz Biotechnology, Inc. Anti-PTEN, anti-Akt, anti–phospho-Akt (Thr308), and anti–phospho-TrkA (Tyr490) were obtained from Cell Signaling Technology, and anti-tubulin monoclonal antibody was obtained from Calbiochem. Anti-Rac1, anti- Cdc42, and anti-p85 monoclonal antibodies came from BD Biosciences. PC12 cells were transfected with the desired pSUPER constructs by using Lipofectamine 2000 (Invitrogen). After recovery, the cells were selected by a 2-d incubation with 3 μg/ml puromycin and then used for further analysis. For FRET imaging, the indicated pRaichu plasmids were transfected into the shRNA-expressing cells 1 d after the addition of puromycin as described. After an additional 1-d incubation with puromycin, the cells were starved and used for imaging. PC12 cells expressing FRET probes were starved for 6–12 h with phenol red–free DME/F12 medium containing 0.1% BSA and then treated with 50 ng/ml NGF. The medium was covered with mineral oil (Sigma-Aldrich) to preclude evaporation. Cells were imaged with an inverted microscope (IX81; Olympus) at 37°C, equipped with a cooled charge-coupled device camera (Cool SNAP-HQ; Roper Scientific), a laser-based auto-focusing system (IX2-ZDC; Olympus), and an automatically programmable XY stage (MD-XY30100T-Meta; SIGMA KOKI), which allowed us to obtain time-lapse images of several view fields in a single experiment. The filters used for the dual-emission imaging were obtained from Omega Optical: an XF1071 (440AF21) excitation filter, an XF2034 (455DRLP) dichroic mirror, and two emission filters (XF3075 [480AF30] for CFP and XF3079 [535AF26] for FRET). Cells were illuminated with a 75-W Xenon lamp through a 12% ND filter and viewed through a 60× oil-immersion objective lens (PlanApo 60×/1.4). The exposure times for 4 × 4 binning were 400 ms for CFP and FRET images and 100 ms for differential interference contrast images. After background subtraction, FRET/CFP ratio images were created with MetaMorph software (Universal Imaging Corp.), and the images were used to represent FRET efficiency. FRET/CFP ratio images were shown after normalization as follows. First, in each sample, we determined the mean ratio over the whole cell before NGF addition and used that ratio as the reference value. Then, the raw FRET/CFP ratio of each pixel was divided by the reference value, and this normalized value was used to generate a normalized ratio image. PC12 cells were transfected with the indicated pSUPER constructs and selected with 3 μg/ml puromycin for 2 d. Then, neurite outgrowth was stimulated with 50 ng/ml NGF and allowed to proceed for 60 h in DME/F12 medium containing 0.1% BSA and 3 μg/ml puromycin. Quantification of neurite outgrowth was performed as described previously (). PC12 cells on polyethyleneimine-coated 35-mm glass-based dishes were transfected with PA-GFP fusion plasmids. After a 36-h incubation, the cells were serum starved for 6 h and stimulated with 50 ng/ml NGF. The cells were imaged with a confocal microscope (FV1000; Olympus) equipped with an argon laser, a 405-nm laser diode, and a 60× oil-immersion objective lens (UPlanSApo). For photoactivation, a region of interest (3-μm diameter) in an NGF-induced protrusion was illuminated with 405-nm laser light. Images were obtained every 0.065 s by illuminating with the argon laser. The filters used were a DM405/488 dichroic mirror and an emission filter (BA505IF; Olympus). All reactions were represented by linear molecule–molecule interactions and enzymatic reactions. The biochemical reactions and the rate constants used in this study are shown in Fig. S3 and Table S1 (available at ), respectively. simulator (version 2.2) with a interface (version 9.0) was used for solving the ordinary differential equations with a time step of 100 ms as described previously (; ). Fig. S1 shows the NGF-induced phosphorylation of TrkA in various conditions and the effect of K252a or LY294002 treatment on Rac1/Cdc42 activity and the levels of PIP and PIP in the absence of NGF. Fig. S2 shows the schematic representation of all reactions in the model of the NGF–PIP–Rac1 signaling pathway and the quantification of endogenous TrkA, p85, Vav2, SHIP2, PTEN, and Rac1. Fig. S3 shows the NGF-induced PIP production in control, SHIP2- or PTEN-depleted PC12 cells, and the effect of depletion of SHIP2 and PTEN on neurite outgrowth of PC12 cells. Fig. S4 depicts the method of semiquantitative measurement of intracellular PIP levels. Fig. S5 demonstrates the subcellular localization of SHIP2 and PTEN in PC12 cells stimulated with NGF and the different diffusion rate among Vav2, SHIP2, and PTEN. Table S1 describes the parameters of the molecule–molecule interactions in the computational model. Table S2 describes the parameters of the enzymatic reactions. Table S3 describes the parameters of transitions. Table S4 describes the initial concentration of NGF and PIP in the computational model. The supplemental text gives details of the quantification of endogenous TrkA, p85, PTEN, SHIP2, Vav2, and Rac1 and the semiquantitative measurement of intracellular PIP level. Online supplemental material is available at .

Dendritic spines are small protrusions from the dendrite that form the postsynaptic component of excitatory synapses. Filopodia are recognized as one origin of dendritic spines (for reviews see ; ). During early stages of synaptogenesis, filopodia rapidly protrude and retract from dendrites. When dendritic filopodia contact presynaptic sites and form synapses, filopodia contract and transform into dendritic spines. Many transmembrane receptors and intracellular molecules have been shown to play a role in spinogenesis (for reviews see ; ; ; ; ), including syndecan-2. Syndecan-2 belongs to the syndecan family of transmembrane heparan sulfate proteoglycans. By virtue of their heparan sulfate modifications, syndecans act as coreceptors for growth or differentiation factors, presenting these molecules to specific receptor tyrosine kinases, including the fibroblast growth factor receptors (). Syndecans also function as adhesion molecules that regulate cell migration, cell–cell interactions, and cell–extracellular matrix interactions (; ; ). During neural development, syndecan-2 expression is elevated during synaptogenesis (; ). The overexpression of syndecan-2 starting at 1 d in vitro (DIV) accelerates spine formation in hippocampal neurons examined at 8 DIV (), suggesting a role of syndecan-2 in spinogenesis. Because syndecan-2 overexpression also promotes filopodia formation in nonneuronal cell lines such as COS-1 and Swiss 3T3 (, ), it is possible that syndecan-2 first promotes filopodia formation and, consequently, transforms filopodia into dendritic spines in neurons. As yet, the molecular mechanism underlying the effect of syndecan-2 on cytoskeleton rearrangement remains unclear. Although the cytoplasmic domain of syndecan-2 is short (∼30 residues) and has no kinase domain, several syndecan-2–interacting proteins have been identified whose activity may provide clues about syndecan-2 signaling. The cytoplasmic domain of syndecan-2 consists of three small regions: two highly conserved regions (C1 and C2) and, between these, a variable (V) region unique to each syndecan. The C2 region contains a type II PDZ-binding motif (residues E-F-Y-A; ; ). This EFYA motif is important for syndecan-2– dependent dendritic spine formation, and syndecan-2 loses the ability to promote spinogenesis when the C2 motif is removed (). Several adaptor proteins such as syntenin, calcium/CaM-dependent serine protein kinase (CASK), synbindin, and synectin all bind to the EFYA motif of syndecans (; ; ; ; ), suggesting that these interactions play a role in synaptic formation. Another syndecan-2–interacting protein is neurofibromin (), which is encoded by the () gene and interacts with the C1 region of syndecan-2. 40–60% of NF1 patients are characterized as having specific learning disabilities (for reviews see ; ). Mice carrying a heterozygous null mutation of the gene also show several features of the learning deficits associated with mutations in humans (for review see ; ). These studies indicate an important role of neurofibromin in neuronal function. At the molecular level, neurofibromin possesses a central Ras GTPase-activating protein–related domain that regulates the Ras–MAPK pathway (for reviews see ; ). In addition, neurofibromin is also involved in the cAMP pathway via the regulation of adenylyl cyclase through two distinct pathways (; ; ). One is the receptor tyrosine kinase pathway, which acts independently of any heterotrimeric G protein; Ras activation by neurofibromin is essential for this pathway. The other is the classic heterotrimeric G-protein pathway, which is Gα dependent and requires the C-terminal region of neurofibromin (). In this study, we elucidate the role of these intracellular interactions of syndecan-2 in neuronal morphogenesis. Filopodia formation in nonneuronal cells was chosen here as a model to study the early downstream signaling of syndecan-2. The common signaling of the syndecan-2–neurofibromin–PKA–Enabled (Ena)/vasodilator-stimulated phosphoprotein (VASP) pathway leading to filopodia formation and spinogenesis was then studied in cultured hippocampal neurons. Our study provides the first evidence that neurofibromin is required for dendritic spine formation, which may explain how mutation leads to deficits in learning and memory. To confirm that syndecan-2 is important for dendritic spine formation, we used an RNAi approach to reduce neuronal syndecan-2 protein levels. First, we examined the ability of syndecan-2 small hairpin RNA (shRNA) to knock down syndecan-2. Syndecan-2 shRNA but not vector control (SUPER.neo+GFP) efficiently down-regulated syndecan-2 protein expression in both human embryonic kidney (HEK) 293T cells and cultured hippocampal neurons (Fig. S1, A and B; available at ). The effect of syndecan-2 shRNA on dendritic spine formation was then investigated. Constructs expressing syndecan-2 shRNA (also expressing GFP) and GFP–actin were cotransfected into hippocampal neurons at 11 DIV. Morphology of dendritic protrusions was then assessed by GFP and GFP-actin signals. At 16 DIV, dendritic spines in our cultures had differentiated and exhibited characteristics of mature spines (, inset). In the presence of syndecan-2 shRNA, the number of dendritic protrusions representing dendritic spines was greatly reduced (). To ensure the sequence-specific effect of syndecan-2 shRNA, a syndecan-2 silent mutant insensitive to syndecan-2 shRNA was coexpressed with syndecan-2 shRNA (). Overexpression of the silent mutant did rescue the effect of syndecan-2 shRNA on protrusion number (). In addition, mature spines were observed along the dendrites in the presence of syndecan-2 mutant (, inset). These results support a critical role for syndecan-2 in dendritic spine formation. Because dendritic filopodia have been proposed to be precursors of dendritic spines and syndecan-2 has been shown to induce filopodia formation in nonneuronal cells (, ), it is likely that the overexpression of syndecan-2 in cultured hippocampal neurons first induces filopodia formation and then promotes dendritic spine maturation. Indeed, when syndecan-2 was transfected into cultured hippocampal neurons at 1 DIV, numerous filopodia emerging from dendrites were observed at 4–5 DIV (). Because filopodia formation induced by syndecan-2 also occurs in nonneuronal cells, this initial process involved in dendritic spinogenesis does not appear to be neuron specific, although the transformation from filopodia to spines should be specific for neurons. To explore the signaling downstream of syndecan-2 that initiates spine formation, we first used filopodia formation in nonneuronal HEK cells as a model system for syndecan signaling and confirmed these observations in cultured hippocampal neurons. To monitor syndecan-2 expression and cell morphology of transfected cells, we generated the syndecan-2 antibody syndecan-2G. This antibody was confirmed as recognizing syndecan-2 (Fig. S2, A and B; available at ) but not syndecan-1, -3, or -4 (Fig. S2 A) in both immunoblotting and immunostaining experiments in transfected COS cells, confirming its high syndecan-2 specificity. Generally, there are very few filopodia on the surface of parental HEK cells (<0.2 filopodia per μm; ). When syndecan-2 was overexpressed in HEK293T cells, numerous filopodia were revealed by syndecan-2G antibody (). Cotransfection of syndecan-2 and GFP into HEK cells was also performed (unpublished data). Perhaps as a result of staining of the plasma membrane, we found that syndecan-2 signal outlined the morphology of HEK cells more clearly than GFP signal. To explore the downstream signaling of syndecan-2, several kinase inhibitors were added to cultures overexpressing syndecan-2, including the phosphatidylinositol 3-kinase (PI3K) inhibitors LY294002 and wortmannin, PKC inhibitors Go6976 and Go6850, and PKA inhibitors KT5720 and H89. Neither PIK3 nor PKC inhibitors prevented filopodia formation induced by syndecan-2 (Fig. S3, available at ), suggesting that syndecan-2–induced filopodia formation is independent of the PI3K–Akt and PKC pathways. In contrast, both PKA inhibitors KT5720 and H89 decreased syndecan-2–induced filopodia formation (), although the effect of KT5720 was weaker than that of H89 for unknown reasons. To further confirm that PKA is involved in syndecan-2 signaling, a construct expressing the PKA-specific peptide inhibitor PKI was cotransfected with syndecan-2 into HEK cells. The presence of PKI significantly reduced the filopodia density on the surface of syndecan-2–expressing HEK293T cells (P < 0.001; ), supporting the idea that PKA signaling is important for syndecan-2–induced filopodia formation in HEK293T cells. Results for cultured hippocampal neurons were similar to those in HEK293T cells: PKA inhibitor H89 efficiently reduced the number of syndecan-2–induced filopodia along neurites (). These results indicated that syndecan-2 induces dendritic filopodia formation via PKA in cultured hippocampal neurons. A series of C-terminal mutants of syndecan-2 () were then used to map the cytoplasmic regions of syndecan-2 that are required for filopodia formation. Mutant genes were transfected into HEK293T cells, and immunoblotting with syndecan-2G antibody confirmed that expression levels of these mutants were comparable with those of wild-type syndecan-2 (Fig. S2 C). In addition, these deletions did not considerably affect the plasma membrane targeting of mutants (unpublished data). The effects of these mutants on filopodia formation in HEK293T cells were then examined. The mutant syndecan-2Δ3 lacking the last three residues of syndecan-2 successfully induced normal filopodia formation (). The ability of the mutant syndecan-2Δ20 lacking the cytoplasmic V and C2 regions to induce filopodia was slightly weaker than that of wild-type syndecan-2 (). When the entire cytoplasmic region of syndecan-2 was removed in the syndecan-2Δ32 mutant, the ability to form filopodia was greatly impaired (). In the syndecan-2ΔC1 mutant lacking only the C1 region, the ability to promote filopodia formation was also greatly reduced (). These results suggested that both the C1 and V regions (variable region of syndecan) of syndecan-2 are involved in filopodia formation in HEK cells, with the C1 region being the most critical. To confirm the role of the syndecan-2 C1 region in dendritic filopodia formation, wild-type syndecan-2, -2ΔC1, and -2Δ3 mutants were transfected into cultured hippocampal neurons. Compared with wild-type syndecan-2, the number of dendritic filopodia was greatly reduced in neurons expressing syndecan-2ΔC1 (). In contrast, syndecan-2Δ3 still efficiently induced dendritic filopodia formation (). These data supported the notion that the C1 region is also critical for syndecan-2–induced filopodia formation in neurons. Because the C1 region of syndecan-2 is the most important region for neurofibromin interaction and neurofibromin is involved in both neurite outgrowth () and cAMP signaling, we investigated whether the interaction of neurofibromin with the C1 region of syndecan-2 mediates signaling from syndecan-2 to PKA. Because mutation of the RKKD motif in the C1 region of syndecan-2 impairs neurofibromin interaction (), we explored this possibility by examining the effects of the alanine replacement mutants syndecan-2 RK/AA and KD/AA on filopodia formation. Indeed, the ability of syndecan-2 RK/AA and KD/AA mutants to induce filopodia formation was much weaker than that of wild type (), suggesting that an interaction with neurofibromin is involved in filopodia formation induced by syndecan-2. In addition to characterizing filopodia density, we also dynamically analyzed filopodia behavior using time-lapse techniques. For syndecan-2–induced filopodia, the motility of filopodia was generally very low. The length of most filopodia did not change substantially during the time recorded (around 10 min; ): only a small fraction of filopodia was altered in length, with a mean velocity of 1.55 ± 0.42 μm/min (). In contrast, the motility of syndecan-2ΔC1–expressing cells was higher. Syndecan-2ΔC1–induced filopodia frequently extended and withdrew at greater amplitudes () and at a mean velocity of 1.86 ± 0.39 μm/min (). The majority (12 out of 15) of syndecan-2ΔC1–induced filopodia extended or withdrew their tips to lengths >1 μm within 6 min (, middle; red lines). In contrast, only the minority (3 out of 12) of mobile syndecan-2–induced filopodia exhibited this ability (, top; red lines). When forskolin (FSK), which increases cAMP levels, was added into culture transfected with syndecan-2ΔC1, filopodia movement was immediately frozen (); only a small fraction still extended or withdrew with a mean velocity of 1.29 ± 0.26 μm/min (), and the movement amplitude was <1 μm within 6 min (, bottom). These results indicated that syndecan-2–induced filopodia are very stable and that the PKA pathway may play an important role in stabilizing formed filopodia. The relative instability of syndecan-2ΔC1–induced filopodia may account for the lower filopodia density shown in . The aforementioned experiments demonstrated that syndecan-2–induced filopodia formation was prevented by the addition of PKA inhibitors. Thus, it is possible that the overexpression of syndecan-2 activates PKA and subsequently promotes filopodia formation. To test this possibility, we compared PKA activity (measured by ELISA) between cells transfected with syndecan-2 or with control vector. Syndecan-2 overexpression in HEK293T cells resulted in 20–30% increases in total PKA activity compared with the vector control (). The mutant syndecan-2Δ32, which lacked the entire cytoplasmic domain of syndecan-2, lost the ability to enhance PKA activity (). The syndecan-2ΔC1 mutant missing the critical C1 region was also unable to activate PKA (). These results support a model for syndecan-2 activation of PKA via its cytoplasmic C1 region, culminating in filopodia formation. To confirm that neurofibromin mediates signaling from syndecan-2 to PKA, leading to filopodia formation, we used two approaches. The first approach was to interrupt the interaction between neurofibromin and syndecan-2 by overexpression of the syndecan-2–interacting domain of neurofibromin, the Jn fragment (residues 1,356–1,473), which was identified in a yeast two-hybrid assay (). Interaction between syndecan and neurofibromin Jn fragments was first confirmed under mammalian cell culture conditions using coimmunoprecipitation (Fig. S1 D). The effect of the Jn fragment on syndecan-2–induced filopodia formation was then examined. When the Jn fragment was overexpressed, filopodia formation induced by syndecan-2 was greatly reduced in HEK293T cells (). More important, the Jn fragment also inhibited filopodia formation in cultured hippocampal neurons (), suggesting an essential role for the interaction between syndecan-2 and neurofibromin in syndecan-2–induced filopodia formation. Overexpression of the Jn fragment alone did not cause an obvious morphological abnormality of cultured neurons. Neither dendrite number nor the length of dendrites was affected by the Jn fragment (unpublished data), supporting the specific effect of the Jn fragment on syndecan-2–induced filopodia formation. If interrupting the interaction between syndecan-2 and neurofibromin by adding the Jn fragment prevents PKA activation and, thus, blocks filopodia formation, it can be predicted that increasing PKA activity in Jn-expressing cells should restore filopodia formation. Indeed, the addition of FSK restored filopodia formation impaired by the Jn fragment in both HEK293T cells () and cultured hippocampal neurons (). The second approach to investigate the role of neurofibromin in syndecan-2–induced filopodia formation was to down-regulate endogenous neurofibromin expression by RNAi. We first examined the down-regulation of endogenous NF1 expression using an NF1 shRNA construct (Fig. S1 C). When syndecan-2 and NF1 shRNA construct were cotransfected into HEK cells, the filopodia density was significantly reduced (P < 0.001; ). Similarly, the expression of NF1 shRNA in cultured hippocampal neurons impaired dendritic filopodia formation compared with nonsilencer control (). Again, to confirm that the effect of NF1 shRNA is to block the signaling from syndecan-2 to the cAMP–PKA pathway, FSK was added into the cultures transfected with syndecan-2 and NF1 shRNA construct. Adding FSK completely restored the filopodia formation in both HEK cells and hippocampal neurons (). These results supported the notion that neurofibromin mediates signaling to the cAMP–PKA pathway that is required for syndecan-2–induced filopodia formation in neurons. The aforementioned study demonstrated that PKA is the important downstream effector of syndecan-2 on the induction of filopodia formation. We then wondered whether the activation of PKA alone is sufficient for filopodia formation in cultured hippocampal neurons. To address this point, FSK and dibutyryl cAMP were added into young hippocampal neurons (5 DIV) without the overexpression of syndecan-2. The results showed that FSK and dibutyryl cAMP could not induce filopodia formation in the absence of syndecan-2 (), suggesting that although PKA is required for syndecan-2–induced filopodia formation, activation of PKA alone is not sufficient for filopodia formation. Multiple signaling provided by syndecan-2 may be involved in filopodia formation. We then addressed how PKA conducts the signal from syndecan-2/ neurofibromin to induce cytoskeleton rearrangement. Previous studies had demonstrated that Ena/VASP proteins, which are important for the formation and elongation of filopodia, are regulated by PKA phosphorylation (; ; ). There are three related Ena/VASP proteins in vertebrates—mammalian enabled (Mena), VASP, and Ena-VASP–like (EVL)—that are highly related and can function interchangeably (; ; ). They promote actin filament elongation by interacting with barbed ends and shielding them from capping proteins (). Mena, EVL, and VASP all share a conserved PKA phosphorylation site, which is critical for the regulation of their function in actin filament elongation (). It has been shown that Mena is present at the tip of growth cone filopodia (), where it is positioned to initiate actin polymerization and promote filopodia elongation. To address whether Ena/VASP is the downstream effector of the syndecan-2–neurofibromain–PKA pathway, we first examined whether Ena/VASP proteins are present at the tips of syndecan-2–induced filopodia. GFP-Mena, -VASP, and -EVL all distributed to the tips of every single filopodia in syndecan-2–transfected HEK293T cells (Fig. S4 A, available at ) as well as cultured neurons (). These results favor the possibility that Ena/VASP proteins are involved downstream of syndecan-2. To further explore this possibility, we examined VASP phosphorylation at the PKA sites in the presence of syndecan-2. Indeed, syndecan-2 overexpression enhanced the phosphorylation levels of VASP in HEK293T cells, as revealed by immunoblotting using phosphopeptide antibodies specifically recognizing PKA-phosphorylated VASP (). This data supported the activation of VASP by syndecan-2 overexpression. To address the role of Ena/VASP in syndecan-2–induced filopodia formation, FP4-mito and AP4-mito constructs were coexpressed with syndecan-2 in cells. FP4-mito is a fusion containing EGFP, four binding motifs for Ena/VASP proteins, and a mitochondria target sequence (). This fusion binds and mistargets Ena/VASP protein to mitochondria instead of plasma membrane, reducing the filopodia formation activity of Ena/VASP proteins (; ). AP4-mito, which contains a mutation in the Ena/VASP-binding motif and fails to interact with Ena/VASP proteins, was used as a negative control. In both HEK293T cells (Fig. S4 B) and cultured hippocampal neurons (), the expression of FP4-mito reduced syndecan-2–induced filopodia formation. These results supported the notion that Ena/VASP proteins are the downstream mediators of syndecan-2. The addition of FSK into the culture cotransfected with syndecan-2 and FP4-mito did not prevent the blocking effect of FP4-mito in HEK293T cells (Fig. S4 B) or in cultured hippocampal neurons (), supporting the idea that PKA works upstream of Ena/VASP. The aforementioned results showed that syndecan-2 promotes filopodia formation via an NF1–PKA–Ena/VASP pathway in both HEK293T cells and cultured hippocampal neurons. To further elucidate the involvement of this signaling pathway in dendritic spine formation, we performed three more experiments. First, we examined the requirement of the cytoplasmic region of syndecan-2 in spine formation. Wild-type syndecan-2, -2ΔC1, and -2Δ3 mutants were transfected into cultured hippocampal neurons at 1 DIV, and their abilities to induce dendritic spine formation were examined at 8–9 DIV. Consistent with previous observations (), wild-type syndecan-2 induced dendritic spine formation at 8–9 DIV (). These spines formed functional synapses because they made contact with presynaptic buttons, as revealed by presynaptic marker synaptophysin staining (). For neurons transfected with syndecan-2Δ3, the protrusion density was not significantly different from that of wild-type syndecan-2 (P = 0.387; ). However, the majority of the protrusions of syndecan-2Δ3–expressing neurons still carried the characteristics of filopodia (being longer than 2 μm and lacking a head; ), supporting the idea that syndecan-2Δ3 can promote filopodia formation; however, the consequent transformation from filopodia to spines was prevented. For syndecan-2ΔC1, the density of the protrusion was much less than that induced by wild-type syndecan-2 (). This outcome was similar to the effect of syndecan-2ΔC1 on filopodia formation at 4–5 DIV (). These results support the notion that the C1 region of syndecan-2 is required for both filopodia and spine formation and that the C2 region is critical for spine formation. The fact that the protrusion densities induced by syndecan-2 and -2Δ3 were similar () also supports the hypothesis that filopodia are the intermediates of spines in this syndecan-2–dependent process. Next, we examined the involvement of neurofibromin and Ena/VASP proteins in spine formation by again expressing NF1 shRNA and FP4-mito constructs, respectively, in hippocampal neurons at 11–12 DIV. The spine density of mature neurons was then examined at 16–17 DIV. Compared with nonsilencer control, the NF1 shRNA construct significantly reduced spine density along dendrites of mature hippocampal neurons (P < 0.001; ), supporting the role of neurofibromin in dendritic spine formation. In the investigation targeting Ena/VASP proteins, expression of the FP4-mito construct also inhibited spine formation in mature hippocampal neurons (), supporting an involvement of Ena/VASP proteins in spine formation. In conclusion, these results indicated that neurofibromin and Ena/VASP proteins contribute to dendritic spine formation, perhaps through the regulation of filopodia formation. Using HEK cells and cultured hippocampal neurons as models, we have demonstrated that the neurofibromin–PKA–Ena/VASP pathway mediates the downstream signaling of syndecan-2. This study shows that syndecan-2 itself can deliver a signal to cells to change cell morphology. Our results also provide the first evidence that both neurofibromin and Ena/VASP proteins are involved in dendritic spine formation. For cell signaling, syndecans are first identified as coreceptors for growth and differentiation factors. However, the previous studies indicated that syndecans are capable of activating intracellular signaling pathways directly via their short cytoplasmic domains. The first example of this was seen with syndecan-4; overexpression of this protein in CHO cells promotes focal adhesion assembly (). The V region of syndecan-4, which binds to phosphatidylinositol-4,5-bisphosphate and activates PKCα, is critical for this function (). Another study showed that the pleiotrophin/heparin-binding growth-associated molecule, which promotes neurite outgrowth in developing neurons, binds to syndecan-3 to induce a syndecan-3–Src–cortactin interaction (). In the present study, we have demonstrated a third signaling pathway involving a direct interaction of syndecan-2 with the neurofibromin–PKA–Ena/VASP pathway. Overexpression of syndecan-2 induces filopodia formation via PKA activity, which is mediated by neurofibromin, and thus regulates the function of Ena/VASP proteins in filopodia formation. In contrast to the heparin-binding growth-associated molecule–syndecan-3 pathway, the need for a specific ligand for the activation of syndecan-2 has not been clarified. In our system, overexpression of syndecan-2 is sufficient to induce downstream signaling; however, it cannot be ruled out that an unknown syndecan-2 ligand is present in our culture systems. Currently, it is also unclear how the overexpression of syndecan-2 enhances the ability of neurofibromin to activate PKA. Perhaps interaction with syndecan-2 changes the conformation of neurofibromin, allowing the activation of adenylyl cyclase. Because syndecans form dimers (and perhaps multimers) via their transmembrane domains and there are two separate syndecan-2–interacting sites in neurofibromin (), a single neurofibromin molecule may simultaneously interact with a syndecan-2 dimer. This interaction might fix the neurofibromin molecule in a conformation that favors the activation of adenylyl cyclase. Therefore, increases in the level of syndecan-2 might fix more neurofibromin molecules in a conformation that activates adenylyl cyclase. More investigation is needed to explore this possibility. The deletion analyses showed that syndecan-2Δ32 and -2ΔC1 mutants were unable to promote filopodia formation. Retention of mutants in the cytoplasm cannot explain this phenotype. Our syndecan-2G antibody recognized the ectodomain of syndecan-2 (unpublished data). It stained both syndecan-2– and -2Δ32–expressing cells in the absence of permeabilization (unpublished data), indicating the normal surface expression of syndecan-2Δ32. Moreover, the expression levels of both syndecan-2ΔC1 and -2Δ32 were comparable with that of wild-type syndecan-2 (Fig. S2 C). The loss of filopodia formation with these mutants is also unlikely to be the result of defects in protein expression or subcellular distribution. It suggests that the regions missing in these deletion constructs are important for filopodia formation. The current findings indicate that the PKA pathway is essential for syndecan-2–induced filopodia formation. However, PKA activation itself is not sufficient for filopodia formation. It suggests that syndecan-2 overexpression is likely to activate multiple signaling pathways that are all required for dendritic spine formation. Consistent with this speculation, the Ras pathway, which is also regulated by neurofibromin, has been shown to contribute to dendritic spine formation (; ; ). It is possible that the Ras and PKA pathways downstream of syndecan-2/neurofibromin coordinate and regulate dendritic spinogenesis. Alternatively, the ectodomain of syndecan-2 involved in cell–cell or cell–matrix interaction may also be essential for filopodia outgrowth. Without adhesion via the syndecan-2 ectodomain, filopodial protrusions may not be stable. The observation that overexpression of the entire cytoplasmic domain of syndecan-2 failed to promote filopodia formation (unpublished data) also supports the idea that the ectodomain of syndecan-2 is required in this process. More investigations are required to resolve this issue. Although the cytoplasmic domain of syndecan-2 is short, it interacts with several proteins. Even the complex formation of neurofibromin, syndecan, and CASK has been shown. However, it is unclear whether there is any cross talk between CASK and neurofibromin. CASK apparently has no influence on the filopodia formation activity of neurofibromin because deletion of the CASK-binding site (EFYA motif) on syndecan-2 did not affect filopodia formation. Instead of being part of the same process, it seems likely that the complex formation of CASK, syndecan-2, and neurofibromin serves to achieve two sequential processes: filopodia formation and dendritic spine maturation. The interaction of syndecan-2 with neurofibromin initiates filopodia formation, and the interaction with CASK may further transform filopodia to dendritic spines. The results of the analysis of cytoplasmic deletion mutants of syndecan-2 in filopodia and spine formation support this speculation. demonstrated that EphB2 receptor tyrosine kinase phosphorylates the tyrosine residues Y189 and Y201 in the C1 and V regions of syndecan-2, respectively. EphB2 phosphorylation is required for syndecan-2 clustering on dendrites and the induction of mature spines. When both of the residues Y189 and Y201 are mutated, the mutant syndecan-2 proteins no longer cluster and promote spinogenesis (). However, EphB2 phosphorylation seems only to affect spine formation but not filopodia formation because filopodia formation is not affected in neurons expressing syndecan-2 Y189F/Y201F mutant (). In addition, overexpression of an EphB2 kinase-dead mutant in hippocampal neurons prevents spine but not filopodia formation (). In addition to tyrosine phosphorylation, the V region of the syndecan-2 cytoplasmic domain contains two PKC phosphorylation sites. syndecan-2 is phosphorylated by PKC and is critical for establishing left-right asymmetry during early development (). However, PKC inhibitors did not prevent filopodia formation induced by syndecan-2, suggesting that PKC phosphorylation on the V region only regulates the function of syndecan-2 at an early embryonic stage for left-right decision. Neither EphB2 nor PKC phosphorylation regulates syndecan-2–induced filopodia formation. In conclusion, our studies show clearly that syndecan-2 plays an active role in delivering a biochemical signal into cells. Syndecan-2 remodels the cytoskeleton and promotes the formation of filopodia via the neurofibromin–PKA–Ena/VASP pathway. Hippocampal neuronal cultures were performed as described previously (; ) with the minor modification that 300,000 cells per well were plated in 12-well plates containing poly--lysine–coated coverslips in each well. Transfection using calcium phosphate precipitation was performed to deliver the plasmid DNA into neurons. To study the effect of syndecan-2 on filopodia formation, transfection of syndecan-2 was performed at 1–2 DIV, and immunostaining was performed at 4–5 DIV. Filopodia formation was determined by the density of protrusions emerging from neurites with a length between 0.75 and 10 μm. The protrusions longer than 10 μm were recognized as dendritic branches. To study the effect of syndecan-2 on spine formation at 8–9 DIV, the length of individual protrusions and protrusion density were measured. Only the protrusions longer than 0.5 μm were counted. To study intrinsic dendritic spine formation, transfection was performed at 11–12 DIV, and immunofluorescence staining was performed at 16–17 DIV. The densities of protrusions longer than 0.5 μm were determined. All results were shown in cumulative probability distribution and statistically analyzed using the Kolmogorov-Smirnov test. Significance was determined based on the D and corresponding p-values. All of the data obtained in Kolmogorov-Smirnov tests and sample sizes of each experiment are summarized in the supplemental text (available at ). The details for transfection of COS cells and immunoblotting of syndecan-2 and neurofibromin have been previously described (,; ). The conditions for HEK293T cell transfection using LipofectAMINE (Invitrogen) were identical to those used for COS cells. Immunoprecipitation using COS cell extract was performed as described previously () with some modifications. In brief, 1 d after transfection, cells were washed twice in PBS and treated with 2 mM dithiobis[succinimidylpropionate] (Pierce Chemical Co.) in PBS at 37°C for 30 min. Free dithiobis[succinimidylpropionate] was then removed by washing twice with PBS. Cells were lysed with radioimmunoprecipitation assay buffer and immunoprecipitated using syndecan-2G antibody. To determine the filopodia density on HEK cell, 24 h after transfection, cells were harvested for immunostaining using syndecan-2G antibodies. Filopodia revealed by syndecan-2G signals on the cell surface were counted and divided by the length of cell circumference. For each treatment, a filopodia density of >50 cells was measured. Cumulative probability distribution was shown, and the Kolmogorov-Smirnov test was used to check the significance. All of the data obtained in Kolmogorov-Smirnov tests and sample sizes of each experiment are summarized in the supplemental text. A rabbit polyclonal syndecan-2 antibody (syndecan-2G) was generated using a GST–syndecan-2 (residues 33–211) recombinant protein as antigen. The antibody was affinity purified using protein A–Sepharose. The monoclonal antibody against GFP was purchased from Invitrogen. VASP antibody was obtained from BD Biosciences. Antibodies against phosphorylated S157 and S239 of VASP were purchased from Chemicon. Plasmid construction for the mammalian expression of syndecans 1–4 has been previously described (). EGFP-Mena, EGFP-VASP, EGFP-EVL, FP4-mito, and AP4-mito constructs (; ) were provided by F.B. Gertler (Massachusetts Institute of Technology, Cambridge, MA). PKI construct was provided by H.-M. Shih (Institute of Biomedical Science, Academia Sinica, Taipei, Taiwan). To construct the C-terminal deletion mutants syndecan-2Δ3, -2Δ20, and -2Δ32, the respective syndecan-2 cDNA fragments were PCR amplified and subcloned into the EcoRI site of the vector GW1-cytomegalovirus. The same sense oligonucleotide primer (5′-CGAT CGGGTACGAGCCACG-3′) was used for PCR construction of all three mutants. The antisense oligonucleotide primers were as follows: for syndecan-2Δ3, 5′-CGCTACTCCTTAGTGGGTGCCTTCT-3′; for syndecan-2Δ20, 5′-CGCTAAAGGTCGTAGCTTCCTTCGTCTTTCTTC-3′; and for syndecan-2Δ32, 5′-CGCTAGTACACCAACAACAGGATGAG-3′. The underlined sequences are EcoRI sites that were appended for cloning purposes. To construct the syndecan-2ΔC1 mutant, a pair of primers (5′-CGATGCGGGTACGAGCCACG-3′ and 5′-GTACACCAACAACAGGATGAG-3′) were used to amplify the region containing the extracellular and transmembrane domains. A pair of primers (5′-GACCTTGGAGAACGCAAACCG-3′ and 5′-CGGAATTCTTATGCATAAAACTCCTTAGT-3′) was used to amplify the variable and C2 domains. These two fragments were ligated by blunt-end ligation and subcloned into the KpnI and EcoRI sites of the vector GW1-cytomegalovirus. A QuikChange XL Site-Directed Mutagenesis kit (Stratagene) was used for generating the syndecan-2 RK/AA, KD/AA, and silent mutants. For syndecan-2 RK/AA, the pair of primers 5′-GTTGGTGTACCGCATGGGAAAGACGAAGGAAGCTAC-3′ and 5′-GCTTCCTTCGTCTTTCCCATGCGGTACACCAACAAC-3′ was used for site-directed mutagenesis. For syndecan-2 KD/AA, the primers 5′-GTACCGCATGCGGAAGAGCGAAGGAAGCTACGACCTTG-3′ and 5′-GTCGTAGCTTCCTTCGCTCTTCCGCATGCGGTACAC-3′ were used. For the syndecan-2 silent mutant, the pair of primers 5′-AGCAGCTCCATTGAGGAAGCTTCAGGTTTATCCTATTG-3′ and 5′-C AATAGGATAAACCTGAAGCTTC-3′ was used. The underlined residues indicate the mutated sites. For vector-based syndecan-2 RNAi construction, a pair of oligonucleotides (syndecan-2is, 5′-GATCCCCTTCAAGAGATTTTTGGAAA-3′; syndecan-2iAs, 5′-GCTTTTCCAAAAATCTCTTGAAGGG-3′) was annealed and inserted into the BgIII and HindIII sites of the siRNA expression vector pSUPER.neo+GFP (OligoEngine). The underlined residues correspond to the nucleotide residues 148–166 of rat syndecan-2 cDNA. The NF1 shRNA expression constructs 27 and 32 in vector pSM2c were purchased from Open Biosystems and corresponded to the nucleotide residues 8,416–8,436 and 3,349–3,370 of human type I NF1, respectively. Because residues 8,416–8,436 are identical in the human and rat NF1 genes, construct 27 can also knock down rat NF1. Nonsilencer control expressing and shRNA sharing no homology with any known mammalian genes were also purchased from Open Biosystems. The Jn and Pn mammalian expression constructs were created by excising the Jn and Pn fragments from the Jn-pGAD10 and Pn-pGAD10 constructs () with EcoRI and subcloning them into the EcoRI site of a modified vector GW1 containing a myc cassette. Plasmids EGFP and EGFP-actin were purchased from CLONTECH Laboratories, Inc. For myc-tagged actin, actin cDNA was digested from EGFP-actin and subcloned into the BglII site of the vector myc-GW1. HEK293T cells were transfected with the vector control or a variety of syndecan-2 constructs. 18 h after transfection, cells were harvested for PKA activity analysis using an ELISA kit from Calbiochem. Equal protein amounts of cell extracts were used. All animal experiments were performed with the approval of and in strict accordance with the guidelines of the Academia Sinica Institutional Animal Care and Utilization Committee and the Republic of China Council of Agriculture Guidebook for the Care and Use of Laboratory Animals. Pregnant rats were housed individually and killed by CO inhalation. All efforts were made to minimize animal suffering and to reduce the number of animals required. Fig. S1 is the characterization of syndecan-2 and neurofibromin RNAi constructs and the dominant-negative mutant of neurofibromin. Fig. S2 shows the specificity of syndecan-2 antibody and expression of syndecan-2 wild-type and mutant constructs. Fig. S3 shows that PKC and PI3K are not involved in filopodia formation downstream of syndecan-2. Fig. S4 shows that Ena/VASP proteins are the downstream effectors of PKA activated by syndecan-2 in HEK293T cells. Supplemental text contains the data obtained in Kolmogorov-Smirnov tests, including D and corresponding p-values and sample sizes of each experiment. Online supplemental material is available at .

Synapses are highly specialized and asymmetric intercellular junctions organized into morphologically, biochemically, and physiologically distinct subdomains. At the presynaptic terminal membrane, active zones mediate Ca-dependent synaptic vesicle fusion, whereas the surrounding periactive zones are essential for synaptic vesicle endocytosis and the control of synaptic terminal growth (; ). Definition of distinct synaptic subdomains is not restricted to the plasma membrane but is also clearly visible within the presynaptic terminal cytoplasm. Notably, synaptic vesicles are clustered at the cell cortex, in the vicinity of active zones. In addition, they seem organized into functional subpools displaying distinct release and recycling properties (). Such an organization requires the precise trafficking and targeting of vesicles to their appropriate location and the specific recruitment and release of subsets of vesicles, depending on the stimulation conditions. One of the main challenges synapses have to face is maintaining such a highly organized structure while constantly adapting their morphology and strength in response to developmental programs and/or external stimuli. Indeed, synaptic terminals can adjust their size; the number, size, and composition of their pre- and postsynaptic membrane specializations; and the availability and release competence of cytoplasmic synaptic vesicles. These dynamic changes require the maintenance of precise physical and functional connections between pre- and postsynaptic compartments, as well as between cytoplasmic and plasma membrane subdomains. To date, the mechanisms allowing such a dynamic reorganization are still poorly understood. However, using the neuromuscular junction (NMJ) as a genetic model, different components of periactive zones, including transmembrane proteins and adaptor molecules, have been implicated in the control of terminal outgrowth (; ; ; ; ). Cell adhesion molecules (CAMs) of the Ig superfamily seem particularly important in maintaining the integrity of synaptic terminals but also in transmitting signals to the cell interior, thereby promoting differentiation of pre- and postsynaptic specializations and regulating synaptic structure and function (; ; ; ; ; ). Moreover, the actin-rich presynaptic cytoskeleton is important for rearranging synaptic domains and for controlling synaptic vesicle distribution and release ability (). How the linkage between cortical cytoskeleton, cytoplasmic vesicle pools, and specialized membrane domains is mediated and, more generally, how plasma membrane and cytoplasmic membranes are spatially and functionally connected largely remain to be elucidated. Here, we identify the transmembrane Ig CAM Basigin (Bsg) as a new component of periactive zones at NMJ synapses. Bsg is the only member of the Basigin/Embigin/Neuroplastin family of glycoproteins, of which mammalian Bsg has been shown to have multiple functions, including in tumor progression (). It seems to regulate cell architecture and cell–cell recognition (; ), act in signaling (; ), and act as a chaperone for transmembrane proteins (; ). By analogy to other mammalian cell surface glycoproteins, and in particular to the CD44 transmembrane protein family (), Bsg may be essential for establishment of transmembrane complexes and for organization of cell structure and signal transduction cascades. Interestingly, mammalian Bsg and Neuroplastin have been suggested to play a role in memory functions and long-term potentiation, respectively, although their precise function has not been determined (; ). Our in vivo study shows that Bsg is required in both pre- and postsynaptic compartments to control formation and growth of synaptic varicosities (or boutons) at larval NMJs. We also show that Bsg is a bona fide Ig CAM because it can promote cell–cell adhesion and its transmembrane and/or juxtamembrane cytoplasmic domains are critical for its function in vivo. Furthermore, down-regulation of affects the size of postsynaptic receptor fields, as well as the distribution of synaptic vesicles within neuronal terminals. These defects are associated with alterations of the actin/Spectrin network, suggesting that Bsg accumulation at the plasma membrane regulates synaptic compartmentalization and architecture. Strikingly, we found that Bsg function is also essential within the presynaptic compartment for the restriction of neurotransmitter release. Based on our in vivo data, we propose that Bsg may be part of a transsynaptic complex surrounding active zones and involved in the coordinated development of pre- and postsynaptic membranes, as well as in the functional coupling of plasma membrane and cortical subdomains. To identify new proteins controlling synapse development, we searched for proteins specifically accumulating at developing NMJs of larvae. We performed a protein-trap screen in which GFP fusion proteins expressed from their endogenous promoters are randomly generated () and screened the expression pattern of ∼350 GFP lines (see Materials and methods). Thereby, we identified 10 lines exhibiting GFP expression at the larval NMJ and focused on three independent lines showing strong GFP accumulation at larval NMJs, but only low GFP levels along the motoneuron axons, and at the surface of muscle fibers (). In these lines, a strong GFP signal is also observed in different neuropil structures of the larval brain (). Using inverse PCR, we found that in each of these three lines the protein-trap cassette was inserted in the gene CG31605, encoding the homologue of the mammalian protein CD147/EMMPRIN/Basigin, Basigin (Bsg; ). According to predictions, the artificial GFP exon should be incorporated upon splicing into mature transcripts whose transcription starts upstream of the insertion, resulting in the in-frame incorporation of GFP. We confirmed this by RT-PCR (unpublished data) and Western blot analysis using anti-Bsg antibodies raised against the protein (). Bsg is a small transmembrane protein composed of two extracellular Ig-like domains, a highly conserved transmembrane domain, and a short cytoplasmic tail (). Mammalian Basigin has been described as a multifunctional protein regulating different processes, including tumor invasion, reproduction, and sensory and memory functions (; ). Interestingly, is highly expressed in the mouse nervous system (), and Bsg protein is present in purified postsynaptic densities (PSDs) of mouse central nervous system synapses (). In , Bsg has been proposed to regulate cellular architecture during eye morphogenesis (). The cellular mechanisms underlying its functions, however, are still poorly understood. Bsg is the only member of a mammalian protein family including Basigin, Embigin, and Neuroplastin/gp65/gp55. All the members of this family have been suggested to regulate cell–substratum adhesion and/or cell–cell adhesion, and are therefore proposed to belong to the Ig CAM family (; ; ; ). To check if the distribution of tagged Bsg reflects that of the endogenous protein, we stained wild-type larvae with anti-Bsg antibodies. Endogenous Bsg shows a localization pattern identical to that of the GFP fusion (), and both precisely colocalize with Discs large (Dlg), a transmembrane protein present both pre- and postsynaptically, but mainly accumulating in stacks of postsynaptic membranes named subsynaptic reticulum (SSR; ; ). Like Dlg, Bsg accumulates to higher levels at type I than at smaller type I boutons (; and not depicted). To exclusively visualize the presynaptic expression of Bsg, we next expressed a GFP-tagged Bsg fusion specifically in the presynaptic compartment () and observed a robust GFP signal at NMJs. Consistent with an accumulation of Bsg at the presynaptic membrane, the inner aspect of both endogenous Bsg and GFP-Bsg protein-trap fusion labels partially overlap with the presynaptic membrane marker HRP (). Further analysis revealed that Bsg is not homogenously distributed at the membrane but is excluded from active zones labeled with anti-Bruchpilot (BRP) NC82 antibodies (; ). Therefore, similar to other transmembrane proteins involved in the structural control of synaptic terminals, such as Dlg or Fasciclin II (Fas II), Bsg localizes to periactive zones (). According to data from the genome project, encodes nine distinct putative transcripts, of which eight encode the same protein (). One of the predicted transcripts, RG, encodes a slightly different protein. However, this transcript is barely detected after quantitative RT-PCR on larval fillet extracts (unpublished data). To address the in vivo requirement for Bsg at the larval NMJ, we searched for element insertions near the transcription starts of the longer transcripts and found five belonging to the same lethal complementation group. Three of these (lk13638, lk14308, and NP3198), when placed in trans over a deficiency covering the locus (, hereafter referred to as ), cause an embryonic/early larval lethality that can be rescued by ubiquitous expression of a transgene (), and represent strong hypomorphic alleles (; unpublished data). Two other insertions, NP6293 and lSH1217, behave genetically as weaker hypomorphic alleles, as, respectively, 30 and 50% of the corresponding hemizygotes reach third larval instar (). This semilethality is reverted after precise excision of NP6293 (). Consistent with these data, Western blot analysis shows that Bsg expression levels are greatly reduced in /NP6293 and /lSH1217 mutant larvae but restored to normal levels after precise excision of NP6293 (). The amount of Bsg specifically accumulating at the NMJ is also significantly reduced in /NP6293 larval fillets, compared with wild type (). Together, these results show that NP6293 and lSH1217 are mutant alleles suitable for analysis of larval NMJ development and maturation. We therefore renamed them and and used them for subsequent studies. To determine whether mutants exhibit defects in their motoneuron connection pattern and/or NMJ morphology, we examined synaptic boutons and motoneuron membranes of both and third instar larvae. Axonal targeting is not altered to a visible degree in these animals. However, the growth of synaptic boutons is strongly altered, as revealed by the considerable increase in their size (; and Table S1, available at ). In particular, the proportion of very large boutons (>12 μm) is greatly increased in mutants compared with controls ( and Table S1; P < 0.001). The observed increase in bouton size is associated with a concomitant reduction of both NMJ branching and bouton number (; and Table S1), keeping the overall NMJ size close to normal (muscle 4 NMJ area: 165.33 ± 9 μm [ = 18] and 163.97 ± 15 μm [ = 12] for and larvae, respectively; P > 0.05). Moreover, defects in bouton size and number are already observed in second instar animals (Fig. S1) and revert after precise excision of NP6293 ( and Table S1). To explicitly determine whether these growth defects could be rescued and whether they reflected pre- and/or postsynaptic function of at the NMJ, we expressed a wild-type copy of in specific compartments of larvae. Expression of wild-type Bsg solely in muscles (using -Gal4 or 24B-Gal4) or in neurons (using -Gal4), partially, but significantly, rescued both the increase in bouton size and the reduction of bouton number observed in mutant larvae (; and not depicted). Near-complete rescue of bouton size and junction growth was obtained only when expressing wild-type Bsg both pre- and postsynaptically (; and Table S1). Collectively, we conclude that Bsg is needed for efficient outgrowth of larval NMJs and that its function is required in both pre- and postsynaptic compartments to define boutons of proper size. Such a dual requirement is documented for the Ig CAM Fas II and is thought to reflect the establishment of transsynaptic homophilic interactions (). We thus tested whether Bsg might also promote cell–cell adhesion. As shown in , S2 cells transfected with a GFP-Bsg construct strongly adhere to each other, whereas S2 cells transfected with a control GFP construct do not. Thus, Bsg promotes cell aggregation, consistent with the idea that Bsg could regulate the addition and growth of synaptic boutons through modulation of cell adhesion. Depending on the cell type and/or the protein partners, different domains of mammalian Bsg are required for its activity (; ; ). Thus, to determine which domains of Bsg are required for its function at the larval NMJ, we generated GFP-tagged truncated variants () and assayed their capacity to rescue morphological defects. Bsg lacking the most C-terminal part of the intracellular domain (Δintra) rescues defects in bouton size and number similarly to the full-length tagged form (fl) when expressed presynaptically (). In contrast, forms composed of the two Ig domains only (Extra) or of the two Ig domains of Bsg fused to the transmembrane and intracellular domains of rat CD2 (Bsg-CD2) do not significantly rescue the decrease in bouton number observed in mutants and only poorly rescue bouton growth defects (). Thus, Bsg transmembrane and/or juxtamembrane cytoplasmic domains appear crucial for regulation of NMJ bouton growth and budding by Bsg. The cytoplasmic juxtamembrane region of Bsg contains a conserved cluster of positively charged residues (KRR; ). We found that when KRR is substituted with NGG, the mutated protein only poorly rescues the reduced bouton number and only to a low extent the increased bouton size of larvae (). Moreover, ubiquitous expression of the KRR→NGG mutated protein does not significantly rescue the early lethality of the strong mutant combination /lk13638, whereas full-length Bsg does (), further indicating a crucial and previously unknown role of this motif for Bsg function. Given that mutants exhibit defective NMJ morphology, we next tested whether the assembly and/or maintenance of pre- and postsynaptic specializations might also be altered. As shown in Fig. S2 (A and B; available at ), the overall distribution and complementary accumulation of markers specific to perisynaptic zones and PSDs seems to be normal at junctions. Moreover, no alteration of SSR integrity could be detected at the light microscopy level (Fig. S2, C and D) or at the ultrastructural level (). Next, we assayed the distribution of receptor fields and active zones, using antibodies recognizing the glutamate receptor subunit GluRIID () in combination with anti-BRP NC82 antibodies (). As illustrated in (A and B), the distribution of these two markers is normal at junctions: BRP and GluRIID remain concentrated in individual puncta of normal intensity and distribution (density of BRP puncta: 1.35 ± 0.3/μm and 1.19 ± 0.2/μm for and third instar larvae, respectively; P > 0.05). Moreover, as described for wild-type animals, BRP release sites are reproducibly found in direct apposition to postsynaptic glutamate receptor clusters in larvae (, insets; and not depicted). Consistent with these observations, transmission EM showed that active zones are found at normal frequency and that their characteristic electron-dense specializations (T-bars) are of normal morphology (). Quantification, however, indicated a slight increase in the electron-dense PSD diameter (), which correlates with a slight, but significant, increase in the mean size of GluRIID clusters observed using light microscopy ( = 0.76 ± 0.01 μm, = 525; = 0.84 ± 0.01 μm, = 501; P < 0.001; ). Thus, these results suggest that, although Bsg is involved in definition of receptor field size, its function is not essential for specifying active and periactive zone domains. Given that Bsg has been suggested to regulate cell architecture, possibly by modulating the cell cytoskeleton (), we checked the integrity of the actin-based cytoskeleton at NMJs. α-Spectrin (α-Spec) closely associates with the NMJ juxtamembrane actin-rich cytoskeleton (). Although it is mainly enriched in the postsynaptic peribouton area, α-Spec is also found at the inner presynaptic bouton cortex (; ). In larvae, even though no major alterations of the postsynaptic Spectrin cytoskeleton are observed, α-Spec aggregates are detected within the bouton lumen () in ∼38% of NMJ branches (). These aggregates are ∼0.5 μm large and contain other α-Spec–associated proteins, such as β-and βH-Spectrin (not depicted), as well as the actin-associated protein Wasp (). In contrast, no enrichment of microtubule-associated proteins was observed in these aggregates (). To more directly and specifically visualize the presynaptic F-actin network, we expressed the F-actin–binding domain of Moesin fused to GFP (GFP-GMA) exclusively in neurons (). As shown in , GFP-GMA accumulates at the cortex of wild-type synaptic boutons. In mutants, although a cortical actin network is still clearly detected at the periphery of boutons, clusters of F-actin filaments are also frequently present within them (). Altogether, these observations indicate that the organization of the presynaptic F-actin network is altered at NMJs. In the course of our ultrastructural analysis, we observed that abnormally large vesicles (diameter of up to ∼300 nm) are present in boutons () but are only rarely observed after presynaptic reexpression of Bsg in this background (not depicted). The exact nature of these vesicles remains unclear, as we have not observed any concomitant alteration in the distribution and/or size of the FYVE-GFP endosomal compartment () at the light microscopy level (unpublished data). To determine whether these defects could be associated with an alteration of the synaptic vesicle compartment, we analyzed synaptic vesicle distribution using specific vesicle markers. In wild-type boutons, synaptic vesicles are clearly enriched at the cortex but are largely excluded from their central core (). In contrast, in larvae, preferential association of vesicles with the bouton cortex is lost in ∼60% of NMJ branches (), and CSP (cysteine string protein) vesicles fill parts of () or even the entire lumen () of the bouton. CSP staining is also abnormally strong in axonal tracts connecting boutons and appears more granular than in control animals (). An essentially identical mislocalization was observed using two other independent markers of synaptic vesicles, Synaptotagmin and Synapsin (). These defects do not indirectly result from the increase in bouton size observed in mutants, as synaptic vesicle localization appears normal in hemizygous larvae, which also form abnormally large boutons (; ; unpublished data). Together with the fact that such a diffuse distribution can be observed upon tracking of freshly endocytosed synaptic vesicles (FM 1–43 loading assay; Fig. S3, C and D, available at ), our data suggest that Bsg specifically regulates the spatial distribution of synaptic vesicles and, in particular, their proper anchoring to the cortex of synaptic boutons. To address whether the observed changes in the distribution of synaptic vesicles might be linked to functional changes in transmitter release, we recorded postsynaptic currents at larval NMJs. As shown in , the amplitude of the postsynaptic response to the fusion of single vesicles (minis) is increased above wild-type levels in mutants (, bottom right; Fig. S3 B; and Table S2, available at ). This effect is most likely related to the observed enlargement of the postsynaptic glutamate receptor clusters (), given that no increase in the size of synaptic vesicles was found in mutants compared with controls (, 34.4 ± 6.6 nm; , 34.8 ± 7.2 nm; P > 0.5). Notably, the frequency of spontaneous release events is strongly elevated in mutants (, bottom left), and these events often occur clustered in “exocytotic bursts” (, asterisk). The mean amplitude of nerve-evoked excitatory junctional currents (eEJCs) is also increased at NMJs (), which largely correlates with the observed enlargement of minis (Table S2). However, the temporal profile of mutant eEJCs is strikingly lengthened, reflecting a dramatic and atypically delayed release of vesicles. Indeed, although the charge carried by mutant minis is only moderately increased (1.5-fold increase; Table S2), a near eightfold elevation of the charge transferred to the postsynapse after exocytosis occurs upon initial nerve stimulation ( and Table S2). Notably, this value decreases progressively after further low-frequency stimulation, which may result from the exhaustion of the abnormally recruited pool of vesicles responsible for the atypically delayed release component. Averaging the charge transferred over 15 consecutive sweeps nonetheless reveals a near fivefold increase in mutants (Table S2); therefore, a more than threefold elevation of the number of vesicles released per action potential (quantal content) is estimated to occur. These defects reflect a requirement for Bsg within the presynaptic terminal, as sole presynaptic expression of wild-type Bsg in the mutant background rescues both the asynchronous evoked release () and the high frequency of spontaneous release (, bottom left), whereas its sole postsynaptic reexpression does not (). Interestingly, the presynaptic reexpression of Bsg even decreases the amplitude of eEJCs and the frequency of minis below control levels, indicating a dose-dependent role of presynaptic Bsg in restricting vesicle release. #text In brief, +/Y; pBac[3xP3-DsRed; GFPexon]/+; pHer[3xP3-ECFP; α-tub-piggyBacK10]/+ jumpstarter males, carrying the protein-trap transposon and a source of transposase (), but not expressing any detectable GFP, were used to mobilize the GFP cassette and create new insertions. Jumpstarter males were crossed en masse with virgins, and embryos were collected at 25°C on apple juice plates for either 5 h (for late embryo sorts) or overnight (for L1 sorts). They were then aged to stage 16–17 embryos and L1 larvae, respectively. Dechorionated embryos or L1 larvae were sorted using an embryo sorter (COPAS Select 500; Union Biometrica), following a procedure adapted from . Sorted larvae were then manually rescreened at later stages using a stereomicroscope (MZFLIII; Leica) equipped with standard GFP filters. About 350 GFP individuals from a total of ∼1.5 million animals sorted (1 million sorted as young larvae and 0.5 million sorted as late embryos) were selected, which roughly corresponds to a frequency of GFP event recovery of 1/2,000–2,500 (estimated after taking into account the percentage of lethality). Emerging GFP individuals were then crossed individually to flies to establish independent lines. Third instar larvae from each line were dissected, briefly fixed (5 min in 4% PFA), and washed several times in PBT. Fillet preparations were examined for GFP distribution using an epifluorescence microscope. For lines of interest, the position of the inserted protein-trap transposon was determined by sequencing of the flanking genomic regions using the inverse PCR protocol described by . The exact positions (in the genomic scaffold AE003619) of the three protein-trap insertions described are the following: 271285 (line 05.02), 277305 (line 91.03), and 278932 (line 79.13). Although two of these insertions (05.02 and 79.13) are homozygous viable, with homozygous larvae showing a rather normal NMJ morphology, one insertion is homozygous lethal (91.03). Given that the three insertions are located in the same intron, and thus generate the same fusion protein, this lethality is not associated with production of the chimeric protein. The NP6293, lSH1217, and lk13638 insertions, and the deficiency covering locus (), were obtained from the Kyoto, Szeged, and Bloomington stock centers, respectively. NP6293 and l1217 lie at position 267161 and 267380 of 2L, respectively. The NP6293 chromosome contains a single element insertion (as verified by inverse PCR), but contains an additional mutation outside of the region covered by , as indicated by the fact that both NP6293 and NP6293 homozygous animals never reach larval stages, whereas a good number of NP6293/ larvae survive until third larval instar. For precise excision of NP6293, NP6293/CyO; Δ2-3, Sb/+ males were individually crossed to If/CyO females. A single w− CyO male per vial was then selected among the progeny and crossed to If/CyO virgin females to establish a stock. Excisions were then evaluated at the molecular level, by PCR amplification and sequencing of genomic DNA of hemizygous larvae. -Gal4, -Gal4, -Gal4, UAS-syt-EGFP, UAS-GMA stocks were obtained from the Bloomington stock center. All crosses were reared at 25°C. The UAS- transgene was generated by insertion of a 2.7-kb EcoRI–XhoI fragment obtained by double digestion of the DGC clone LD19437 into an EcoRI–XhoI digested pCasper3 vector. The following procedure was used to construct GFP-tagged full-length and mutated Bsg variants in which GFP insertion mimics that generated upon insertion of the protein-trap transposon. Total RNA was isolated from protein-trap larval body wall preparations by standard Trizol extraction and used for reverse transcription with Superscript II RT (Invitrogen) and oligo dT primers. Reverse transcription products were then subjected to further PCR and cloned into a pENTR/DTOPO vector (Gateway technology; Invitrogen). One of the full-length GFP-tagged Bsg clones obtained was used as a template to construct the other Bsg variants. Insertion of the different Bsg-tagged proteins into pUASt vector was achieved through a LR recombination reaction (Gateway technology; Invitrogen) using pTW as a destination vector (Drosophila Gateway Vector Collection; ). Lines expressing similar levels of GFP-tagged mutant proteins were used for the rescue experiments described in (not depicted). Dissections and immunostainings were performed as described by , using the following antibodies: rabbit anti-GFP (1:1,000; Invitrogen), mouse anti-Dlg (1:1,000; Developmental Study Hybridoma Bank [DSHB]), mouse anti-CSP (1:40; a gift from E. Buchner, Theodor-Boveri-Institute, Würzburg, Germany), mouse NC82 anti-BRP (1:50; []), mouse anti-Synapsin (1:20; a gift from E. Buchner), rabbit anti– α-Spectrin (1:100; a gift from R. Dubreuil, University of Illinois at Chicago, Chicago, IL), mouse anti–α-Spectrin (1:50; clone 3A9; DSHB), mouse anti-Futsch (1:50; clone 22C10; DSHB), guinea pig anti-Wasp (1:400; ), FITC- and Cy5-conjugated anti-HRP (Cappel and Jackson ImmunoResearch Laboratories, respectively). Alexa Fluor 488 (Invitrogen), Cy3 or Cy5 (Jackson ImmunoResearch Laboratories) conjugated secondary antibodies were used at a 1:500 dilution. Unless specified, confocal pictures are those of muscles 6/7 NMJs (segments A2–A4) and were taken with a microscope (DMR-E; Leica) equipped with a scan head (TCS SP2 AOBS; Leica) and an oil-immersion 63× 1.4 NA objective. For bouton number quantifications, type I boutons of muscles 6/7 on segment A2 were counted using anti-CSP and anti-HRP double stainings. Muscles were photographed at 20× magnification and then traced and measured using ImageJ. The normalized bouton number was calculated by the dividing bouton number by the muscle surface area (data are expressed as a percentage of the controls in each experiment). Although larvae exhibited muscles of a rather normal size, larvae have smaller muscles. Larvae with extremely thin muscles were excluded from the quantification of normalized bouton numbers. For bouton surface area quantification, confocal pictures of muscle 4 NMJs (segment A2) stained with anti-HRP antibodies were taken with a 63× magnification. Sections along the z axis were projected, and individual type I boutons were then manually traced and measured using ImageJ. For measurement of BRP puncta density, sections of NMJ 6/7 branches were projected along the z axis, and the density was calculated as the ratio between the total number of puncta and the projected surface of the branch (using Image J). Rat polyclonal anti-Bsg antibodies were raised against an N-terminal synthetic peptide (QSLDKLVPNYD) obtained from Thermo Scientific. Crude serum was used at a 1:200 dilution for immunostainings and at a 1:1,500 dilution for Western blot analysis. Insertion of GFP within the epitope used to generate anti-Bsg antibodies probably explains the lower signal observed for tagged versus untagged Bsg proteins (). The highest molecular weight isoform detected in total larval extracts is not detected in body wall extracts () and might correspond to a glycosylated form, as described for mammalian Bsg. The following antibodies were used for Western blot analysis: rabbit anti-GFP (1:500 [Santa Cruz Biotechnology, Inc.] or 1:2,500 [Torrey Pines]) and HRP-conjugated anti–mouse, –rat, or –rabbit (1:2,000; GE Healthcare) antibodies. 3 ml of a 10 cells/ml culture of met-Gal4–expressing S2 cells were plated and transfected the next day with 3.5 μg of either UAS-Golgi-GFP or UAS-GFP-Bsg. After overnight recovery, cells were induced for 30 h with 0.7 mM CuSO, shortly centrifuged, and resuspended at 2 × 10 cells/ml in SFM medium (Invitrogen) containing 18 mM -Glutamine, 50 U/ml penicillin, and 50 g/ml streptomycin and puromycin. 1 ml of each suspension was placed in a 2-ml tube and shaken for 1 h at room temperature. Aliquots were spotted on slides and examined using Nomarski optics and epifluorescence microscopy. Transmission micrographs were obtained from dissected preparations of third instar larvae (NMJ 6/7 and 12/13, segment A2/A3), as described by . All measurements were done using ImageJ. The data are reported as mean ± SEM, and where included, p-values denote the significance according to the Mann-Whitney Rank Sum test. Measurement of synaptic vesicle diameter was performed on populations of vesicles found within a radius of 300 nm around active zone T-bars. Two-electrode voltage clamp recordings were obtained at 22°C from VLM 6 in segments A2 and A3, of late third instar larvae, essentially as previously described (). The composition of the extracellular haemolymph-like saline (HL-3) was as follows: 70 mM NaCl, 5 mM KCl, 20 mM MgCl, 10 mM NaHCO, 5 mM trehalose, 115 mM sucrose, 5 mM Hepes, and 1 mM CaCl, pH adjusted to 7.2. Minis (voltage clamp at −80 mV) and eEJCs (voltage clamp at −60 mV) were recorded with intracellular microelectrodes filled with 3 M KCl to give final resistances of 8–21 MΩ. The data are reported as mean ± SEM. indicates the number of cells examined, and where included, p-values denote the significance according to the Mann-Whitney Rank Sum test. In the figures, the level of significance is marked with asterisks: *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001. The quantal content was roughly estimated as the ratio between the mean charge transferred per action potential and the mean charge carried by single minis (see Tables S2 for raw values). The procedure used for FM1-43 labeling was adapted from . For quantification of FM1-43 distribution upon loading (Fig. S3 C), fillets of third instar larvae were dissected in Ca-free saline (130 mM NaCl, 36 mM sucrose, 5 mM KCl, 4 mM MgCl, and 5 mM Hepes) and incubated for 6 min with high-K saline (45 mM NaCl, 36 mM sucrose, 90 mM KCl, 2 mM MgCl, 2 mM CaCl, 5 mM Hepes, and 0.5 mM EGTA) containing 7 μl/ml (∼10 mol.l) of a 1 μg/μl FM1-43X solution (Invitrogen). Fillets were then briefly washed two times in Ca-free saline, fixed for 5 min in 4% formaldehyde, and washed quickly before being mounting on a slide. FM1-43 fluorescence was imaged right after using a confocal microscope (SP2; Leica). For analysis of FM1-43 loading and unloading, fillets of third instar larvae were dissected in Ca-free saline, transferred into a microscope chamber, and incubated for 5 min with high-K saline containing 6 μl/ml (∼10 mol.l) of FM1-43 dye (Invitrogen). Preparations were then briefly rinsed three times with Ca-free saline, further washed (once for 5 min and once for 10 min) with Ca-free saline, and imaged using confocal microscopy and a 40× immersion objective (“after loading” pictures). We could not quantify the amount of FM1-43 dye loaded after the first part of the procedure, as muscles contract longer than wild-type muscles upon transfer to Ca-free saline, and thus start unloading part of the dye. We also noticed that contractions provoked by the high-K solution are weaker in muscle 4 than in muscles 6/7, and therefore imaged exclusively muscle 4 NMJs. For unloading, two consecutive rounds of high-K saline stimulation (1 min each), separated by washes with Ca-free saline (three times for 5 min), were then applied in the absence of any dye. Junctions were imaged after further washing (“after unloading” pictures), using the same confocal settings as for the “after loading” pictures. Fig. S1 shows second instar larvae NMJ morphology and cell–cell aggregation assays performed with Bsg variants. Fig. S2 shows the distribution of pre- and postsynaptic markers in larvae. Fig. S3 shows the localization of synaptic vesicle–associated proteins (serial confocal sections), the distribution of mini amplitudes, and FM1-43 labeling assays. Table S1 shows raw bouton number and size in mutant and rescued contexts. Table S2 shows recorded electrophysiological properties of mutant and rescued larvae. Online supplemental material is available at .

Neurons are exquisitely polarized cells with axonal and somatodendritic compartments organized into distinct ion channel domains (; ). A striking example is the localization of sodium channels to the axon initial segment (AIS) and nodes of Ranvier, sites of action potential generation and propagation, respectively (). The mechanisms responsible for the formation of these two related axonal domains remain poorly understood. The molecular composition of the AIS and of nodes is remarkably similar (; ; ). Both domains are enriched in voltage-gated sodium channels complexed with the neural cell adhesion molecules (CAMs) NrCAM and the 186-kD isoform of neurofascin (NF; ). Sodium channels also associate in cis with one or more β subunits (), which are likewise concentrated at nodes (, ). Sodium channels are proposed to interact with NrCAM and NF186 via two distinct mechanisms: a direct cis interaction of the β1 channel subunit with NF186 () and indirectly via interactions with ankyrin G, a cytoskeletal scaffold to which nodal CAMs, sodium channels, and their β subunits all bind (; ; ). Specific ankyrin G isoforms of 480 and 270 kD are expressed at the node and the AIS (; ; ). Ankyrin G, in turn, is linked to the cytoskeletal protein βIV spectrin, which is also highly enriched at nodes and initial segments (). The signals that drive assembly of the AIS and nodes are distinct. Although the AIS is intrinsically specified, forming in neurons cultured in the absence of glia (; ; ; ), glial signals are required for node formation (; ). The sequence in which proteins accumulate at these two domains is also different, further suggesting that they assemble by distinct mechanisms. In the peripheral nervous system (PNS), early nodal intermediates contain NrCAM and NF186 (). These are overlain by Schwann cell processes (; ) enriched in the adhesion molecule gliomedin, which binds to NrCAM and NF186 (). After a slight delay, ankyrin G, βIV spectrin, and sodium channels concentrate at nodes (; ; ; ). In contrast, ankyrin G appears to accumulate before βIV spectrin, sodium channels, and NF at the AIS (). Together, these results suggest that the AIS and PNS nodes are likely to assemble by distinct mechanisms. Important insights into the assembly of these domains have emerged from recent functional studies of individual components. Mice deficient in NF have major defects of PNS node formation, including disrupted ankyrin G and sodium channel localization (). It has not been reported whether the AIS is also defective in the absence of the NF186. These results indicate that NF186 plays an essential role in node assembly, potentially via extracellular interactions with gliomedin, which is also required for PNS node formation based on knockdown studies (). In contrast, sodium channels still localize at nodes of mice deficient in NrCAM (), the β1 or -2 subunits (, ), or βIV spectrin (). Although the role of ankyrin G at the node has not been examined directly, it has a key role in the localization of sodium channels and CAMs at the AIS. Thus, mice, which lack the major ankyrin G isoform at Purkinje cell initial segments, have profound defects in AIS formation, including loss of NF186 and sodium channels (; ). Whether binding to ankyrin G is required for the localization of proteins at the node and whether ankyrin G is itself critical for node assembly are major remaining questions. To address these issues and further elucidate the mechanisms of AIS and PNS node assembly, we have compared the targeting and function of NF186 at these two axonal domains. NF186 is dispensable for AIS formation, where it is targeted via intracellular interactions that require ankyrin G. In contrast, NF186 is targeted via its ectodomain to the node, where it is essential for recruitment of ankyrin G, which we now demonstrate is required for sodium channel localization and the stability of the entire nodal complex. The ectodomain of NF186 also independently mediates its clearance from the internode, further contributing to its restricted localization at the node. Thus, sodium channel complexes at these two sites on the axon assemble by distinct mechanisms: initial segments are intrinsically specified and form via an inside-to-outside mechanism nucleated by ankyrin G, whereas PNS nodes form from outside to inside directed by glial signals that recruit NF186 and thereby ankyrin G and sodium channels. The AIS and nodes of Ranvier form reliably in primary cultures of hippocampal (Hc) neurons and co-cultures of dorsal root ganglia (DRG) and Schwann cells, respectively (). DRG neurons develop proximal segments (PSs) in vitro that are intrinsically specified and have a composition similar to the AIS (). These neurons therefore provide convenient models to analyze and compare the mechanisms of assembly of sodium channel domains at different sites along the axon. We first examined the mechanisms regulating targeting of NF186 to these domains, in view of its key role in AIS function () and node formation (). We verified that heterologous NF186 is indeed targeted to the AIS and nodes by nucleofecting Hc and DRG neurons, respectively, with cDNAs encoding epitope-tagged NF186; DRG neurons were then co-cultured with Schwann cells under myelinating conditions. All NF constructs contained an HA tag at their N terminus (); live staining for the HA epitope confirmed their expression at the cell surface (not depicted). For studies of node targeting, NF constructs also contained a GFP tag at their C terminus. HA-tagged NF186 was consistently targeted to the AIS of Hc neurons, identified by staining for ankyrin G and the absence of microtubule-associated protein 2 (MAP2), a somatodendritic marker ( and Fig. S1 A, available at ). HA-NF186 containing a C-terminal GFP tag was similarly enriched at the AIS but also detected more distally in the axon, especially in cells with high expression levels. The location of the GFP tag blocks a candidate PDZ binding sequence (YSLA) at the C terminus of NF186 (). We therefore examined the expression of HA-NF186 with or without this putative PDZ binding sequence. Deletion of the last four amino acids at the C terminus of NF186 resulted in expression of NF186 that frequently extended beyond the AIS (Fig. S1 B), indicating that C-terminal interactions contribute to the AIS localization. Subsequent studies of NF186 targeting to the AIS were therefore performed using HA-tagged NF186 without the GFP tag. HA-NF186-GFP was targeted appropriately to nodes of Ranvier. NF186-GFP was initially expressed at the membrane of the cell body and uniformly distributed along the neurites (). Over time, in myelinating co-cultures, NF186 expression became increasingly apparent at the PS (), as its expression elsewhere along the axon was down-regulated. Of note, NF186-GFP was strikingly localized to mature nodes of Ranvier and heminodes, where it colocalized with ankyrin G (); it remained diffuse in nonmyelinated fibers. The C-terminal GFP tag did not impair targeting to the node and was therefore used as a marker in studies of node targeting and assembly. As a control, we expressed intercellular adhesion molecule 1 (ICAM1), a lymphocyte IgCAM () that is not expressed by neurons. When nucleofected into Hc neurons, ICAM1 had a nonpolarized, predominantly somatodendritic distribution with variable but low level expression in axons (). In DRG neurons, ICAM1 was expressed at the membrane of the soma and axons, with minimal expression in the PS. In co-cultures of DRG neurons and Schwann cells, ICAM1 was uniformly distributed at the membrane of nonmyelinated and myelinated axons, although its expression was frequently reduced at nodes of Ranvier (). These results are quantitated in . Collectively, these data indicate that exogenously expressed NF186 is targeted appropriately and can be used in combination with ICAM1, which is nonspecifically localized, to study mechanisms of targeting to PNS nodes and the AIS. To identify the sequences that target NF186 to the AIS and nodes, we constructed chimeras of NF186 and ICAM1 and generated deletions of various NF186 domains ( and ). We first examined whether NF186 is targeted to the AIS via extracellular or intracellular interactions. Replacement of ICAM1's cytoplasmic segment with that of NF186 directed ICAM1 to the AIS of Hc neurons ( and ) and the PS of DRG neurons (Fig. S2 A, available at ). Similarly, NF186 constructs containing deletions of the major extracellular domains were also targeted predominantly to the AIS (). In contrast, NF186 constructs in which the cytoplasmic segment of NF186 was replaced by that of ICAM1 had a nonpolarized distribution and failed to concentrate at the AIS. In each case, targeting to the AIS (not depicted) and to the PS (Fig. S2 A) was not affected by whether the transmembrane domain was from NF186, which has a palmitoylated cysteine (), or from ICAM1, which does not. Previous experiments demonstrated that the FIGQY sequence in the cytoplasmic segment of NF186 is required for binding to ankyrin G (). Constructs in which this ankyrin binding domain (ABD) was deleted (NF186ΔABD) similarly failed to localize to the AIS and had a nonpolarized distribution instead (). Mutating the tyrosine residue in this sequence to either phenylalanine or histidine also disrupted targeting to the AIS (unpublished data). These results indicate that interactions of the cytoplasmic domain of NF186 with ankyrin G are required for targeting to the AIS and PS and that this targeting does not depend on its extracellular or transmembrane segments. We next performed a similar analysis of the targeting of NF186 to PNS nodes. We transfected DRG neurons with NF186/ICAM1 chimeras () and analyzed their targeting in myelinated co-cultures several weeks later (; quantitated in ). All chimeras were initially uniformly distributed at the membrane of the soma and processes. In striking contrast to the AIS, a chimera of the NF186 ectodomain fused to ICAM1 transmembrane and cytoplasmic domains was targeted appropriately to nodes and heminodes (), whereas a chimera containing the ICAM1 ectodomain and transmembrane domain fused to the NF186 cytoplasmic domain failed to accumulate at the nodal regions (). These findings indicate that the ectodomain of NF186 is necessary and sufficient for targeting to the node. To delineate the specific regions of the ectodomain that direct nodal targeting, we expressed NF186 constructs with deletions of the Ig, fibronectin type III (FNIII), or mucin domains. Deletion of the Ig domains completely abolished nodal accumulation (), whereas deletion of the FNIII or mucin domains did not (). These results indicate that NF186 targeting to the node requires its Ig domains. Interactions with ankyrin G were not required for targeting to the node, as indicated by the proper localization of the NF-ED/ICAM1 chimera as well as NF186ΔABD (). However, in some cases, NF186ΔABD staining extended proximal to the nodal region; it also appeared to turn over more quickly, as indicated by fewer positive nodes in the co-cultures, especially at later time points. These results suggest that although interactions with ankyrin G are dispensable for targeting to the node, they may promote stable and restricted expression at this site (see below). In parallel with myelination, proteins that concentrate at the node are down-regulated along the internode, i.e., the region underneath the compact myelin sheath (). In agreement, full- length NF186 accumulated at nodes and was down-regulated along the internode ( and ). This down-regulation could potentially result from redistribution of NF186 from the internode to the node or, alternatively, down-regulation may be independent of targeting to the node. The latter result is suggested by the fact that NF186ΔIg, which is not expressed at nodes, is still cleared from the internode (). To extend these studies, we examined the targeting of a construct in which the Ig domains of NF186 were replaced by those of ICAM1; this recombinant protein, ICAM1/NFΔIg, has a domain structure and molecular weight similar to that of wild-type (wt) NF186 (). Both wt NF186 and ICAM1/NFΔIg were consistently cleared from the internodal region. However, although the wt protein was targeted to 100% of nodes, the ICAM1/NFΔIg construct was detected (weakly) at only 8% of nodes. These results indicate that clearance from the internode is not dependent on targeting to the node. They also argue that ICAM1 persists along the internode () because of a failure of clearance rather than specific targeting to this site mediated by its ectodomain. These findings further suggest that clearance requires NF186 ectodomain sequences. In agreement, a chimera in which the ectodomain of NF was fused to the cytoplasmic tail of ICAM1 () was cleared, whereas a construct in which the ICAM1 ectodomain was fused to the NF cytoplasmic domain () was expressed at high levels along the internode but not at the node, similar to full-length ICAM1 (). These results corroborate that NF186 ectodomain sequences promote clearance and indicate that the cytoplasmic domain of NF186 is neither necessary nor sufficient for clearance. To identify the ectodomain sequences involved, we analyzed a series of domain deletions. Unexpectedly, NF186 constructs containing deletions of the FNIII or mucin domains were also cleared from the internode (, left), much like NF186ΔIg (). These results indicate that several distinct regions of the ectodomain can promote internodal clearance. The source of the transmembrane domain does not affect internodal clearance promoted by the NF ectodomain (compare , with Fig. S2 B, a and b, respectively). Together, these findings indicate that sequences in the ectodomain, but not the membrane or cytoplasmic segments, are required for internodal clearance. In contrast to internodal clearance, down-regulation along the adjacent unmyelinated (myelin basic protein negative) regions of the axon and discrete targeting to heminodes both require an intact NF186 ectodomain. Full-length NF186 is reliably down-regulated along the adjacent, nonmyelinated portions of the axon and is concentrated at heminodes (). However constructs containing deletions of the Ig, FNIII, or mucin domains, all of which are down-regulated along myelin internodes, continue to be expressed along the adjacent, unmyelinated portions of the axon (; and ). As a consequence, the NF186ΔFNIII and Δmucin constructs are not discretely localized to heminodes as they are at nodes (, compare left and right panels). These findings also indicate that myelinating, but not premyelinating, Schwann cells must have mechanisms to clear these deletion constructs from the internodal axon that ensures their discrete expression at nodes. Although NF186ΔABD was targeted appropriately, it was expressed at fewer nodes and less robustly than full-length NF186, suggesting that ankyrin G interactions may stabilize its expression. To test this possibility directly, DRG neurons were nucleofected with NF186-GFP and NF186ΔABD-GFP and maintained as neuron-only cultures or seeded with Schwann cells and cultured under myelinating conditions. At weekly intervals after nucleofection, lysates were prepared, and the expression of the two NF186 constructs was compared by Western blotting for the GFP tag. Expression of full-length NF186-GFP and NF186ΔABD-GFP in neuron-only cultures (, arrowheads) was similar, peaking 1 wk after nucleofection, remaining strong at 2 wk, and declining thereafter. Expression of NF186ΔABD-GFP decreased at a slightly more rapid rate than NF186-GFP (see Fig. S3 A, available at , for quantitation), indicating that ankyrin G enhances its stability. A minor band of ∼150 kD (, asterisk) was present for both constructs over the time course shown. As this band was recognized by the anti-GFP antibody, it likely reflects cleavage of ∼60 kD from the N terminus of NF186. Endogenous NF186 similarly undergoes proteolysis of its N-terminal segment (Fig. S3 C, left, asterisk). We next examined the effects of Schwann cells on NF186 and NF186ΔABD expression. Nucleofected DRG neurons were seeded with Schwann cells after 1 wk in culture; detergent lysates were prepared at weekly intervals thereafter and blotted for GFP (). Addition of Schwann cells accelerated NF186 turnover, which decreased substantially with the onset of myelination at ∼3 wk, indicated by expression of the myelin protein P0. Deletion of the ankyrin binding sequence further enhanced turnover of NF186ΔABD (compare expression of full-length and deletion constructs at 3 wk; quantitation is shown in Fig. S3 A). For both constructs, the ∼150-kD NF fragment was initially more prominent in the co-cultures and then progressively decreased. Together, these results indicate that NF186 is cleaved in the ectodomain, that this cleavage pattern is altered and turnover accelerated by Schwann cell interactions, and that ankyrin G interactions promote more stable expression. In contrast, the levels of endogenous NF186, detected by blotting nontransfected cultures with a pan-NF antibody, remained stable over several weeks, even in Schwann cell co-cultures (Fig. S3 C, right), presumably reflecting ongoing synthesis. To examine specifically whether NF186 expression at the node was stabilized by interactions with ankyrin G, we extracted myelinated co-cultures expressing NF186-GFP or NF186ΔABD-GFP with Triton X-100. Cultures were then fixed and stained for exogenous NF constructs. In nonextracted cultures, NF186 and NF186ΔABD were localized at nodes and nonmyelinated fibers (). After extraction, no staining in nonmyelinated axons was detected for either protein. Strikingly, after extraction, full-length NF186 persisted at nodes () and PSs (not depicted), whereas NF186ΔABD was not detected at either site, or anywhere in the neuron. These results provide further evidence that binding of NF186 to ankyrin G stabilizes the expression of NF186 at nodes and PSs. To corroborate that the neuronal form of NF, i.e., NF186, is required for node formation () and to examine its role in the AIS, we performed short hairpin RNA (shRNA) knockdown studies, targeting a sequence in the NF186-specific mucin domain. Neurons were infected with the shRNA construct or, as controls, with the pLL3.7 vector alone; Schwann cells were added and co-cultures maintained in myelinating conditions for an additional 3–4 wk. shRNA-treated co-cultures appeared normal although they myelinated at a slower rate than controls. GFP staining revealed nearly uniform rates of infection in both cases. Western blot analysis of DRG neurons demonstrated a substantial (>90%) reduction of NF186 (); there was no effect on NrCAM levels, underscoring the specificity of the knockdown. Staining with a pan-NF antibody revealed NF expression in the nodal axolemma and the glial paranodes of control cultures but only in the glial paranodes of the shRNA-treated cultures (), corroborating the neuron-specific knockdown. Of note, ankyrin G and sodium channels failed to cluster at the great majority of nodes in shRNA-treated cultures (). NrCAM expression was also absent at most nodes (). Gliomedin was detected at some nodes in the shRNA-treated cultures, although it was frequently present at reduced levels (). In contrast to the nodes, shRNA treatment of NF186 did not inhibit ankyrin G and sodium channel accumulation at the AIS of Hc neurons () or at the PS of DRG neurons (not depicted), despite effective knockdown of NF186 at these sites. Together, these results provide compelling evidence for a role of NF186 in node but not AIS/PS formation, indicating that sodium channels accumulate at these regions by distinct mechanisms. We next investigated how NF186 promotes PNS node formation. An important insight was provided by examination of nodes that expressed the NF186ΔABD construct at high levels under lentiviral control. Such nodes frequently had reduced levels of ankyrin G and sodium channels compared with nontransfected nodes or those expressing full-length NF186 (, arrowheads; quantitation is shown in Fig. S4, available at ). These reduced levels may reflect competition between the exogenous NF186ΔABD and endogenous NF186 for binding to glial receptors and suggest that NF186ΔABD is unable to recruit ankyrin G and, hence, sodium channels to the nodes. To test this possibility directly, without the confounding effect of endogenous NF186, we performed a knockdown-rescue experiment. The essential strategy was to knock down endogenous NF186 by shRNA while expressing full-length NF186 or NF186ΔABD constructs containing codon substitutions that preserved their amino acid sequence but rendered them insensitive to shRNA treatment (). As we obtained nearly uniform infection of the neurons by the shRNA lentivirus, we removed the vector GFP sequences to facilitate analysis of the exogenous GFP-tagged NF186 constructs. Western blotting analysis () was performed with an NF antibody specific for a cytoplasmic epitope that includes the ABD; hence, this antibody only reacts with full-length endogenous and exogenous NF186. We also blotted for GFP to identify both exogenous constructs; these run at a higher molecular weight than endogenous NF because of their GFP tags. These blots demonstrate that the shRNA effectively suppressed endogenous NF186 expression without impairing expression of the modified, exogenous wt NF186 or NF186ΔABD proteins. Strikingly, although both full-length NF186 and NF186ΔABD were targeted to nodes, only the full-length construct was effective in recruiting ankyrin G and sodium channels (). Ankyrin G and sodium channels were largely absent from nontransfected nodes (, orange arrowhead), which serve as an internal control for the specificity of the knockdown and rescue. Results from this study are quantitated in . The effective rescue of node assembly by NF186 but not by NF186ΔABD provides compelling evidence that a key role of NF186 is to recruit ankyrin G, and thereby sodium channels, to nodes. Finally, in complementary studies, we examined the effect of a knockdown of ankyrin G on the formation of nodes and PSs in myelinating co-cultures. We infected DRG neurons with lentiviruses expressing shRNAs to sequences corresponding to the membrane and spectrin binding domains of ankyrin G. Knockdown of ankyrin G in neurons was highly effective, as demonstrated by immunostaining () and Western blotting (); ankyrin B expression was not affected based on staining (not depicted) and blotting (), highlighting specificity of the knockdown. Blots also demonstrated that NF186 levels were modestly reduced, whereas NrCAM and sodium channel levels were unchanged. shRNA to ankyrin G eliminated essentially all ankyrin G and the great majority of sodium channel expression at PSs () and nodes (; quantitated in ). Sodium channels were expressed along neurites and did not accumulate in the neuronal soma in knockdown cultures (unpublished data), suggesting that defective expression at nodes resulted from impaired localization, not deficient axon expression or transport. NF186, NrCAM, and βIV spectrin were similarly absent from nodes (). Finally, although gliomedin persisted at some nodes, overall expression was reduced (unpublished data). Collectively, these data demonstrate that a major function of NF186 at nodes is to recruit ankyrin G, which, in turn, is essential for the subsequent localization and assembly of the entire nodal complex. #text Primary cultures of Hc neurons were isolated from E18 rat hippocampi treated with 0.05% trypsin (Invitrogen) in PBS for 30 min at 37°C and dissociated by repeated passage through a fire-polished constricted Pasteur pipette. Cells were plated onto 12-mm coverslips coated with poly--lysine (0.1 mg/ml in 0.1 M sodium borate, pH 8.1) in MEM containing Earl's salts and glutamine with 10% FBS (Gemini), 0.45% glucose (Sigma-Aldrich), 1 mM pyruvate, and penicillin-streptomycin (Atlanta Biologicals). After 2 h, the medium was replaced by Neurobasal medium (Invitrogen) with 2% B-27 (Invitrogen), 0.5 mM -glutamine (Invitrogen), and penicillin-streptomycin. Cultures were maintained at 37°C in a humidified 5% CO atmosphere until used. Primary rat Schwann cells and DRG neurons were established as described previously () with the following modifications. Schwann cells were expanded in DME (BioWhittaker) supplemented with 10% FBS, 2 mM -glutamine, 2 mM forskolin (Calbiochem), and 10 mg/ml pituitary extract (Sigma-Aldrich). DRG were isolated from E16 rat embryos, dissociated with 0.25% trypsin, and plated onto 12-mm glass coverslips coated with matrigel (BD Biosciences). The plating medium (C-medium) consisted of MEM (Invitrogen), 10% FBS, 2 mM -glutamine, 0.4% glucose, 50 ng/ml 2.5 S NGF (Harlan Bioproducts for Science), and penicillin-streptomycin. Neuronal cultures were cycled in Neurobasal medium supplemented with 2% B-27, 2 mM -glutamine, 0.4% glucose, 50 ng/ml 2.5 S NGF, and 10 mM each of 5-fluorodeoxyuridine and uridine (Sigma-Aldrich) every other feeding for 7 d to remove nonneuronal cells. To establish DRG-Schwann cell co-cultures, ∼150,000 Schwann cells were added to the purified neurons in C-medium. The co-cultures were kept in C-medium for 3 d and then switched to C-medium supplemented with 50 mg/ml ascorbic acid (Sigma-Aldrich) to initiate myelination. Rat NF186, tagged with an N-terminal HA epitope tag and a C-terminal GFP tag was provided by S. Lambert (University of Massachusetts Medical Center, Worcester, MA). To generate NF186 without GFP tag (NF186-HA), cDNA encoding amino acids 1138–1240 of NF186 C-terminal domain was PCR amplified with introduction of 3′ stop codon and NotI. The PCR product was subcloned into ApaI–NotI sites of NF186-GFP, with concurrent removal of GFP. Other NF constructs were derived from either HA-NF186-GFP or HA-NF186. A cDNA for mouse ICAM1 was provided by M. Dustin (New York University School of Medicine, New York, NY). NF-ICAM1 chimeric constructs, as well as constructs of NF186 with various domain deletions, were generated by the patch PCR technique () unless otherwise indicated; Pfu Turbo Polymerase (Stratagene) was used in PCR reactions. All constructs were designed as duplicates, i.e., with or without the GFP tag at the C terminus. GFP-tagged proteins were obtained by subcloning cDNA into pGFP-N1 vector (CLONTECH Laboratories, Inc.); proteins without the GFP tag were generated by subcloning cDNA into GFP-C3 (CLONTECH Laboratories, Inc.) with concurrent removal of GFP. All constructs were verified by sequencing. In some cases, live cultures were extracted with Triton X-100 before staining. Nucleofected DRG neurons were rinsed with dPBS and incubated in extraction buffer (30 mM Pipes, 1 mM MgCl, 5 mM EDTA, and 0.5% Triton X-100) for 10 min at 37°C. Subsequently, neurons were rinsed in dPBS, fixed in 1% PFA, and processed for immunofluorescence. Slides were examined by epifluorescence on a microscope (Eclipse E800; Nikon) and on a confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.). Confocal images were acquired with Neoflura 40×/1.3 oil or Apochromat 100×/1.4 oil objectives on an eight-bit PTM (Carl Zeiss MicroImaging, Inc.) using AxioVision software (Carl Zeiss MicroImaging, Inc.); brightness and contrast were adjusted using Photoshop (Adobe). Cultures of DRG neurons and myelinated co-cultures were lysed in a solution containing 1% SDS, 95 mM NaCl, 10 mM EDTA, 1 mM PMSF, 10 mg/ml aprotinin, and 20 mM leupeptin in 50 mM Tris, pH 7.4. Lysates were boiled for 10 min and cleared by centrifugation at 14,000 rpm for another 10 min. Protein concentrations were determined by the BCA method (Pierce Chemical Co.); protein samples (10–40 mg) were fractionated by SDS-PAGE and blotted onto nitrocellulose (Whatman). Appropriate regions of the blot were cut out and incubated with specific primary and secondary antibodies. Blots were developed using the SuperSignal chemiluminescent substrate (Pierce Chemical Co.). In some cases, NF186 and NF186ΔABD levels were quantitated by probing Western blots with I protein A and quantitative comparison to peripherin levels on a phosphorimager. NF186-GFP and NF186ΔABD-GFP were subcloned into the pFUGW lentiviral vector (), modified by introduction of a unique NheI site to facilitate subcloning. 293FT cells were transfected with FUGW constructs and packaging plasmids Δ8.9 and pCMV-VSVG (provided by J. Milbrandt, Washington University in St. Louis, St. Louis, MO) using Lipofectamine 2000 (Invitrogen). The viral supernatants were collected 72 h after transfection, centrifuged at 2,500 rpm for 10 min, aliquoted for one-time use, and frozen at −80°C. DRG neurons were infected 2 d after dissection/plating with the lentiviruses diluted 1:3 in Neurobasal medium. The cells were incubated with the viruses for 48 h. Viral expression in DRG neurons was confirmed by immunostaining for GFP and by Western blot analysis. To generate shRNAs to NF186 and ankyrin G, we used the pLentiLox (pLL3.7) vector in which the U6 promoter drives shRNA expression and GFP is expressed under separate promoter control (; ; provided by L. Van Parijs). For NF186 knockdown studies, shRNA was designed to target a 21-nt sequence GAAGCCACAACAGTTGCCATC (E956-I962) within the NF186-specific mucin-like domain. shRNA sequences to knock down the expression of ankyrin G were designed using Easy siRNA (ProteinLounge), BLOCK-iT RNAi designer (Invitrogen), and siRNA sequence selector (CLONTECH Laboratories, Inc.). Three 19–21-nt sense shRNA sequences were commonly identified by all three software programs: CT19, GAGACATAAACTGGCCAAC (within the ANK repeats of the membrane binding domain); PL11, GGCTGACATAGTGCAACAACT (within the spectrin binding domain); and PL20, GCGCATCTGCAGAATCATCAC (within the spectrin binding domain) Fig. S1 shows targeting of NF186 to the initial segment of Hc neurons. Fig. S2 shows determinants of targeting to PSs and clearance from the internode. Fig. S3 shows quantitation of the turnover of exogenous NF186 constructs and blots of endogenous NF186 expression. Fig. S4 shows that expression of NF186ΔABD at nodes of Ranvier inhibits ankyrin G accumulation. Fig. S5 shows representative images of NF186 and ankyrin G. Online supplemental material is available at .

In mammals, definitive erythropoiesis first occurs in the fetal liver with progenitor cells from the yolk sac (). Within the fetal liver and the adult bone marrow, hematopoietic cells are formed continuously from a small population of pluripotent stem cells that generate progenitors committed to one or a few hematopoietic lineages. In the erythroid lineage, the earliest committed progenitors identified ex vivo are the slowly proliferating burst-forming unit–erythroids (BFU-Es; , ). Early BFU-E cells divide and further differentiate through the mature BFU-E stage into rapidly dividing colony-forming unit–erythroids (CFU-Es; , ). CFU-E progenitors divide three to five times over 2–3 d as they differentiate and undergo many substantial changes, including a decrease in cell size, chromatin condensation, and hemoglobinization, leading up to the enucleation and expulsion of other organelles (). Erythropoietin (Epo) has long been understood to be the major factor governing erythropoiesis; its role in regulating the expansion, differentiation, apoptosis, and activation of erythroid-specific genes is well characterized (). The first phase of erythroid differentiation is highly Epo dependent, whereas later stages are no longer dependent on Epo (). Consistent with this, Epo receptors are lost as erythroid progenitors undergo terminal proliferation and differentiation (). This raises the question of what other signals, if any, these differentiating erythroblasts require to support terminal proliferation, differentiation, and enucleation. The extracellular matrix protein fibronectin has been identified as an important part of the erythroid niche in both the adult bone marrow and fetal liver (), but its precise role in erythropoiesis and potential interaction with Epo-mediated signals is unknown. Fibronectin is a ubiquitous extracellular matrix molecule that presents developmental cues to many cell types, including hematopoietic cells ( ). Interactions with fibronectin are essential for proper erythropoiesis, as adhesion to fibronectin is required for the enucleation of murine erythroleukemia cells (). Human bone marrow erythroid progenitor cells expand in the presence of fibronectin in a dose-dependent manner and do not form enucleated erythroid colonies in the absence of fibronectin (). Collectively, these findings suggest that fibronectin not only provides a supportive niche for erythroid progenitor cells but also plays a role in ensuring proper terminal expansion and differentiation. Fibronectin is a large multidomain glycoprotein that contains binding sites for heparin, collagen, fibrin, and gelatin in addition to a number of cell surface receptors. Adhesion of hematopoietic cells to fibronectin is mediated by at least two integrin pairs. αβ integrin (VLA-5) mediates adhesion to the canonical RGDS sequence in the 10th type III repeat. There are two cell-binding sequences in the type III connecting segment that mediate adhesion to αβ integrin (VLA-4). The LDV sequence forms a high affinity binding site, whereas the REDV sequence forms a binding site with much lower affinity (; ). α integrins appear to be essential for the efficient differentiation and expansion of erythroid progenitors in vivo and in vitro. Deletion of α integrin has no effect on the number of fetal liver erythroid progenitors but results in decreased numbers of differentiated erythroid cells. In in vitro erythropoiesis assays, fetal liver α-null erythroid cells formed only small pale colonies, whereas wild-type cells transmigrated beneath the stroma, expanded, and formed large red colonies (). Early studies have shown that fetal liver erythroid cells express both α and α integrins and that these integrins mediate attachment of the CFU-E to fibronectin and stromal cells (; ; ). However, the biological significance of this adhesion is not yet known. In this study, we sought to characterize the precise role that fibronectin plays in erythropoiesis by using an in vitro model of fetal erythropoiesis (). In this system, populations of erythroid progenitors at varying phases of differentiation can be purified from the fetal liver based on expression of the cell surface markers CD71 and TER119; these same markers can then be used to track the differentiation of progenitor cells cultured in vitro. We present data to support a novel model for erythropoiesis wherein Epo and fibronectin each play a distinct, essential role. We show that erythroid expansion proceeds in two phases, with an early Epo-dependent phase followed by a fibronectin-dependent phase. We also determine that αβ integrin is the primary molecular mediator of the observed fibronectin response and that signals emanating from αβ engagement by fibronectin act to protect cells from apoptosis in a manner similar to the binding of Epo to its receptor. We used murine fetal liver cells to study erythroid differentiation. As previously described, fetal liver erythroid cells can be separated into five distinct populations of progressively differentiated cells based on the expression of CD71, the transferrin receptor, and TER119, an erythroid-specific glycoprotein (; ). Early erythroid progenitors (the R1 population) express moderate levels of CD71 and are TER119. Later progenitor cells (R2) express higher levels of CD71 but are still TER119. As cells continue to divide and differentiate, TER119 expression is induced, and CD71 expression is down-regulated, as indicated by the R3, R4, and R5 populations. R3–5 populations contain no CFU-E activity. The expression of CD71 and TER119 can also be used to track erythroid differentiation in vitro. In this system, purified embryonic day (E) 14.5 TER119 progenitor cells are cultured on fibronectin-coated plates in two phases: first in the presence of Epo for 16–18 h and then without Epo for an additional 24 h. On each day, cells are dissociated with PBS containing EDTA, counted, and stained with CD71 and TER119 antibodies for FACS analysis. In the first phase, CD71 is up-regulated, and there is a modest expansion in cell number. By day 3, TER119 is up-regulated, and CD71 is down-regulated (). We used this in vitro model to study the role of fibronectin in erythropoiesis. We observed a dramatic increase in cell number between days 1 and 2 for TER119 progenitor cells cultured on fibronectin but not on uncoated substrates despite the presence of Epo between days 0 and 1 in both cases (). As previously shown, the withdrawal of Epo during the first day of culture leads to an expansion defect even in the presence of fibronectin (, dashed lines). This result led us to hypothesize that Epo and fibronectin regulate erythroid expansion in temporally distinct regimes. To test this hypothesis, we varied the presentation of extracellular matrix and growth factor cues during the course of the 2-d culture period. When TER119 erythroid progenitors were cultured on a control substrate for 1 d and were transferred to a fibronectin substrate for the second day, the level of expansion was indistinguishable from that achieved in the presence of fibronectin for the entire 2-d period, suggesting that fibronectin is only required during the second day of erythroid culture (). In another test, we isolated TER119 cells from the fetal liver. As described previously, TER119 is a marker of differentiated erythroid cells (), so we hypothesized that the expansion of these cells is fibronectin but not Epo dependent. Indeed, when we cultured differentiated TER119 erythroid cells, we found that they expanded on fibronectin and not on the control substrate and that the addition of Epo had no effect (). It is important to note that these freshly isolated TER119 cells have not been exposed to Epo ex vivo, and, thus, the response can be attributed solely to the presence of fibronectin. Collectively, these results suggest a two-phase model for erythroid expansion in which the presence of Epo on the first day of culture followed by the presence of fibronectin on the second day of culture is each essential for proper erythroid expansion. It is well known that Epo protects early erythroid progenitors from apoptosis, () so we tested the hypothesis that fibronectin acts in a similar manner. First, we examined the roles of Epo and fibronectin in preventing the apoptosis of early erythroid progenitors in the first phase of erythropoiesis. We found that nearly 40% of TER119 cells cultured overnight on fibronectin in the absence of Epo were annexin V positive (). When Epo was present in the culture medium, the percentage of annexin V–positive cells dropped to 23.6%, confirming that Epo plays an antiapoptotic as well as proliferative role during the first phase of erythropoiesis. However, the absence of fibronectin either in the presence or absence of Epo did not have an effect on the level of apoptosis (, 23.6% FN + Epo vs. 18.7% PBS + Epo and 39.9% FN-Epo vs. 42.0% PBS-Epo). To test the role of Epo and fibronectin in preventing apoptosis during the second phase of erythropoiesis, TER119 progenitors were cultured overnight on uncoated substrates with Epo as in and were transferred to fresh fibronectin or control substrates in the absence of Epo. 4 h after Epo removal, 28.8% of the cells from control wells were annexin V positive versus 17.1% of those on fibronectin (). Thus, the presence of fibronectin during the second phase partially protects erythroid cells from apoptotic death. One of the mechanisms of apoptosis protection mediated by Epo is through up-regulation of the antiapoptotic protein bcl-xL (), so we tested whether fibronectin protects against apoptosis in a similar manner in the second phase of erythropoiesis. TER119 progenitors were cultured overnight on uncoated substrates in the presence of Epo and were serum starved for 1 h before being transferred to fresh fibronectin or control substrates in the absence of Epo. Cells were lysed at 30 min, and bcl-xL protein expression was assessed via Western blotting. As shown in , bcl-xL expression is distinctly higher in cells cultured on fibronectin than in those cultured in control wells. Similar results were obtained via quantitative PCR as well as flow cytometry (unpublished data). These data support the idea that fibronectin induces signals that act to prevent apoptosis in the second phase of erythropoiesis, much like the binding of Epo to its receptor does in the first phase. We were interested in determining the molecular mediators of the observed fibronectin response, so we focused on integrins as fibronectin receptors. As previously shown, E14.5 fetal liver cells can be separated into five distinct phases of erythroid development on the basis of the relative expression of CD71 and TER119 (; ). This system allowed us to examine the expression of candidate fibronectin receptors at distinct phases in erythropoiesis that correlate with our two-phase model. We used flow cytometry to determine fibronectin receptor expression on the surface of fetal liver erythroid progenitors. By costaining E14.5 fetal liver cells with antibodies to CD71 and TER119 as well as a panel of integrin subunits, we determined that α, α, and β are the most highly expressed integrins on these cells. The expression of α, α, β, and β integrins was not distinguishable from the background fluorescence on erythroid progenitor R1 and R2 cells, a pooled population of progenitors we refer to as R1 + R2 cells (unpublished data). The expression of α, α, and β integrins is down-regulated as cells differentiate (). R1 + R2 cells express the highest levels of each integrin subunit, and the expression levels in the R3, R4, and R5 populations progressively decrease. Although R4 and R5 cells have completely lost the expression of α integrin, the expression of α and β integrins is clearly distinct from the background in these populations. To test the functional role of the loss of integrin expression on fetal liver erythroid cells, we adapted a quantitative cell–extracellular matrix adhesion assay for use with the fetal liver erythroid system. A centrifugation assay is ideal for this application because it is highly quantitative and repeatable while requiring no special equipment (). Using this assay, we demonstrated a progressive loss in adhesion to fibronectin during erythroid development (). Sorted R1–5 cells were allowed to attach to plates coated with 10 or 3 μg/ml human plasma fibronectin and were centrifuged at an acceleration of ∼1,000 . Whereas 75% of R1 + R2 cells adhere to 10 μg/ml fibronectin under these conditions, the fraction of adherent cells decreases progressively in the R3–5 populations. Adhesion of all cell populations was greater on 10 μg/ml fibronectin than on 3 μg/ml. The progressive decrease in adhesion on fibronectin parallels the loss of α-, α-, and β-integrin expression shown in . The optimum centrifugal force for cell detachment was determined by applying a range of centrifugal accelerations to R1 + R2 and R5 cells attached to plates coated with 10 μg/ml fibronectin (). A low centrifugal force was not enough to distinguish between R1 + R2 and R5 cell populations, whereas higher forces removed all of the R5 cells. A centrifugal acceleration of 1,000 provided enough force to distinguish between the adhesion of R1 + R2 and R5 cells without removing all of the R5 cells; thus, 1,000 was used for all further studies. To test the role of specific integrin subunits in mediating adhesion to fibronectin, we used recombinant fibronectin fusion proteins. These recombinant proteins contain either the αβ-binding site (VRGD), the αβ-binding site (Vo), or both binding sites (V; ). Cell adhesion assays were performed as in . As shown in , adhesion to all fragments mirrors that on fibronectin in that the fraction of adherent cells on V, Vo, and VRGD decreases progressively in R1 + R2, R3, R4, and R5 populations. For all populations of cells, adhesion was greatest to the V fragment, which contains both αβ- and αβ-integrin–binding sites. We then repeated the adhesion assay with the addition of function-blocking antibodies. Sorted R1 + R2 cells were incubated with function-blocking antibodies to α or α integrin before being added to precoated plates. When α integrins were blocked on R1 + R2 cells, adhesion to the αβ-binding fragment VRGD was almost completely abrogated, whereas adhesion to the αβ-binding fragment Vo was unaffected (). Similarly, blocking α integrins on R1 + R2 cells had no effect on cell adhesion to αβ-binding fragments but abrogated adhesion to αβ-binding fragments. These results indicate that αβ and αβ integrins on fetal liver erythroid progenitor cells mediate adhesion to distinct sequences on fibronectin. Changes in the levels of integrin expression and corresponding adhesion to fibronectin is physiologically relevant, as erythroid progenitors and erythroblasts must be retained in the bone marrow or fetal liver, whereas the more differentiated reticulocytes must be released into the circulation. Because fibronectin is a major extracellular matrix protein in both the fetal liver and bone marrow, we conclude that adhesion of erythroid progenitors by both αβ and αβ integrins is crucial in retaining immature erythroid cells in the marrow, and loss of these integrins is likely crucial in triggering their release. We were interested to determine whether both αβ and αβ integrins were responsible for the observed proliferative and antiapoptotic effects of fibronectin in the second phase of erythropoiesis. To this end, we cultured TER119 erythroid progenitors on the various recombinant fibronectin fragments in Epo-containing media for 1 d and then in Epo-free media for a second day. Importantly, only those cultured on αβ-binding substrates V and VRGD underwent a dramatic expansion between days 1 and 2 similar to cells cultured on intact fibronectin (). In contrast, cells cultured on the αβ-binding fragment Vo exhibited a defect in cell expansion, indicating that αβ-mediated adhesion to fibronectin is necessary for maximum numbers of terminal erythroid divisions. We obtained the same result when we blocked integrin engagement with antibodies. TER119 cells were isolated from the fetal liver as in and were incubated with function-blocking antibodies against α or α integrins for 15 min on ice. Cells were then added to fibronectin-coated or control plates in Epo-free media. As shown in , blocking α integrin blocked expansion significantly on fibronectin (P = 0.0023). Blocking α integrin had a lesser effect. Our two-phase model of erythroid expansion predicts that αβ-integrin engagement is only necessary during the second, Epo-independent phase. We tested this prediction by varying the extracellular matrix cues presented to the cells during the first and second phases of culture, as in . As seen in , cells cultured on the αβ-binding substrate Vo for the first day exhibit normal expansion when transferred to αβ-binding substrates (fibronectin and VRGD) for the second day. Conversely, there is a marked defect in expansion if cells are cultured in the absence of αβ-mediated adhesion on the second day even if αβ integrin is engaged during the first day of culture (). Collectively, these results indicate that the proliferative effect of fibronectin on the second day of erythropoiesis is αβ integrin dependent. We went on to characterize the role of αβ integrin in preventing the apoptosis of differentiating erythroid cells. TER119 progenitor cells were cultured overnight on PBS in the presence of Epo and were transferred to fresh fibronectin or PBS wells in the presence or absence of function-blocking antibodies to α and α integrins. Blocking α integrin increased the percentage of annexin V–positive cells 4 h after Epo removal, whereas blocking α integrin had little effect on the percentage of apoptotic cells (). Representative flow cytometry data are shown in , and the normalized means of three independent experiments are shown in . Collectively, these results establish that αβ integrin protects differentiating erythroid cells from apoptosis much like the action of Epo on early erythroid progenitors. Over the course of erythroid differentiation, progenitor cells undergo three to five cell divisions, during which they undergo profound morphological and biochemical changes. These include up-regulation of the cell surface transferrin receptor CD71, which is required for import of the iron essential for hemoglobin function, followed by the induction of many erythroid-important enzymes and other proteins, including α and β globin; red cell–specific membrane proteins such as glycophorin, TER119, and the anion exchange protein AE1 (also called Band III); and enzymes in the heme biosynthetic pathway. Subsequently, during the final two or three cell divisions, there is a marked decrease in cellular volume, increase in chromatin condensation, and inactivation of DNA transcription before the final steps of enucleation and expulsion of other organelles (). These changes are used to define the sequential stages of erythroid differentiation: proerythroblasts, basophilic, and polychromatophilic erythroblasts, orthochromatic erythroblasts, and reticulocytes. We have used the sequential developmental up-regulation of CD71, the induction of TER119, and the down-regulation of CD71 as a tool for monitoring these stages of erythroid differentiation by FACS analysis both during culture in vitro and in vivo (; ). Epo and its specific receptor are crucial for promoting the survival, proliferation, and differentiation of mammalian erythroid progenitors (). The first stage of erythroid differentiation, from CFU-E to late basophilic erythroblast, is highly Epo dependent, whereas differentiation beyond this stage is no longer dependent on Epo (). Consistent with this, Epo receptors are lost as erythroid progenitors undergo terminal proliferation and differentiation (). In our culture system for primary fetal liver erythroid progenitors, Epo is required only during the first day, when the progenitors undergo the approximately two initial divisions, and is not essential for the later stages of terminal cell proliferation and differentiation. This raises the question of what extracellular signals, if any, these differentiating erythroblasts require to support terminal proliferation, differentiation, and enucleation. The main point of this study is that engagement of αβ integrin by fibronectin provides signals necessary for the terminal expansion of these differentiating erythroblasts. Our experiments indicate that erythroid expansion proceeds in two temporally distinct phases, the first governed by the binding of Epo to its receptor and the second by engagement of αβ integrin by fibronectin. In each phase, Epo and fibronectin act to promote cell expansion by inducing antiapoptotic signals through bcl-xL, but it is likely that both induce and repress the expression of many other genes as well. Although fibronectin is known to be the major extracellular matrix protein in the erythroid niche, its role in erythropoietic differentiation and survival was unknown. We used our established in vitro model of fetal erythropoiesis to systematically vary the extracellular matrix and growth factor cues presented to erythroid progenitors. Our experiments indicate that withdrawal of Epo during the first day of the culture period leads to defects in cell expansion resulting from increased apoptosis ( and ). In contrast, the presence or absence of Epo on the second day of culture had no effect on the extent of expansion of these more differentiated erythroid cells (). Fibronectin plays a reciprocal role in that it has no effect on the survival or expansion of early erythroid CFU-E progenitors, but its absence on the second day of culture leads to increased apoptosis and a reduction in cell division (, , and ). These results establish that erythropoiesis is governed by an early Epo-dependent phase followed by a later fibronectin-dependent phase. To determine the molecular mediators of the observed fibronectin response, we measured integrin expression levels in erythroid progenitor cells; on these cells, α, α, and β were the mostly highly expressed integrins. Furthermore, we showed a progressive down-regulation of α, α, and β integrins over the 3-d course of erythroid differentiation (). A quantitative adhesion assay indicated that the loss of integrin expression is correlated with the loss of adhesion to fibronectin (). Changes in the levels of integrin expression and corresponding adhesion to fibronectin are physiologically relevant, as erythroid progenitors and erythroblasts must be retained in the bone marrow or fetal liver, whereas the more differentiated reticulocytes must be released into the circulation. In this respect, α and α integrins play a similar role in that they both function as adhesion receptors (). Because fibronectin is a major extracellular matrix protein in the bone marrow, we conclude that adhesion of erythroid progenitors by both αβ and αβ integrins is crucial in retaining immature erythroid cells in the marrow, and the loss of these integrins is likely crucial in triggering their release. By culturing erythroid progenitor cells on recombinant fibronectin fragments, we demonstrated that only αβ-integrin engagement supports expansion on fibronectin-coated surfaces (), indicating that α integrins play an additional role in transducing signals from the extracellular matrix. A previous study has identified αβ and αβ as the predominant integrins on erythroid progenitors (), but in the present study, for the first time, we distinguish between the roles of these two fibronectin receptors on erythroid cells. As predicted by our two-phase model, αβ-integrin engagement is necessary only during the second phase of erythropoiesis, when erythroblasts undergo the approximately two final cell divisions and enucleate. As shown in , robust cell expansion is observed when αβ-integrin engagement is provided in the second, Epo-independent phase of the culture period regardless of whether fibronectin or any of its fragments was present during the first, Epo-dependent phase. Thus, our work suggests that αβ integrin plays two roles in erythroid development. In the first phase of development, αβ seems to function solely as an adhesion receptor for fibronectin, whereas in the second phase, it functions to activate pathways that are necessary for erythroid expansion. Interestingly, we found that integrin engagement is not essential for the differentiation of erythroid progenitors, as measured by the expression levels of CD71 and TER119 (unpublished data), supporting the view that erythroid differentiation and expansion are decoupled and regulated by separate pathways (). Withdrawal of Epo during the first phase of erythropoiesis and the lack of engagement of αβ integrin during the second phase both lead to increased apoptosis. Dead cells cannot proliferate or differentiate, and, thus, it is difficult to determine whether signals downstream of the Epo receptor or αβ integrin directly activate signal transduction pathways leading to cell proliferation or differentiation. Alternatively, by preventing apoptosis, these signals could allow previously inscribed signaling pathways to support both the cell division and induction of erythroid-specific genes. In other systems, cytokine and integrin-mediated signals interact to direct cell behavior. For example, the differentiation of mammary epithelia requires signals initiated by the binding of prolactin to its receptor as well as the simultaneous engagement of β integrins by the basement membrane protein laminin (). Similar coordination is evident in neurogenesis, in which simultaneous signals from FGF and β integrins are necessary for neural stem cell maintenance and proliferation (). One important distinction of these developmental pathways with erythroid development is that growth factor and integrin-mediated signals regulate erythroid cells at very different stages of differentiation, whereas in these other systems, they occur simultaneously. Integrin-mediated adhesion to the extracellular matrix initiates a diverse set of intracellular signaling pathways that are specific to each integrin dimer and cell type. Little is known about the precise downstream pathways activated after integrin receptor–ligand binding in erythroid cells. In fibroblasts, integrin engagement often results in the formation of focal adhesions, which form linkages to the cytoskeleton and bring together many different scaffolding proteins and enzymes to activate downstream signaling pathways such as MAPK and Ras (). Although focal adhesion kinase (FAK) has emerged as the key signaling component of focal adhesions, our preliminary data and that of others suggest that the FAK family member PYK2 rather than FAK itself is activated in erythroid cells in response to integrin activation (). Identifying the downstream signaling pathways and transcription factors that are activated by integrin-mediated adhesion in these cells is one important area for future work. In addition, examining any interaction between Epo- and integrin-mediated signals would provide additional insight into the regulation of erythroid development. For example, it is likely that Epo-mediated signals during the first day of proliferation and differentiation of erythroid progenitors generate cells that are primed to receive and transduce integrin-mediated signals during the final days of erythropoiesis. In particular, signals activated downstream of the Epo receptor may affect integrin activation. Determining how αβ integrin affects apoptosis and cellular expansion and how these signals are integrated with Epo-mediated signals is the focus of our future work. C57BL/6 mice were purchased from The Jackson Laboratory and maintained at the Whitehead Institute animal facility. C57BL/6 E14.5 fetal livers were dissected into Hank's Balanced Salt Solution containing 2% FBS, 100 U/ml penicillin, 100 μg/ml streptomycin, and 10 mM Hepes, pH 7.4 (called HBSS+) at a concentration of two livers per milliliter. Liver tissue was disaggregated by vigorous pipetting followed by passage though a 70-μm cell strainer. Cells were blocked with a 1:50 dilution of ChromPure Rat IgG (The Jackson Laboratory) and labeled with a 1:200 dilution of phycoerythrin TER119 and FITC CD71 antibodies (BD Biosciences) for 15 min on ice. Cells were washed and resuspended in HBSS+ and labeled with 1 μg/ml propidium iodide for cell sorting on a MoFlow3 cell sorter (Becton Dickinson). Fetal liver differentiation assays were performed as previously described (). In brief, day 14.5 fetal livers were suspended in PBS + 2% FBS at a concentration of two fetal livers/milliliter. Cells were incubated with 1:50 rat IgG for 15 min on ice followed by 1:100 biotinylated anti-Ter119 antibody (BD Biosciences). Cells were washed, resuspended at the same concentration, and further labeled with 1:10 tetrameric streptavidin complex (StemCell Technologies) for 15 min followed by 60 μl of magnetic colloids per milliliter of cells (StemCell Technologies) for 15 min. Ter119 cells were purified using a StemSep magnetic column (StemCell Technologies) according to the manufacturer's instructions. Cells were then seeded into 24-well plates coated overnight with 0.5 ml of 20 μg/ml human plasma fibronectin, V, Vo, or VRGD recombinant fusion proteins and washed twice with PBS. Cells were seeded at a density of 100,000 cells/ml. For the first 16–18 h, they were cultured in day 1 media, which consists of Iscove's Modified Dulbecco's Medium (Invitrogen) supplemented with 15% FBS (StemCell Technologies), 1% BSA (StemCell Technologies), 10 μg/ml recombinant human insulin (Sigma-Aldrich), 200 μg/ml recombinant human holotransferrin (Sigma-Aldrich), 10 M β-mercaptoethanol, 1% penicillin/streptomycin, 2 mM glutamine, and 2 U/ml Epo (Amgen). After 16–18 h, the medium was changed to day 2 media, which consists of Iscove's Modified Dulbecco's Medium supplemented with 20% FBS (Invitrogen), 10 M β-mercaptoethanol, 1% penicillin/streptomycin, and 2 mM glutamine. At each time point, cells were dissociated with PBS containing 5 mM EDTA and 10% FBS for 5 min at 37°C. Cells were counted using a hemocytometer (BrightLine; Hausser Scientific) and were stained with 1:200 FITC CD71 and phycoerythrin TER119 for 15 min on ice. Cells were washed and analyzed used a flow cytometer (FACSCalibur; Becton Dickinson). FACS data were analyzed using FlowJo 6.0 software (Tree Star, Inc.). In experiments in which the substrate was varied from days 1 to 2, cells were cultured in day 1 media for the first day, dissociated with PBS containing 5 mM EDTA and 2% FBS for 5 min at 37°C, counted using the BrightLine hemocytometer, and added to fresh plates coated with 20 μg/ml human plasma fibronectin, Vo, or VRGD recombinant fusion proteins in day 2 media. After 24-h culture, cells were dissociated and counted. In experiments using TER119 cells, E14.5 fetal livers were blocked and labeled with biotinylated TER119 antibody. Cells were then washed, resuspended in HBSS+, and incubated with streptavidin-coupled microbeads (Miltenyi Biotec) according to the manufacturer's instructions. TER119 cells were isolated on an autoMACS column (Miltenyi Biotec) using the Possel_s program. Recombinant fibronectin fusion proteins were expressed in grown to late log phase and induced with 1 mM IPTG for 4 h. Cells were harvested by centrifugation and stored at –80°C in PBS containing a cocktail of protease inhibitors (Sigma-Aldrich). Bacterial pellets were thawed, and the buffer was adjusted to 50 mM sodium phosphate, pH 8.0, 10 mM imidazole, and 300 mM NaCl. Pellets were incubated with 2 mg/ml lysozyme and DNase for 30 min at 4°C, lysed using a French press, brought to 1% Triton X-100, and centrifuged at 20,000 for 20 min. Supernatants were incubated with TALON resin (CLONTECH Laboratories, Inc.) and washed in 50 vol of wash buffer (50 mM sodium phosphate, pH 8.0, 300 mM NaCl, and 10 mM imidazole), and proteins were eluted with 2 vol of elution buffer (wash buffer adjusted to 250 mM imidazole). Proteins were dialysed against CAPS buffer (20 mM CAPS, pH 11.0, and 150 mM NaCl). The final protein concentrations were determined by UV absorption, and protein sizes were confirmed on SDS-PAGE. To measure apoptosis, TER119 cells were isolated and cultured on uncoated surfaces in day 1 media for 16–18 h. At this time, cells were dissociated, washed, and incubated with 10 μg/ml of function-blocking antibodies to either α (clone PS/2; Serotec) or α (5H10-27; BD Biosciences) integrin for 15 min on ice. The cells were then transferred to fresh fibronectin or control substrates in Epo-free day 2 media. 4 h after Epo removal, cells were dissociated and stained with annexin V–allophycocyanin (BD Biosciences) and 7-AAD (BD Biosciences) according to the manufacturer's instructions. bcl-xL expression was assayed by fixing and permeabilizing the cells and staining with 1:500 antibody to bcl-xL (clone 54H6; Cell Signaling Technology) for 15 min at room temperature followed by 1:1,000 AlexaFluor647 secondary antibody for 15 min at room temperature (Invitrogen). Expression data were then collected on a flow cytometer (FACSCalibur; BD Biosciences) and analyzed using FlowJo software. Cells were gated based on forward scatter and side scatter properties before then analyzing annexin v and 7-AAD levels. C57/Bl6 E14.5 fetal livers were dissected and stained as above with the addition of a 1:100 dilution of the following biotinylated antiintegrin antibodies, which were all obtained from BD Biosciences: α (9C10), α (5H10-27), and β (Ha2/5). Cells were washed, resuspended, and stained with 1:1,000 streptavidin phycoerythrin Cy5.5 (Caltag Laboratories) for 15 min on ice. Cells were washed, resuspended in HBSS+, and labeled with 1 μg/ml propidium iodide for FACS analysis. Data were collected on a cell sorter (MoFlow3; Becton Dickinson) and analyzed with FlowJo 6.0 software (Tree Star, Inc.). The force experienced by the cells is given by the following equation: where V is the volume of the cell, is the density of the cell, is the density of the medium, RCF is the relative centrifugal force, ω is the centrifugal velocity, and is the radius of the centrifugation. In our centrifuge, was 10 cm. For integrin inhibition studies, cells were incubated with 10 μg/ml of the following function-blocking antibodies before cell seeding: α clone PS/2 (Serotec), α clone 5H10-27 (BD Biosciences), and β clone Ha2/5 (BD Biosciences). Cells were stained for 30 min at 37°C. Significance levels were determined using the test algorithm in Excel (Microsoft). For the data in , , , and , a homoscedastic two-tailed distribution was used. In , a heteroscedastic two-tailed distribution was used. TER119 cells were isolated and cultured in day 1 media on uncoated surfaces overnight. Cells were then dissociated and serum starved in Iscove's Modified Dulbecco's Medium + 1% BSA for 1 h before being transferred to fresh fibronectin or uncoated wells. At the indicated time points, cells were lysed in modified radioimmunoprecipitation assay buffer (50 nM Tris, pH 7.4, 250 nM NaCl, 2 nM EDTA, and 0.5% NP-40) containing protease inhibitors (Complete mini [Roche], 4 mM NaF, and 4 mM NaVO) for 20 min at 4°C. Lysates were ultracentrifuged at 13,000 for 10 min at 4°C and analyzed via SDS-PAGE. Membranes were blocked with Odyssey Blocking Buffer (LI-COR Biosciences) for 1 h and incubated with 1:1,000 bcl-xL antibody clone 54H6 (Cell Signaling Technology) and 1:1,000 actin antibody I-19 (Santa Cruz Biotechnology, Inc.) overnight. Proteins were visualized using the Odyssey infrared imaging system (LI-COR Biosciences).

Communication between adjacent cells through gap junctions occurs in nearly every tissue and is fundamental to coordinated cell behavior. Gap junctions are composed of connexins, consisting of an intracellular N terminus, four transmembrane domains, and a cytosolic C-terminal tail. Six connexins oligomerize into a pore-forming connexon, and alignment of two connexons in apposing cell membranes forms a gap junction channel. These channels allow direct cell-to-cell diffusion of ions and small molecules (<1–2 kD), including nutrients, metabolites, second messengers, and peptides, without transit through the extracellular space (; ; ). Gap junctions play important roles in normal tissue function and organ development (; ; ) and have been implicated in a great diversity of biological processes, notably, electrical synchronization of excitable cells, energy metabolism, growth control, wound repair, tumor cell invasion, and antigen cross-presentation (; ; ; ; ; ; ). The importance of gap junctions is highlighted by the discovery that mutations in connexins underlie a variety of genetic diseases, including peripheral neuropathy, skin disorders, and deafness (; ). Connexin43 (Cx43) is the most abundant and best-studied mammalian connexin. Cx43-based gap junctional communication is of a particular interest because it is regulated by both physiological and pathophysiological stimuli. In particular, Cx43-based cell–cell coupling is rapidly disrupted after stimulation of certain G protein–coupled receptors (GPCRs), such as those for endothelin, thrombin, nucleotides, and bioactive lipids (; ; ; ; ; ; ). Disruption is transient, as communication is restored after ∼20–60 min, depending on the GPCR involved (). GPCR-mediated inhibition of intercellular communication will have broad consequences for long-range signaling in cells and tissues where Cx43 is vital, such as dermal fibroblasts, glial cells, and heart. However, the link between receptor stimulation and Cx43 channel closure has remained elusive to date. Numerous studies on the “gating” of Cx43 channels have focused on a possible role for phosphorylation of Cx43 by various protein kinases, in particular, PKC, MAP kinase, and c-Src, but the results remain ambiguous (; ; ). One of the difficulties with unraveling the regulation of Cx43 channel function is that Cx43 functions in a multiprotein complex that is currently ill understood (). One established component of this assembly is the scaffold protein zona occludens 1 (ZO-1), which binds directly to the C terminus of Cx43 via its second PDZ domain (; ). ZO-1 has been suggested to participate in the assembly and proper distribution of gap junctions, but its precise role in the Cx43 complex remains unclear (; ). In the present study, we sought to identify the signaling pathway that leads to inhibition of Cx43 gap junctional communication in fibroblasts. Using a variety of experimental approaches, we show that the levels of phosphatidylinositol 4,5-bisphosphate (PtdInsP) at the plasma membrane dictate the inhibition (and restoration) of Cx43 gap junctional communication in response to GPCR stimulation, with no role for PtdInsP-derived second messengers. We further show that ZO-1, via its third PDZ domain, interacts with phospholipase Cβ3 (PLCβ3) and is essential for G/PLC-coupled receptors to abrogate Cx43-based cell–cell communication. Our results suggest a model in which ZO-1 serves to organize Cx43 and PLCβ3 into a complex to allow exquisite regulation of Cx43 channel function by localized changes in PtdInsP. Rat-1 fibroblasts are ideally suited for studying Cx43 channel function because they express Cx43 as the only functional connexin (; ). Stable knockdown of Cx43 expression (using pSuper short hairpin RNA [shRNA]) resulted in a complete loss of intercellular communication, consistent with Cx43 being the only functional gap junction protein in Rat-1 cells (). shows that the Cx43 binding partner ZO-1 retains its submembranous localization in Cx43-knockdown cells. To assess which G proteins mediate inhibition of gap junctional communication, we introduced active versions of Gα, Gα, Gα, and Gα subunits into Rat-1 cells and examined their impact on cell–cell coupling. Expression of active Gα resulted in complete inhibition of intercellular communication, whereas active Gα, Gα, and Gα left cell–cell coupling unaltered, as indicated by Lucifer yellow (LY) diffusion and electrophysiological assays (). Disruption of gap junctional communication induced by active Gα was persistent, as opposed to the transient inhibition observed after GPCR stimulation (). Similarly, treatment of Rat-1 cells with toxin, a direct activator of Gα (), caused persistent abrogation of cell–cell coupling (). Gα couples to PLCβ to trigger PtdInsP hydrolysis, leading to production of the second messengers inositol-1,4,5-trisphosphate (IP) and DAG (). We monitored PtdInsP in living cells by using a GFP fusion protein of the pleckstrin homology (PH) domain of PLCδ1 (GFP-PH) as a probe (; ; ). In control cells, the PtdInsP probe was concentrated at the plasma membrane. In cells expressing active Gα, however, the probe was spread diffusely throughout the cytosol, indicative of PtdInsP depletion from the plasma membrane (). Although these results are consistent with Gα mediating agonist-induced inhibition of intercellular communication, they should be interpreted with caution because constitutive depletion of PtdInsP from the plasma membrane promotes apoptosis (; ). To monitor the kinetics of PtdInsP hydrolysis and resynthesis with high temporal resolution, we made use of the fluorescence resonance energy transfer (FRET) between the PH domains of PLCδ1 fused to CFP and YFP, respectively (). When bound to plasma membrane PtdInsP, CFP-PH and YFP-PH are in close proximity and show FRET. After PtdInsP breakdown, the probes dilute out into the cytosol, and FRET ceases. The prototypic G-coupled receptor agonist endothelin, acting through endogenous endothelin A receptors, induced an acute and substantial decrease in PtdInsP, reaching a maximum after 30–60 s; thereafter, PtdInsP slowly recovered to near basal levels over a period lasting as long as 45–60 min (, top, red trace). Sustained PtdInsP hydrolysis by endothelin A receptors has been reported previously () and may be explained by the fact that activated endothelin A receptors follow a recycling pathway back to the cell surface rather than the lysosomal degradation route (). The kinetics of endothelin-induced inhibition and recovery of cell–cell communication followed those of PtdInsP hydrolysis and resynthesis, respectively, with communication being restored after ∼75 min (, top, black trace). More transient PtdInsP depletion and recovery kinetics were observed with a thrombin receptor (PAR-1 [protease-activated receptor 1]) activating peptide (TRP), which correlated with a more short-lived inhibition of gap junctional communication (, bottom). Furthermore, a desensitization-defective mutant NK2 receptor (for neurokinin A) that mediates prolonged PtdInsP hydrolysis () inhibits gap junctional communication for prolonged periods of time when compared with the wild-type NK2 receptor (). Although these results reveal a close correlation between the duration of PtdInsP depletion and that of communication shutoff, we note that the restoration of cell–cell communication consistently lagged behind the recovery of PtdInsP levels. Nevertheless, our findings strongly suggest that the G/PLCβ-mediated hydrolysis and subsequent resynthesis of PtdInsP dictate the inhibition and restoration of Cx43 gap junctional communication, respectively. The Gα-activated PLCβ enzymes comprise four members (β1–4; ). PLCβ1 and -β3 are ubiquitously expressed, whereas PLCβ2 and -β4 expression is restricted to hematopoietic cells and neurons, respectively. Rat-1 cells express PLCβ3 but no detectable PLCβ1 ( and not depicted). We stably suppressed PLCβ3 expression using the pSuper shRNA expression vector (). Four different target sequences were selected to correct for clonal variation and off-target effects. Immunoblot analysis shows a marked reduction in PLCβ3 expression in different clones (). When comparing PtdInsP dynamics in PLCβ3-knockdown versus control cells, agonist-induced PtdInsP breakdown was strongly reduced in the PLCβ3-deficient cells (). PLCβ3-knockdown cells showed normal basal cell–cell communication but failed to shut off cell–cell communication after GPCR stimulation (). These results indicate that PLCβ3 is a key player in the control of intercellular communication, supporting the view that GPCRs inhibit gap junctional communication through the G/PLCβ3–PtdInsP hydrolysis pathway. PLC-mediated PtdInsP hydrolysis generates the second messengers IP and DAG, leading to Ca mobilization and PKC activation, respectively. Previous pharmacological studies have already suggested that neither Ca nor PKC have a critical role in GPCR-mediated inhibition of cell–cell coupling (), a notion supported by additional experiments using “caged” IP, the cell-permeable Ca chelator BAPTA-AM (1,2-bis[2-aminophenoxy]ethane-N,N,N,N-tetra-acetic acid), and a PKC-activating bacterial PLC (; Table S1, available at ). Whether PtdInsP-derived second messengers are dispensable for Cx43 channel closure upon GPCR activation remains debatable, however, as the supporting pharmacological evidence is indirect. To examine whether the depletion of PtdInsP suffices to inhibit Cx43 gap junctional communication, we used a newly developed method to rapidly deplete PtdInsP without activating PLC. In this approach, PtdInsP at the plasma membrane is converted into PtdInsP and free phosphate by rapamycin-inducible membrane targeting of the human type IV phosphoinositide 5-phoshatase (5-ptase; ; ). The method is based on the rapamycin-induced heterodimerization of FRB (fragment of mammalian target of rapamycin that binds FKB12) and FKB12 (FK506 binding protein 12), as schematically illustrated in . In this approach, a mutant version of 5-ptase with a defective membrane targeting domain (CAAX box) is fused to FKB12 and tagged with monomeric red fluorescent protein (mRFP), whereas its binding partner FRB (fused to CFP) is tethered to the plasma membrane through palmitoylation (construct PM-FRB-CFP; ). In the absence of rapamycin, 5-ptase resides in the cytosol and leaves PtdInsP levels at the plasma membrane unaltered (, left). Upon addition of 100 nM rapamycin, FKB12 and FRB undergo heterodimerization and the 5-ptase is recruited to the plasma membrane (, right). We expressed the mRFP–FKB12–5-ptase and PM-FRB-CFP fusion proteins in Rat-1 cells and confirmed their proper intracellular localization by confocal microscopy (unpublished data). Addition of rapamycin caused a rapid and complete depletion of PtdInsP, as shown by the disappearance of the PtdInsP sensor YFP-PH from the plasma membrane (, top trace; = 10). As expected, no Ca signal was detected after the 5-ptase–mediated conversion of PtdInsP into PtdInsP (; = 4). To determine how the 5-ptase–induced hydrolysis of PtdInsP affects gap junctional communication, we measured the intercellular diffusion of calcein (added as membrane-permeable calcein-AM) using FRAP (; ). Rat-1 cells expressing mRFP–FKB12–5-ptase and PM-FRB-CFP showed efficient intercellular transfer of calcein. At 2 min after rapamycin addition, however, intercellular dye diffusion was inhibited as inferred from a strongly reduced fluorescence recovery rate (; = 15; P < 0.005; ∼0.25× the recovery rate before rapamycin addition). The recovery of calcein fluorescence could not be decreased any further by addition of 50 μM of the gap junction blocker 2-aminoethoxy-diphenylborane (; ). Rapamycin did not affect cell–cell communication in nontransfected cells (unpublished data). Thus, PtdInsP depletion by 5-phosphatase activation is sufficient to inhibit Cx43 gap junctional communication, with no need for PtdInsP-derived second messengers. PtdInsP at the plasma membrane is generated mainly from PtdInsP by PIP5K (; ). As a further test of the PtdInsP hypothesis, we stably overexpressed PIP5K (type Iα, fused to GFP) in Rat-1 cells in an attempt to prevent PtdInsP depletion after GPCR stimulation (). As shown in , transfected GFP-PIP5K localizes to the plasma membrane. In the PIP5K-overexpressing cells, PtdInsP levels remain elevated (i.e., above FRET threshold levels) after agonist addition (). Nonetheless, GPCR agonists still induced transient rises in IP and Ca (), indicating that excessive synthesis of PtdInsP does not interfere with its hydrolysis. Basal cell–cell communication in PIP5K-overexpressing cells was not significantly different from that in control cells. However, the PIP5K-overexpressing cells failed to close their gap junction channels upon addition of TRP and, to a lesser extent, endothelin (). That endothelin is still capable of evoking a residual response in PIP5K-overexpressing cells may be explained by the fact that endothelin is by far the strongest inducer of PtdInsP depletion (). Expression of a kinase-dead version of PIP5K had no effect on either PtdInsP hydrolysis or inhibition of cell–cell communication (). We conclude that Cx43 channel closure is prevented when PtdInsP is maintained at adequate levels. PtdInsP can modulate the activity of various ion channels and transporters, apparently through direct electrostatic interactions (; ). By analogy, regulation of Cx43 channels by PtdInsP would imply that basic residues in Cx43 bind directly to the negatively charged PtdInsP. Indeed, the regulatory cytosolic tail of Cx43 (aa 228–382) contains a membrane-proximal stretch of both basic and hydrophobic residues (VFFKGVKDRVKGK/R) that could constitute a potential PtdInsP binding site. Local depletion of PtdInsP might then dissociate the juxtamembrane region of the Cx43 tail from the plasma membrane, leading to channel closure. We reasoned that if the Cx43 juxtamembrane domain binds PtdInsP in situ, mutations within this domain might interfere with PtdInsP-regulated channel closure. We therefore neutralized the membrane-proximal Arg and Lys residues by mutation to alanine, resulting in eight distinct Cx43 mutants, notably, K237A,K241A; R239A,R243A; K241A,R243A; R239A,K241A; K237A,R239A; R239A,K241A,R243A; K237A,R239A,K241A; and the “4A” mutant, K237A,R239A,K241A,K243A. When expressed in Cx43-deficient cells, however, all these mutants were trapped intracellularly and failed to localize to the plasma membrane (Fig. S1, available at ). Although this result reveals a previously unknown role for the membrane-proximal Arg/Lys residues in Cx43 trafficking, it precludes a test of the Cx43–PtdInsP interaction hypothesis. We next examined whether PtdInsP can specifically bind to either the Cx43 C-terminal tail (Cx43CT; aa 228–382) or a Cx43CT-derived juxtamembrane peptide (Cx43JM; aa 228–263) in vitro. We generated a GST-Cx43CT fusion protein and determined its ability to bind phosphoinositides in vitro using three distinct protocols. GST-PH(PLCδ1) was used as a positive control. In the first approach, agarose beads coated with either PtdInsP or PtdInsP were incubated with GST-Cx43CT or GST-PH and then pulled down by centrifugation. PtdInsP beads readily brought down the GST-PH polypeptide but not GST-Cx43CT (Fig. S2 A, available at ). Second, we incubated GST-Cx43CT with P-labeled PtdInsP and examined the ability of excess phosphoinositides to displace bound P-PtdInsP. Although GST-PH showed again strong PtdInsP binding that could readily be displaced by excess PtdInsP, there was no detectable binding of PtdInsP to Cx43CT above that observed with GST alone (Fig. S2 B). Finally, we found that PtdInsP (and other phosphoinositides) immobilized on nitrocellulose strips failed to bind either Cx43CT or a 35-amino-acid juxtamembrane domain peptide (Cx43JM; aa 228–263; ; unpublished data). Thus, PtdInsP does not detectably bind to the juxtamembrane domain of Cx43, or to the full-length regulatory tail (aa 228–362), at least in vitro. The very C terminus of Cx43 binds directly to the second PDZ domain of ZO-1, but the functional significance of the Cx43–ZO-1 interaction is not understood. We asked if ZO-1 has a role in modulating gap junctional communication in response to GPCR stimulation. We already showed that RNAi-mediated depletion of Cx43 does not significantly affect the levels and localization of ZO-1 (). Conversely, when ZO-1 expression was knocked down by shRNA, Cx43 levels were unaltered (). ZO-1–knockdown Rat-1 cells retained their fibroblastic morphology and showed normal Cx43 punctate staining and cell–cell coupling (), showing that ZO-1 is dispensable for the formation of functional gap junctions. When ZO-1–knockdown cells were stimulated with endothelin, however, the inhibition of cell–cell communication was severely impaired (). Importantly, overall PtdInsP-dependent Ca mobilization was not affected in the ZO-1–knockdown cells (). We conclude that ZO-1 is essential for the regulation of gap junctional communication by G/PLC-coupled receptors, but not for linking those receptors to PLC activation. A plausible explanation for these findings is that ZO-1 serves to bring the PtdInsP-metabolizing machinery into proximity of Cx43 gap junctions. As a test of the above hypothesis, we examined whether ZO-1 can interact with PLCβ3. PLCβ3 can associate with at least two scaffold proteins, NHERF2 (in epithelial cells) and Shank2 (in brain), via a C-terminal PDZ domain binding motif (; ). We coexpressed HA-PLCβ3 and GFP–ZO-1 in HEK293 cells and performed immunoprecipitations using anti-GFP antibody (). Cell lysates and immunoprecipitates were blotted for GFP and HA. As shown in , PLCβ3 and ZO-1 can indeed be coprecipitated. Next, we coexpressed ZO-1 and a PLCβ3 truncation mutant that lacks the C-terminal 14 residues (HA-PLCβ3-ΔPBD; ) and performed anti-GFP immunoprecipitations. shows that truncated PLCβ3 fails to interact with ZO-1, indicating that PLCβ3 interacts with ZO-1 through its very C terminus, containing the PDZ domain binding motif. Considering that ZO-1 has three distinct PDZ domains, we examined which, if any, PDZ domain binds PLCβ3. We expressed GFP-tagged versions of the three individual PDZ domains in HEK293 cells, either alone or together with HA-PLCβ3. We immunoprecipitated PLCβ3 using anti-HA antibody and blotted total cell lysates and precipitates for both HA and GFP. As shown in , we find that PLCβ3 binds to PDZ3 but not to PDZ1 or PDZ2. To verify that the ZO-1–PLCβ3 interaction exists endogenously, we precipitated ZO-1 from Rat-1 cells and blotted for both ZO-1 and PLCβ3. shows that PLCβ3 coprecipitates with ZO-1. The reverse coprecipitation could not be done, as precipitating antibodies against PLCβ3 are presently not available. Nonetheless, these results suggest that ZO-1, through its respective PDZ2 and PDZ3 domains, assembles Cx43 and PLCβ3 into a signaling complex and thereby facilitates regulation of gap junctional communication by PLC-coupled receptors. #text Materials were obtained from the following sources: endothelin, TRP (sequence SFLLRN), neurokinin A, Cx43 polyclonal, and α-tubulin monoclonal antibodies from Sigma-Aldrich; toxin from Calbiochem-Novabiochem; Cx43 NT monoclonal antibody from the Fred Hutchinson Cancer Research Center; actin monoclonal from Chemicon International; polyclonal PLCβ3 antibody from Cell Signaling; ZO-1 monoclonal antibody from Zymed Laboratories; HRP-conjugated secondary antibodies from DakoCytomation, and secondary antibodies for immunofluorescence (goat anti–mouse [Alexa488] and goat anti–rabbit [Alexa594]) from Invitrogen. HA, Myc, and GST monoclonal antibodies were purified from hybridoma cell lines 12CA5, 9E10, and 2F3, respectively. GFP antiserum was generated in our institute. Constructs encoding active (GTPase-deficient) Gα subunits, EGFP-PH, ECFP-PH, EYFP-PH, and EGFP-tagged mouse type Iα PIP5K have been described (; ; ). Mouse PLCβ3 cDNA was obtained from MRC Geneservice, cloned into pcDNA3-HA by PCR (see Table S2, available at , for primers) and ligated into pcDNA3-HA XhoI–NotI sites. HA-PLCβ3-ΔPBD was obtained by restriction of the full-length construct with Eco47III, cleaving off the very C-terminal 14 residues. Human ZO-1 was cloned into XhoI and KpnI sites of pEGFP C2 (CLONTECH Laboratories, Inc.). GFP-based yellow cameleon 2.1 has been described (). Constructs encoding cytosolic 5-phosphatase fused to FKB12-mRFP and PM-FRB-CFP have been described (). Cells were cultured in DME containing 8% fetal calf serum, -glutamine, and antibiotics. For cell–cell communication assays, cells were grown in 3-cm dishes and serum starved for at least 4 h before experimentation. Monitoring the diffusion of LY from single microinjected cells and single-electrode electrophysiological measurements of cell–cell coupling were done as described previously (). Images were acquired on an inverted microscope (Axiovert 135; Carl Zeiss MicroImaging, Inc.), equipped with an Achroplan 40× objective (NA 0.60) and a camera (F301; Nikon). Cells were harvested in Laemmli sample buffer (LSB), boiled for 10 min, and subjected to immunoblot analysis according to standard procedures. Filters were blocked in TBST/5% milk, incubated with primary and secondary antibodies, and visualized by enhanced chemoluminescence (GE Healthcare). For immunoprecipitation, cells were harvested in 1% NP-40 and 0.25% sodium desoxycholate lysis buffer. Lysates were spun down, and the supernatants were subjected to immunoprecipitation using protein A–conjugated antibodies for 4 h at 4°C. Proteins were eluted by boiling for 10 min in LSB and analyzed by immunoblotting. Cells grown on coverslips were fixed in 3.7% formaldehyde in PBS for 15 min. Samples were blocked and permeabilized in PBS containing 1% BSA and 0.1% Triton X-100 for 30 min. Subsequently, samples were incubated with primary and secondary antibodies for 30 min each in PBS/1% BSA, washed five times with PBS, and mounted in MOWIOL (Calbiochem). Confocal fluorescence images were obtained on a confocal system (TCS NT; Leica), equipped with an Ar/Kr laser. Images were taken using a 63× NA 1.32 oil objective. Standard filter combinations and Kalman averaging were used. Processing of images for presentation was done on a PC using Photoshop (Adobe). All live imaging and time-lapse experiments were performed in bicarbonate-buffered saline containing 140 mM NaCl, 5 mM KCl, 1 mM MgCl, 1 mM CaCl, 10 mM glucose, 23 mM NaHCO, and 10 mM Hepes, pH 7.2, kept under 5% CO, at 37°C. Images of live cells expressing GFP-PH and GFP-PIP5K were recorded on a confocal microscope (TCS-SP2; Leica) using a 63× lens (NA 1.4). Temporal changes in PtdInsP levels in living cells were assayed by the FRET-based PtdInsP sensor, PH-PLCδ1, as described previously (). In brief, Rat-1 cells were transiently transfected with CFP-PH and YFP-PH constructs (1:1 ratio) using Fugene transfection agent and placed on a inverted microscope (Nikon) equipped with an Achroplan 63× oil objective (NA 1.4). Excitation was at 425 ± 5 nm. CFP and YFP emissions were detected simultaneously at 475 ± 15 and 540 ± 20 nm, respectively, and recorded with PicoLog Data Acquisition Software (Pico Technology). FRET is expressed as the ratio of acceptor to donor fluorescence. At the onset of the experiment, the ratio was adjusted to 1.0, and FRET changes were expressed as relative deviations from baseline. Temporal changes in IP levels were monitored using a FRET-based IP sensor, in which the IP binding domain of the human type I IP receptor (aa 224–605) is fused between CFP and YFP, essentially analogous to the sensor described previously (). of ∼5 nM. Intracellular Ca mobilization was monitored using the CFP/YFP-based Ca sensor yellow cameleon 2.1 (; ; ). Traces were smoothened in Excel (Microsoft) using a moving average function ranging from 3 to 6. Rat-1 cells were transiently transfected with PM-CFP-FRB and mRFP–FKB12–5-ptase (). Cells were selected for experimentation when sufficient protein levels were expressed as judged by CFP and mRFP fluorescence. For PtdInsP measurements, the YFP-PH construct was cotransfected. For Ca measurements, cells were loaded with Oregon green-AM. To monitor gap junctional communication, cells were loaded with calcein-AM and analyzed by FRAP (). These experiments were performed on a confocal microscope (TCS-SP2; Leica) using 63× lens (NA 1.4). To overexpress PIP5K (type Iα; ), virus containing the LZRS-PIP5K constructs was generated as described in the following paragraph. Rat-1 cells were incubated with 1 ml viral supernatant supplemented with 10 μl Dotap (Roche). 48 h after infection, cells were plated in selection medium. Transfected cells were selected on 200 μg/ml zeocin (Invitrogen) for 2 wk, and colonies were examined for PIP5K expression. To generate Cx43-deficient Rat-1 cells, Cx43 was knocked down by stable expression of retroviral pSuper (pRS; ) containing the RNAi target sequence GGTGTGGCTGTCAGTGCTC. pRS-Cx43 was transfected into Phoenix-Eco package cells, and the supernatant containing viral particles was harvested after 72 h. For infection, cells were incubated with 1 ml viral supernatant supplemented with 10 μl Dotap (1 mg/ml). 48 h after infection, cells were selected on 2 μg/ml puromycin for 2 wk. Single cell-derived colonies were tested for Cx43 expression and communication. PLCβ3 was stably knocked down by retroviral expression of PLCβ3 shRNA. Four different target sequences were selected, namely, ACTACGTCTGCCTGCGAAATT, GATTCGAGAGGTACTGGGC, TTACGTTGAGCCCGTCAAG, and CCCTTTGACTTCCCCAAGG). Nonfunctional shRNA was used as a control. ZO-1 was transiently knocked down by adenoviral expression of ZO-1 RNAi. First, ZO-1 RNAi oligos containing the ZO-1 target sequence GGAGGGCCAGCTGAAGGAC were ligated into pSuper after oligo annealing. Next, the oligos, together with the H1 RNA promotor, were subcloned into pEntr1A (BamHI–XhoI) and recombinated into pAd/PL-Dest according to protocol (Virapower Adenoviral Expression System; Invitrogen). Virus was produced in 293A packaging cells according to standard procedures. Supernatant containing virus particles was titrated on Rat-1 cells to determine the amount needed for ZO-1 knockdown. Fig. S1 shows intracellular accumulation of the Cx43-4A mutant. Fig. S2 shows lack of detectable binding of phosphoinositides to the Cx43 C-terminal tail. Table S1 summarizes the results of pharmacological experiments. Table S2 shows sequences of the oligos used. Online supplemental material is available at .

The identification and study of tumor suppressor genes has provided insight into the normal mechanisms of cell proliferation control (). Most tumor suppressors function intracellularly to control the cell division cycle; however, the interface between a cell and its environment also plays a critical role in tumor development and metastasis. The product of the neurofibromatosis type 2 (NF2) tumor suppressor gene, Merlin, localizes to and appears to act at this interface (). Loss of NF2 function is associated with the development of multiple cancers in humans and mice (; ; ). In humans, mutations are associated with familial and sporadic nervous system tumors and with other sporadic cancers such as mesothelioma, whereas heterozygous mutant mice develop bone, liver, and other tumors that are highly metastatic. Treatment strategies for NF2 are currently limited given the often intractable location and multiplicity of tumors, together with their tendency to recur. Surgical approaches are the current standard therapy and pharmacological treatments are not available. Merlin is closely related to the ERM (Ezrin/Radixin/Moesin) proteins that are thought to organize cortical membrane domains that interface with the extracellular environment, via linking membrane-associated proteins to the actin cytoskeleton (; ; ; ). Although Merlin can functionally and physically interact with several proteins, including p21-activated kinase (; ), CD44 () and the two PDZ domain–containing adaptors EBP50/NHE-RF1 and E3KARP/NHE-RF2 (; ), the mechanism whereby Merlin controls cell proliferation remains poorly understood (). We have recently found that a signature of Nf2 deficiency in several types of primary cells, including both mesenchymal and epithelial cells, is a failure to undergo contact-dependent inhibition of proliferation and to establish stable cadherin-mediated adherens junctions (AJs) between cells (). Merlin is regulated by cell–cell contact (), localizes to AJs, and physically associates with AJ components. cells, stable AJ structures are not maintained. Defective AJs and loss of contact-dependent inhibition of proliferation may explain the tumorigenic and metastatic consequences of Nf2 deficiency. cells with defective AJs is not known; indeed, the general mechanism of contact-mediated inhibition of proliferation is not well understood. Here we present novel mechanistic insight into a critical role for the NF2 tumor suppressor, Merlin, in coordinating the processes of AJ stabilization with contact-dependent inhibition of epidermal growth factor receptor (EGFR) activity. In the absence of Merlin, confluent cells are unable to silence mitogenic signaling from the EGFR, and their continuous proliferation is blocked by specific pharmacological inhibitors of the EGFR. Upon cell–cell contact, Merlin associates with EGFR via NHE-RF1 and prevents both ligand-induced EGFR internalization and the association of EGFR with its canonical effectors, precluding downstream signaling. Our data are consistent with a model whereby upon cell contact Merlin restrains EGFR into a membrane compartment from which it can neither signal nor be internalized. These studies reveal a novel mechanism of tumor suppressor function, linking the function of Merlin to that of a well-known oncogene and suggesting a possible therapeutic strategy for mutant tumors. cells continue to proliferate after reaching confluence. cells—in essence the mechanism whereby Nf2-expressing cells normally undergo contact-dependent inhibition of proliferation. cells (Fig. cells (). cells exhibit sustained activation of signaling molecules that are not known β-catenin targets (). Alternatively, accumulating evidence suggests that cadherin-dependent cell adhesion can control mitogenic signaling by negatively regulating receptor tyrosine kinases (RTKs) at the plasma membrane (; ; ). mouse embryo fibroblasts (MEFs) as they progress to high cell density with an increasing area of cell–cell contact in the presence of serum (see Fig. S2 A for a definition of confluence; available at ). MEFs. membranes, indicating their dependence upon soluble growth factors (, right). Thus, in proliferating wild-type MEFs, serum growth factors maintain a physiological level of tyrosine kinase activity that is down-regulated as confluence progresses; this down-regulation is defective in the absence of Merlin. Genetic cooperativity has been demonstrated between NF2 and EGFR pathway mutations in (). Moreover, the EGFR localizes to cell junctions, regulates cell adhesion, and can be negatively modulated by cadherin-dependent cell contact (; ; ; ; ). However, the basis of coordination between cell–cell contact and EGFR activity is not known. membranes (), we used antibodies against active, phosphorylated EGFR (pEGFR) to monitor EGFR activation in confluent wild-type and cells. MEFs (). To determine whether EGFR deregulation is a general signature of Nf2 deficiency, we examined EGFR activation in confluent primary osteoblasts (OBs) and liver-derived epithelial cells (LDCs), two key targets of -associated tumorigenesis in mice (; ). OBs nor LDCs undergo contact-dependent inhibition of proliferation (Fig S2 B; see ). OBs maintain elevated levels of both (; not depicted). LDCs retain high levels of pEGFR (). MEFs, OBs, and LDCs restores contact-dependent inhibition of proliferation, electrondense AJs, and low levels of both pTyr and pEGFR (, ; Fig. S2 C; not depicted). In contrast, a version of Merlin containing a patient-derived missense mutation (Nf2) fails to stably localize to AJs (; ), inhibit proliferation, or reduce pEGFR levels (). cells, suggesting that a program of EGFR signaling fails to be down-regulated () (). cells; this is consistent with the fact that they do not proliferate faster than wild-type cells, but proliferate continuously despite reaching confluence. Thus, three different cell types all fail to undergo contact-dependent inhibition of proliferation and to down-regulate EGFR signaling at high cell density in the absence of Merlin. To understand how Merlin normally controls EGFR activity, we examined EGFR signaling as wild-type cells reach high density. Confluent epithelial cells of breast and kidney origin become unresponsive to stimulation by EGF and other growth factors via a cadherin-dependent mechanism (; ). Accordingly, we found that while acute EGF stimulation of confluent Nf2-expressing cells does induce tyrosine phosphorylation of the receptor itself, activation of EGFR effectors such as Src and Raf does not increase, suggesting that signal propagation from the activated EGFR is prevented once wild-type cells reach high cell density (). cells results in a slight increase in the already elevated membrane phosphotyrosine content (). Reintroduction of Nf2 but not Nf2 restored the block of EGFR signaling at high cell density (). Merlin does not appear to be a general inhibitor of RTK activation because signaling from the IGF-I receptor, including its ability to transactivate the EGFR (), occurs in confluent MEFs regardless of the status (). Induced phosphorylation of EGFR without activation of downstream targets in confluent, Nf2-expressing cells suggested that in this context the ability of the activated receptor to acquire signaling competence might be physically restricted by Merlin at cell confluence. Because Merlin is membrane associated and internalization of liganded EGFR is intimately linked to its signaling output (), we asked whether the absence of Merlin had any effect on the surface levels of EGFR. OBs. cells, consistent with an increased rate of activation and internalization (). To further explore the role of Merlin in EGFR membrane localization we used Texas red–conjugated EGF (Tr-EGF) to visually track the EGFR in the presence and absence of Merlin. For these experiments we chose to use epithelial LDCs that are amenable to immunofluorescence localization analyses. LDCs we found that within 30 min after stimulation Tr-EGF localizes to intracellular vesicles in virtually every cell, consistent with ligand-activated EGFR internalization ( and Fig. S3 A, available at ). LDCs to neighboring cells into which Nf2 or Nf2 were reintroduced. Tr-EGF internalization was prevented by expression of Nf2 that, like EGFR, is enriched at cell–cell boundaries ( and Fig. S3 B), but not by Nf2, which is excluded from cell–cell boundaries (; ). An identical response was seen after basolateral exposure to Tr-EGF (not depicted). Internalization of fluorescent transferrin proceeded similarly in the presence or absence of Merlin (Fig. S3 C). Importantly, Nf2 did not prevent Tr-EGF internalization from the free edge of cells bordering a scrape wound or small colony (; not depicted), consistent with the hypothesis that Merlin limits EGFR internalization specifically upon cell–cell contact. In fact, disruption of cadherin-based intercellular adhesion by EGTA/Ca depletion resulted in the appearance of internalized Tr-EGF () and increased EGFR signaling in Nf2-expressing cells (). Importantly, endogenous levels of Merlin also prevented EGFR internalization in similar mosaic cultures (Fig. S3 D). These results suggest that upon cell contact Merlin functions to physically restrict ligand-activated EGFR from signaling. This interpretation is supported by the altered distribution of pEGFR in fractionated Triton-insoluble membranes in the absence of Merlin (). Although EGF stimulation of confluent wild-type cells yields the appearance of pEGFR that is confined to higher density fractions (II and III) that also contain Merlin, both pEGFR and Merlin are excluded from fraction I (). cells, a substantial pool of pEGFR appears in the lowest density fraction (I; ). These results suggest that the physical state of pEGFR is altered in the absence of Merlin. Merlin, EGFR, and AJ components are all normally enriched in the Triton-insoluble membrane fraction, a poorly defined biochemical compartment enriched in signaling molecules and cytoskeletal components and variously referred to as detergent-resistant membranes, lipid rafts, cholesterol-rich domains, etc. (; ; ; ). Notably, the membrane distribution of Rac and RhoGDI, two proteins implicated in Merlin function, is unaffected by the absence of Merlin; in fact, in contrast to a recent report (), we do detect recruitment of Rac to detergent-resistant membranes in both the presence and absence of Merlin (; not depicted). Consistent with this interpretation, reintroduction of Nf2 but not Nf2 alters the solubility of EGFR in confluent LDCs (Fig. S4 A, available at ). As shown in and , the propagation of signaling from the activated EGFR to its downstream targets is blocked in confluent Nf2-expressing cells. Therefore, we asked whether Merlin directly interferes with the ability of ligand-activated EGFR to interact with its canonical signaling effectors. LDCs causes EGFR to interact with Cbl, Grb2, Sos, and PLCγ; however, these interactions do not occur in the presence of Merlin (). Reintroduction of Nf2, but not Nf2, prevents EGFR association with its immediate effectors in response to EGF despite phosphorylation of the EGFR itself (); in fact, we do not detect changes in the responsiveness of EGFR to EGF ligand (Fig. S4 B). Importantly, under these conditions wild-type Merlin, but not Nf2, physically associates with the EGFR (). Altogether, these data suggest that Merlin prevents EGFR from interacting with its immediate targets by sterically hindering the interaction and/or by sequestering the EGFR into a non-signaling membrane compartment from which both access to its downstream effectors and internalization are impeded. These data also indicate that Merlin acts at a step that precedes endocytosis of the activated EGFR. Indeed, EGF-induced Src activation and EGFR interaction with Grb2 and Cbl, early events that are required for EGFR internalization, do not occur in confluent Nf2-expressing cells (; ; ). NHE-RF1 is a PDZ domain–containing adaptor that interacts with Merlin and the ERM proteins (; ; for review see ) and is thought to play an important role in controlling the surface availability of certain membrane receptors including the β-adrenergic receptor and cystic fibrosis transmembrane conductance regulator (for review see ). Importantly, recent studies indicate that NHE-RF1 can also interact with and alter the surface availability of the EGFR (). This raises the possibility that Merlin regulates the surface availability of EGFR via NHE-RF1. To determine whether NHE-RF1 mediates the association between Merlin and EGFR, we performed shRNA-mediated knockdown of NHE-RF1 expression in Nf2-expressing LDCs. Lentiviral expression of a shRNA targeting NHE-RF1 revealed that reduced NHE-RF1 expression nearly eliminated the association of Nf2 and EGFR (). In contrast, Nf2 associates with Ezrin regardless of the level of NHE-RF1 expression (). Importantly, NHE-RF2 does not detectably associate with EGFR in these cells and shRNA-mediated knockdown of NHE-RF2 expression has little effect on the association between Merlin and EGFR (; not depicted). These data suggest that Merlin–EGFR association is mediated specifically by NHE-RF1. Merlin localizes to AJs and is required for AJ stabilization (). Our previous studies suggest that upon cell–cell contact, Merlin is recruited to and activated at nascent AJs; indeed, Merlin also associates with E-cadherin in epithelial cells (). The simplest interpretation of our data is that active, cadherin-associated Merlin “captures” the NHE-RF1–EGFR complex, thereby retaining it. Consistent with this hypothesis, we found that the association between EGFR and E-cadherin in confluent Nf2-expressing cells is NHE-RF1 dependent, whereas the association of Merlin with E-cadherin is NHE-RF1 independent (). A key prediction of this model is that the association between Merlin and both NHE-RF1 and EGFR is dependent on cell–cell contact; indeed, as shown in , the association between Merlin and NHE-RF1 or EGFR is dramatically enhanced with increasing cell density. In contrast, the association between EGFR and NHE-RF1 is not adhesion dependent (Fig S5 A, available at ). Importantly, after acute disruption of intercellular contacts by Ca depletion, Merlin rapidly dissociates from EGFR (), indicating that cell–cell adhesion is a strict determinant for EGFR–Merlin association. cells, we treated MEFs, OBs, and LDCs with pharmacologic EGFR inhibitors. cells of all three cell types in the presence of serum (; not depicted). The specificity of each compound was demonstrated by its ability to block EGF- but not PDGF-induced membrane pTyr (; not depicted). MEFs, OBs, and LDCs in the presence of serum (; not depicted), indicating that proliferation of non-contacting cells can be sustained by serum-derived signals other than those mediated by EGFR. cells can undergo contact-dependent inhibition of proliferation if EGFR activity is blocked. cells because withdrawal of the inhibitor led to reentry into the proliferative state despite the high cell density. These results suggest that EGFR inhibitors, already in clinical use for several human cancers, could also be of therapeutic benefit for NF2-deficient tumors. cells rather than the high levels driven by oncogenic EGFR mutations. Notably, signaling via EGFR family members (ErbBs) is critical for the proliferation and survival of Schwann cells, the principle target of NF2-associated tumorigenesis in humans (). The discovery, in 1993, that the NF2 tumor suppressor, Merlin, is a member of a family of membrane/cytoskeleton-associated proteins suggested a novel mechanism of tumor suppression (; ). Amidst the identification of many Merlin-interacting proteins and Merlin-controlled activities, a clear role for Merlin in controlling contact-dependent inhibition of proliferation has emerged (; ; ). Loss of contact-dependent inhibition of proliferation is a signature of cell transformation, but the molecular basis of this phenomenon is not known. Our previous work identified a role for Merlin in stabilizing AJs between cells, but did not pinpoint the mitogenic signal that drives proliferation in the absence of Merlin and normal AJs (). An intimate relationship between RTK activity and AJ stability clearly exists, but its molecular underpinnings are only beginning to emerge (for review see ). We have now found that Merlin directly interferes with EGFR signaling in a contact-dependent manner, providing key insight into the molecular basis of contact-dependent inhibition of proliferation and directly linking the functions of a novel tumor suppressor and a well-known oncogene. cells, together with the localization of EGFR to AJs, suggests that Merlin normally coordinates the processes of AJ stabilization and negative regulation of the EGFR by establishing their interdependence as they occur. Our data are consistent with a model wherein the following sequence of events occurs (): Merlin is recruited to nascent AJs () where it is activated and begins to stabilize the developing junctions and sequester a pool of EGFR into a compartment from which it does not have access to its immediate downstream targets and cannot be internalized. Whether Merlin is in an “open” or “closed” conformation at this stage remains to be established; although many studies suggest that “active” Merlin is not phosphorylated at serine 518, it is not clear whether this hypophosphorylated form of Merlin is consistently self-associated. The ensuing localized reduction of EGFR tyrosine kinase activity at contact sites, in turn, may further stabilize AJs, perhaps via altering the phosphorylation of AJ components such as p120 (see ) and/or the activity of Rac. Indeed, while initial stages of AJ formation are accompanied by increased tyrosine kinase activity and activation of the small GTPase Rac, a classic target of EGFR signaling that mediates dynamic membrane/cytoskeletal remodelling at cell–cell interfaces (; ; ), later stages are often associated with negative regulation of both EGFR and Rac (). This could explain previously reported roles for Merlin in negatively regulating both Rac and its effector Pak and in modulating actin cytoskeleton remodelling (; ; ; ). Although the precise mechanism whereby Merlin associates with the actin cytoskeleton is not yet clear, we have found that association with the cortical actin cytoskeleton is necessary for the growth-suppressing and EGFR-inhibiting activities of Merlin (Cole, B.K., personal communication). Molecularly, our studies indicate that Merlin directly links the AJ and EGFR. Indeed, the trilobed structure of the four-point-one, ERM (FERM) domain appears well-designed for assembling multiple proteins (). We found that the association between Merlin and EGFR is mediated by the tandem PDZ domain–containing adaptor NHE-RF1, which is known to interact with the third lobe of the ERM FERM domain (; ). NHE-RF1 can associate with and is thought to regulate the surface abundance of several different receptors (for review see ). In fact, it has been reported that NHE-RF1 stabilizes and slows the down-regulation of surface EGFR (); however, neither this nor other studies have explored how functional specificity is applied to such a wide range of receptor interactions, and it is not clear how temporal and spatial regulation of NHE-RF1–associated receptors is achieved. Our studies suggest that Merlin confers one such level of specificity by locally engaging NHE-RF1–associated EGFR at the AJ. The ERM proteins also interact with NHE-RF1 and are required for stable apical localization of NHE-RF1 in the intestinal epithelium (). However, the ERM proteins likely engage a distinct subset of NHE-RF–associated receptors. Consistent with the tendencies of the ERM proteins and Merlin to be apically or apical-junctionally concentrated, respectively, the ERM proteins appear to be functionally dedicated to the apical membrane and Merlin to the junctional domain (). By analogy to the roles of Merlin in both stabilizing the association between adhesion proteins and the actin cytoskeleton and locally “capturing” NHE-RF1–EGFR complexes, Ezrin may stabilize the association between the apical membrane and cortical cytoskeleton while capturing apical NHE-RF receptor complexes (). Indeed, Ezrin is required for establishing or maintaining the integrity of the apical surface of intestinal epithelial cells in vivo (). In contrast to Merlin, Ezrin does not associate with EGFR or E-cadherin, mediate contact-dependent inhibition of proliferation, or effect EGFR internalization/signaling in the cells studied here (Fig. S5 B), and overproliferation is not detected in the ERM-deficient intestine (). It is clear that AJs are continuously remodelled both in confluent monolayers in culture and in tissues in vivo. In vivo, all cells in solid tissues are in contact and junctional remodeling and cell proliferation must be exquisitely coordinated. Localization to intercellular adhesions may render the EGFR uniquely able to sense and modulate changes in cell contact and to fine-tune its activity accordingly. Our molecular model of Merlin-mediated coordination of AJ stabilization and EGFR down-regulation provides ample opportunity for flexibility. For example, phosphorylation or phospholipid binding may alter Merlin self-association and/or membrane distribution, coordinately affecting junction stability and EGFR signaling. Indeed, both hypo- and hyperphosphorylated forms of Merlin are associated with EGFR (, ), suggesting that S518 phosphorylation may regulate Merlin-associated EGFRs. We found that the association of Merlin with E-cadherin and with NHE-RF1–EGFR are both contact dependent; however, it is interesting to note that while the association of Merlin with E-cadherin and NHE-RF1 is maintained after acute disruption of intercellular contacts (not depicted), the association between Merlin and EGFR is rapidly lost, suggesting disengagement of NHE-RF1–EGFR in this specific context. Our studies also indicate that the status of cell–cell contact has profound implications for the propagation of EGFR signaling. Conversely, in the context of EGFR-driven tumorigenesis, a critical line of investigation will be to determine whether oncogenic variants of the EGFR can evade the contact-dependent inhibition of signaling imposed by Merlin. cells in the presence of serum, suggesting that EGFR activation is necessary and sufficient to cause these phenotypes. This also suggests a novel avenue of therapeutic exploration for NF2. However, given that NHE-RF1 can associate with multiple receptors, Merlin may well affect other receptors by a similar mechanism. In fact, the results presented here are complementary to those of , who reported altered surface availability of EGFR and other membrane receptors in tissues lacking both Merlin and the related tumor suppressor, Expanded. Similarly, Merlin could coordinate regulation of EGFR or other receptors with alternative adhesion receptors such as CD44; it has been proposed that Merlin mediates contact-dependent inhibition of proliferation via CD44 in other cell types (). Indeed, the theme of Merlin-mediated coordination of cell adhesion and membrane receptor signaling is echoed by recent work in suggesting that Merlin inhibits signaling through the Hippo/Warts/Yorkie pathway (; ), corresponding to the conserved Mst/Lats/Yap pathway in mammals. Activation of this pathway in response to extracellular signals appears to be coordinately regulated by Merlin and Expanded, which signals from the Fat cadherin receptor (; ; ). However, neither the source of that extracellular signal nor the signaling receptor(s) involved have been identified in mammals or flies (). The data presented here indicate that Merlin could regulate signaling through this pathway by directly coordinating EGFR signaling output with cadherin-dependent intercellular adhesion. primary MEFs were prepared as described previously (). Wild-type primary OBs were prepared from calvaria of newborn mice as described previously (). deletion in OBs was achieved via adenoviral expression of the Cre-recombinase as we described for MEFs (). Primary MEFs and OBs were used between passages 3 and 6. LDCs were derived by liver-specific, deletion of in vivo by crossing mice to transgenic mice () (; Jackson Laboratories). In brief, the liver of a 12-wk-old mouse was removed, minced, dissociated in Liver Dissociation Medium (Invitrogen), and cultured in 10% FBS-DME. Wild-type epithelial embryonic liver cells were derived from the liver of a day-14.5 embryo as described by , and subsequently adapted to the standard growth conditions used for all other cell lines. Clonal cell lines were established by limiting dilution. The generation and use of adenoviral vectors expressing Nf2 and Nf2 have been described previously (). Primary antibodies against the following antigens were from Upstate Biotechnology (active-β-catenin: ABC, 05–665, 1:1,000 dilution); Transduction Laboratories (pTyr: RC20, 610023; β-catenin: 610153; E-cadherin: 610182; p120ctn: 610133; p120ctn-pY228: 612536; Caveolin-pY14: 611338; c-Cbl: 610441; Grb2: 610111; Sos1: 610095; PLCγ: 610027; Rac1: 612652; all at 1:1,000 or 2,000 dilution); Santa Cruz Biotechnology, Inc. (Merlin: sc331, 1:40,000 dilution; EGFR: sc1005); Cell Signaling (EGFR-pY845: 2231; EGFR-pY992: 2235; EGFR-pY1068: 2234; STAT3-pY705: 9131; STAT5-pY694: 9351; Shc-pY239/240: 2434; MAPK-pT202/Y204: 5120; AKT-pS473: 9271; Raf-pS259: 9421, all at 1:1,000 dilution); Biosource International (Src: 44–656; Src-pY418: 44–660, used at 1:1,000); Abcam (EBP50/NHERF-I: ab3452); NeoMarker (EGFR: Ab17; Ezrin: Ab1); Sigma-Aldrich (actin: A-2547). Monoclonal anti-Merlin 1C4 (a gift of Vijaya Ramesh, Massachusetts General Hospital, Boston, MA) was used at 1:1,000. Polyclonal anti-NHE-RF2 (B70; gift of Anthony Bretscher, Cornell University, Ithaca, NY) was used at 1:1,000. HRP-conjugated secondary antibodies to rabbit, mouse, or rat were from GE Healthcare. Equal protein amounts of total cell and membrane extracts were analyzed by Western blot as described previously () with one modification: the membrane pellet was directly solubilized in RIPA buffer containing 0.5% SDS. For density-gradient separation, postnuclear membrane pellets from three 150-mm dishes of late confluent MEFs were lysed on ice for 30 min in Triton-lysis buffer (1% Triton X-100, 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, and 10 mM β-glycerophosphate plus protease and phosphatase inhibitors), resuspended in buffer A (250 mM sucrose, 1 mM EDTA, 20 mM tricine, and 1% Triton X-100 and inhibitors) containing 35% Optiprep, sequentially overlayed with 25, 15, 5, and 0% Optiprep-buffer A and centrifuged for 3.5 h at 160000 . Fractions collected at the interfaces I = 0–5%, II = 5–15%, III = 15–25%, IV = 25–35%, and the pellet V > 35%, were resuspended in RIPA buffer and analyzed (7 μg/lane) by immunoblotting. LDCs plated on glass coverslips were infected with Ad.Nf2 or Ad.Nf2 when ∼50% confluent. After 4–5 d, confluent monolayers were serum starved in 1% BSA in DME for 2 h, incubated for 30 min at 37°C with 2 μg/ml Tr-EGF (E3480; Molecular Probes) or 10 μg/ml Alexa Fluor 488-Transferrin (T13342; Molecular Probes), and fixed in 4% PFA-cytoskeletal buffer (10 mM MES, pH 6.3, 2 mM EGTA, 3 mM MgCl, and 138 mM KCl) for 15 min at room temperature. After permeabilization in 0.2% Triton X-100, cells were incubated with primary anti-NF2 antibody (sc331; 1:300 in 1% BSA-PBS) overnight at 4°C. After incubation with FITC- or rhodamine-conjugated anti–rabbit secondary antibody (Jackson Immunoresearch Laboratories; 1:200), coverslips were mounted with Vectashield (Vector Laboratories). To create noncontacting free edges in Ad.Nf2-infected LDCs, monolayers growing in 10% FBS-DME were scrape-wounded with a pipet tip, allowed to recover for ∼6 h, and starved (2 h) before adding Tr-EGF as described above. To disrupt intercellular adhesion by depletion of extracellular Ca, monolayers were serum starved for 2 h, washed twice in Ca-free DME (Invitrogen), incubated in 2 μg/ml Tr-EGF, 1% BSA, and 5 mM EGTA in Ca-free DME and fixed as above at the indicated time points. Images were acquired using a 63× 1.4NA oil objective lens (Carl Zeiss MicroImaging, Inc.) on an Axioplan microscope (Carl Zeiss MicroImaging, Inc.) with IP Lab software and a Sony CCD camera. Final images were prepared using Adobe Photoshop 7.0. Late confluent MEFs or OBs were serum starved overnight in DME, shifted to 4°C, rinsed twice in cold PBS and incubated for 1 h with 0.5 μg/ml EZ-Link Sulfo-NHS-LC-Biotin (Pierce Chemical Co.) in PBS. After quenching the reaction (50 mM NHCl, 1 mM MgCl, and 0.1 mM CaCl for 10 min) and rinsing in PBS, cells were returned to 10% FBS-DME at 37°C and lysed in Triton-lysis buffer containing 60 mM Octylglucoside (-Octyl-β--glucopyranoside; Calbiochem) at the indicated time points. Normalized extracts (600 μg total protein/400 μl) were precleared with protein A–Sepharose prebound to normal rabbit IgG for 2 h at 4°C. Anti-EGFR antibody (Ab17, NeoMarker; 8 μg/sample) or streptavidin-coupled agarose beads (50 μl; Pierce Chemical Co.) were added to precleared extracts and incubated overnight at 4°C. The following day, EGFR-containing immunocomplexes were precipitated with protein A–Sepharose beads (40 μl, 2 h, at 4°C). Beads from either immunoprecipitation or biotin pull-down were washed five times in the above buffer and boiled 5 min in 2× sample buffer. Complexes were separated by 8% SDS-PAGE and analyzed by Western blot as described above. Biotinylated-immunoprecipitated EGFR was detected with HRP-conjugated streptavidin. Immunoprecipitations of EGFR (Ab17; 8 μg) and E-cadherin (3 μg) from LDCs were from total membrane extracts in the above Triton-Octylglucoside buffer (800 μg total protein/400 μl). To disrupt intercellular adhesion by depletion of extracellular Ca, confluent monolayers were washed twice in PBS/5 mM EGTA, and incubated in Ca-free DME (Invitrogen) containing 10% Ca-chelated FBS for 45 min. MEFs, OBs (5 × 10) or LDCs (7.5 × 10) were seeded in triplicate 15-mm wells in 5% FBS-DME. The following day, 1 μM Gefitinib (Iressa; AstraZeneca) or 0.5 μM Compound 56 (Calbiochem) were added to culture wells; cells were trypsinized and counted every other day. Fresh medium with or without inhibitors was added each day of counting. For drug withdrawal, 1 μM Gefitinib was added to LDCs daily until day 5 post-seeding, when half of the wells were returned to 5% FBS-DME only. Beginning that day, cells were counted and fresh medium with or without inhibitor was added to the remaining wells every other day. Several shRNA constructs against NHE-RF1 (#68583-68587) and NHE-RF2 (#68613-68615, 68617) in the lentiviral pLKO.1 vector were obtained from Open Biosystems and tested for NHE-RF1/2 knockdown in LDCs; #68587 and #68617 were used for experiments. Lentiviral production and infection was performed as described previously (). Fig. S1 shows level and distribution of adhesion molecules in various cell types used in this work. Fig. S2 defines confluence states in mesenchymal cells and shows transmission electron micrographs that reveal restoration of electrondense AJs upon expression of Nf2 in LDCs. Fig. S3 shows internalization of fluorescent EGF and transferrin in LDCs, EGFR localization in LDCs, and inhibition of fluorescent EGF internalization in confluent embryonic liver cells expressing endogenous . Fig. S4 shows decreased EGFR solubility in the presence of Merlin and a similar dose dependency of EGFR auto-phosphorylation in the presence or absence of Merlin. Fig. S5 shows that the association of endogenous EGFR and NHE-RF1 is contact independent and reveals the lack of association between Ezrin and EGFR or Ezrin and E-cadherin in LDCs. Online supplemental material is available at .

Protease-activated receptor-1 (PAR1), a G protein–coupled receptor (GPCR) for thrombin, is important for hemostasis and thrombosis, inflammation, embryonic development, and cancer progression (; ). Unlike most GPCRs, PAR1 is irreversibly activated by proteolysis. The PAR1 N terminus is cleaved by thrombin, which unmasks a new N terminus that acts as a tethered ligand and binds intramolecularly to the receptor to trigger transmembrane signaling (). Synthetic peptides that mimic the newly formed N terminus can activate PAR1 independently of proteolysis. Because of the irreversible proteolytic nature of PAR1 activation, rapid desensitization and receptor trafficking tightly regulate PAR1 signaling (; ; ). PAR1 displays two modes of trafficking that are important for the regulation of receptor signaling. Unactivated PAR1 constitutively cycles between the cell surface and an intracellular compartment, generating an intracellular pool of uncleaved receptor that replenishes the cell surface after thrombin exposure and leads to rapid resensitization to thrombin signaling independent of de novo receptor synthesis (; ). Unlike most GPCRs, which internalize and recycle, activated PAR1 is internalized, sorted directly to lysosomes, and degraded (; ). Sorting of activated PAR1 to lysosomes is critical for signal termination (). Constitutive and agonist-induced PAR1 internalization are clathrin and dynamin dependent (). However, in contrast to most GPCRs, neither constitutive nor activated PAR1 internalization requires arrestins (). Arrestins interact with clathrin and adaptor protein complex-2 (AP2) to facilitate the internalization of activated GPCRs through clathrin-coated pits (; ). We recently showed that AP2 and not arrestins is critical for PAR1 constitutive internalization and is essential for the cellular recovery of thrombin signaling (). Interestingly, activated PAR1 internalization through clathrin-coated pits is independent of AP2, suggesting that constitutive and activated receptor internalization require different endocytic machinery. The mechanisms that regulate activated PAR1 internalization through clathrin-coated pits is not known. Ubiquitin (Ub) modification of integral membrane proteins can function as an internalization and endosomal sorting signal (). Ub, a 76–amino acid protein, is recognized by Ub-binding domains (UBDs), which are found in proteins of the endocytic sorting machinery. Ubiquitination regulates internalization of the yeast Ste2 and Ste3 GPCRs. Studies using yeast strains that lack specific Ub-conjugating enzymes and Ub-defective Ste2 mutants or chimeras indicate that monoubiquitination is both necessary and sufficient for constitutive and agonist-induced receptor internalization (; ). In contrast, recent studies suggest that mammalian GPCR ubiquitination is essential for lysosomal sorting but not for receptor internalization (; ). Direct β-adrenergic receptor (βAR) ubiquitination is not required for internalization but regulates activated receptor lysosomal sorting and degradation (). Similar to βAR, ubiquitination of chemokine receptor 4 (CXCR4) is essential for agonist-promoted receptor lysosomal degradation but not for internalization (). Although ubiquitination does not have a direct role in mammalian GPCR internalization, it has been shown to function indirectly. Indeed, activation-dependent ubiquitination of arrestins is required for βAR internalization (). However, the function of ubiquitination in the regulation of mammalian GPCRs that do not require arrestins for endocytosis is not known. We have shown that constitutive and agonist-induced PAR1 internalization is clathrin and dynamin dependent and independent of arrestins (). We recently found that the clathrin adaptor AP2 is critical for constitutive but not for agonist-induced PAR1 internalization (). Given these observations and the previous findings that Ub regulates yeast GPCR internalization (; Terrel et al., 1998), we examined the function of ubiquitination in PAR1 trafficking. Our findings here reveal a novel role for ubiquitination in the negative regulation of PAR1 constitutive internalization and in specifying a distinct clathrin adaptor requirement for activated receptor internalization. To define the importance of the posttranslational modification of lysine residues with Ub in the regulation of PAR1 trafficking, we constructed a PAR1 lysineless mutant by replacing the 10 intracytosolic lysine (K) residues with arginines, which is designated PAR1 0K (). To ensure that the PAR1 0K mutant was not globally disrupted in receptor function, the capacity of the activated receptor to promote Gα-stimulated phosphoinositide hydrolysis was examined. HeLa cells expressing comparable amounts of cell surface PAR1 wild type and 0K mutant were incubated with various concentrations of thrombin, and the amounts of [H]inositol phosphates (IPs) formed were then measured (). Both PAR1 wild type and 0K mutant were equally effective at stimulating [H]IP accumulation as well as at inducing a maximal effect at saturating concentrations of thrombin. The ability of the PAR1 0K mutant to couple to G protein activation like the wild-type receptor indicates that receptor function is intact. We next determined whether the receptor was modified with Ub by directly comparing the ubiquitination status of PAR1 wild type to the lysineless 0K mutant. HEK293 cells transiently coexpressing similar amounts of PAR1 wild type or 0K mutant together with HA-tagged Ub and dynamin K44A were incubated in the absence or presence of agonist peptide for 10 min at 37°C. Cells were lysed, PAR1 was immunoprecipitated, and the extent of receptor ubiquitination was determined. Strikingly, in untreated cells, wild-type PAR1 displayed a substantial amount of basal ubiquitination that was detected as prominent high molecular weight species migrating above ∼107 kD (, lane 1), which is consistent with the addition of multiple HA-tagged Ubs. Remarkably, incubation with agonist decreased wild-type PAR1 ubiquitination (, lane 2). In contrast to wild-type receptor, ubiquitinated species of PAR1 0K were undetectable in both untreated or agonist-treated cells (, lanes 3 and 4). These findings suggest that PAR1 is basally ubiquitinated and that activation promotes deubiquitination. To investigate the function of ubiquitination in PAR1 trafficking, we examined the constitutive and agonist-induced loss of cell surface wild-type and Ub-deficient PAR1 by ELISA. HeLa cells and Rat1 fibroblasts stably expressing similar amounts of surface PAR1 wild type or 0K mutant were incubated with M1 anti-FLAG antibody for 1 h at 4°C to label cell surface receptors and were treated with or without agonist for various times at 37°C, and the amount of receptor remaining on the cell surface was then quantified by ELISA. In wild-type PAR1–expressing cells, agonist induced rapid receptor internalization within 10 min, and the receptor continued to slowly internalize, leading to an ∼70% loss of surface PAR1 after 30 min (). PAR1 0K mutant internalization was comparable with wild-type receptor after agonist exposure in both cell types (). We next examined the constitutive internalization of wild-type PAR1 and observed a slow rate of internalization resulting in a 10–20% loss of cell surface receptor after 30 min (), which is consistent with that previously reported for these cell types (; ). In contrast, Ub-deficient PAR1 0K mutant displayed an increased rate of constitutive internalization in which 50–60% of receptor was lost from the cell surface after 30 min of incubation in both cell types (). Immunofluorescence microscopy studies also revealed a substantial amount of internalized Ub-deficient PAR1 0K mutant in early endosomes compared with wild-type receptor even without agonist exposure or antibody prebinding (), suggesting that the absence of ubiquitination enhances PAR1 constitutive internalization. In contrast, both PAR1 wild type and 0K mutant showed robust internalization after agonist exposure (). To confirm the enhanced constitutive internalization of Ub-deficient PAR1, we used a receptor-bound antibody capture assay as a measure of receptor internalization. HeLa cells expressing similar amounts of surface PAR1 wild type or 0K mutant were labeled with antibody at 4°C and incubated in the absence or presence of agonist for the indicated times at 37°C. After incubation, antibody bound to cell surface receptor was stripped, cells were lysed, and the amount of internalized receptor-bound antibody was quantified by ELISA. In wild-type PAR1–expressing cells, ∼10% of antibody that initially bound to cell surface receptor was internalized at the steady state (), whereas virtually no internalized antibody was detected in untransfected cells (not depicted). Remarkably, PAR1 0K mutant displayed an increased rate of constitutive internalization, with ∼50% of receptor-bound antibody internalized after 15 min (). Agonist caused a robust increase in PAR1 wild-type internalization, whereas activated PAR1 0K mutant internalization was comparable with constitutive internalization (). A minimal amount of constitutively internalized wild-type PAR1 was detected in endosomes by immunofluorescence microscopy, whereas PAR1 0K mutant showed substantial redistribution to endosomes in the absence of agonist (). The addition of agonist triggered comparable wild-type and Ub-deficient PAR1 internalization (). Together, these findings provide further evidence that the absence of ubiquitination increases PAR1 constitutive internalization. To define the lysine residues that are important for PAR1 internalization, we examined mutants in which lysines in the first or third intracytosolic loops or cytoplasmic tail (C tail) were mutated to arginines. A mutant PAR1 in which the four C-tail lysine residues were replaced with arginines (4K/R) displayed an enhanced rate of constitutive internalization similar to Ub-deficient PAR1 (), whereas PAR1 intracytosolic loop lysine mutants internalized like wild-type receptor (not depicted). However, the extent of constitutive internalization displayed by PAR1 4K/R was less than that observed with PAR1 0K mutant, indicating that in addition to PAR1 C-tail lysines, other lysine residues or regulatory domains may contribute to constitutive internalization. To test the role of individual C-tail lysines in PAR1 internalization, we constructed receptor mutants in which K, K, and K (3K/R) or K and K (2K/R) were converted to arginines (). The rate of PAR1 3K/R and 2K/R constitutive internalization was increased substantially compared with wild-type receptor (), suggesting that the distal K and K are important for constitutive internalization. PAR1 3K/R and 2K/R also showed marked redistribution to endosomes compared with wild-type receptor in immunofluorescence microscopy studies (), whereas agonist caused a similar increase in wild-type and mutant receptor internalization (). These data suggest that K and K residues are critical for the regulation of PAR1 constitutive internalization. We next examined whether PAR1 C-tail lysines were important for receptor ubiquitination. HEK293 cells transiently coexpressing HA-tagged PAR1 wild type, 4K/R, and 2K/R together with FLAG-Ub and dynamin K44A were treated with or without agonist for 10 min at 37°C, and the extent of receptor ubiquitination was examined. In untreated cells, the major ubiquitinated PAR1 wild-type species migrated above ∼107 kD, whereas minor species were detected below ∼94 kD (, lane 1), which is consistent with multiple ubiquitinated PAR1 species. Strikingly, the addition of agonist caused a marked decrease in wild-type PAR1 ubiquitination (, lane 1). In contrast to wild-type receptor, basal ubiquitination of both PAR1 4K/R and 2K/R C-tail mutants was substantially reduced irrespective of agonist addition (, lanes 3–6). Together, these data suggest that the PAR1 C-tail K and K residues are the major sites for receptor ubiquitination. Our results suggest an important regulatory role for the ubiquitination of C-tail lysines in PAR1 internalization. Therefore, we tested whether constitutive internalization was affected when Ub was fused in frame to the C tail of the PAR1 0K Ub-deficient mutant. We used a modified Ub in which K was mutated to arginine (K48R), and the two terminal glycine residues (ΔGG) were deleted to avoid extensive modification of Ub and to increase the efficiency of PAR1 0K–Ub chimera expression at the cell surface. We first examined PAR1 wild-type and mutant expression in transfected HeLa cells by immunoblotting using anti-PAR1 antibodies. As expected, PAR1 wild type and 0K mutant appeared as one broad major transfection–dependent band migrating between ∼64 and 98 kD (), which is indicative of posttranslational glycosylation of the receptor protein as previously reported (). The PAR1 0K–Ub chimera migrated as two high molecular weight species, which is consistent with the appearance of multiple ubiquitinated receptor species (). PAR1 0K mutant displayed an increased rate of constitutive internalization (). In contrast, the attachment of Ub to the C tail of the PAR1 0K mutant reduced the rate of constitutive internalization comparable with that observed with wild-type PAR1 (). The addition of agonist induced similar increases in PAR1 wild type, 0K mutant, and 0K–Ub chimera internalization, indicating that activated PAR1 0K–Ub internalization remained intact (). Immunofluorescence microscopy studies were consistent with a marked inhibition of PAR1 0K–Ub chimera constitution internalization compared with Ub-deficient PAR1, whereas agonist induced a comparable increase in wild-type and mutant receptor internalization (). Together, these data suggest that modification of PAR1 0K with Ub negatively regulates constitutive internalization. The endosomal accumulation of Ub-deficient PAR1 may be caused by an increased rate of constitutive internalization as well as to an inability of the PAR1 0K mutant to recycle back to the cell surface. To address this possibility, we examined the internalization and recycling of wild-type and Ub-deficient PAR1. HeLa cells expressing PAR1 wild type or 0K mutant were labeled with antibody at 4°C and incubated for 30 min at 37°C to facilitate constitutive internalization. After incubation, surface-bound antibody was stripped, and the recovery of previously internalized receptor-bound antibody was followed for various times. An initial 30-min incubation at 37°C caused an ∼25% decrease in the amount of PAR1 wild type, whereas a substantially greater amount (∼50%) of PAR1 0K mutant was initially lost from the cell surface (, ), which is consistent with an increased rate of constitutive internalization. In both PAR1 wild-type– and 0K mutant–expressing cells, a comparable amount of previously internalized receptor recycled back to the cell surface (). These results are consistent with the previously described extent of constitutive internalization and recycling of PAR1 () and indicate that PAR1 recycling is not attenuated with Ub-deficient PAR1. We next examined whether Ub-deficient PAR1 internalized through a clathrin- and dynamin-dependent pathway like wild-type receptor (). HeLa cells expressing PAR1 wild type or 0K mutant were transiently transfected with GFP-tagged dynamin wild type, K44A, or vector, and the amounts of cell surface receptor at steady state were measured. In PAR1 wild-type–expressing cells, neither dynamin wild-type nor K44A mutant expression affected the amounts of cell surface receptor expression when compared with vector (). The amount of surface PAR1 0K mutant was also comparable in dynamin wild-type and vector-transfected cells. In contrast, the coexpression of dynamin K44A increased PAR1 0K expression detected on the cell surface compared with wild-type receptor or vector (). In immunofluorescence microscopy studies, the constitutive internalization of PAR1 wild type and 0K mutant were virtually abolished in K44A-expressing cells, whereas in adjacent untransfected cells, PAR1-containing endosomes were clearly evident (, arrowheads). Internalization of activated PAR1 wild type and 0K mutant were similarly inhibited by dynamin K44A (, arrowheads), suggesting that wild-type and Ub-deficient PAR1 internalization are dynamin dependent. To determine the role of clathrin in Ub-deficient PAR1 internalization, we used siRNA targeting the clathrin heavy chain (CHC) to deplete HeLa cells of endogenous clathrin. The expression of CHC was considerably decreased in cells transiently transfected with siRNA specifically targeting CHC compared with nonspecific control (, inset). In control siRNA-transfected cells, PAR1 wild-type and Ub-deficient 0K mutant surface levels were comparable. In contrast, the amount of PAR1 0K mutant surface expression was considerably increased in CHC siRNA–transfected cells compared with wild-type receptor, which is consistent with the inhibition of Ub-deficient PAR1 constitutive internalization (). Constitutive and agonist-induced internalization of wild-type and Ub-deficient PAR1 was virtually ablated in CHC siRNA–transfected cells compared with control cells as assessed by immunofluorescence microscopy (). Together, these findings are consistent with a clathrin- and dynamin-dependent regulation of wild-type and Ub-deficient PAR1 internalization. We recently reported that the clathrin adaptor AP2 function is critical for constitutive but not agonist-induced internalization of wild-type PAR1 (). To assess AP2 function in Ub-deficient PAR1 internalization, we used siRNA targeting the μ2 subunit to deplete cells of the endogenous AP2 complex (, A [inset] and C). In PAR1 wild-type–expressing cells, constitutive internalization was completely inhibited in μ2-siRNA–transfected cells compared with nonspecific (ns) siRNA control cells (). The PAR1 0K mutant displayed enhanced constitutive internalization in ns-siRNA control cells, which was virtually abolished in μ2-siRNA–transfected cells (), strongly suggesting a critical role for AP2 in both wild-type and Ub-deficient PAR1 constitutive internalization. In ns-siRNA control cells, agonist caused the substantial internalization of wild-type PAR1 that was partially diminished in μ2-siRNA–transfected cells (), indicating that even in the absence of AP2, activated wild-type PAR1 is capable of internalization. Remarkably, activated PAR1 0K mutant internalization was virtually abolished in μ2-siRNA–transfected cells, suggesting that AP2 function is critical for agonist-induced Ub-deficient PAR1 internalization (). Immunofluorescence microscopy experiments of PAR1 0K mutant–expressing cells were consistent with a critical role for AP2 in activated Ub-deficient receptor internalization. In the absence of agonist exposure, PAR1 wild type and 0K mutant failed to redistribute to endosomes in AP2-depleted cells (). In contrast, agonist peptide caused a marked increase in wild-type PAR1 internalization in μ2-siRNA–transfected cells comparable with siRNA control cells (), which is consistent with an AP2-independent pathway for activated wild-type PAR1 internalization. In contrast, activated PAR1 0K mutant internalization was markedly inhibited in μ2-siRNA–transfected cells (), suggesting a critical role for AP2 in agonist-promoted Ub-deficient PAR1 internalization. Together, these data suggest a novel function for ubiquitination in specifying a distinct clathrin adaptor requirement for activated PAR1 internalization. We next determined whether ubiquitination functions in agonist-induced PAR1 lysosomal degradation by comparing the extent of wild-type versus Ub-deficient PAR1 degradation after prolonged agonist exposure. HeLa cells stably expressing similar amounts of surface PAR1 wild type or 0K mutant were incubated with or without agonist for 90 min at 37°C, and the amount of receptor remaining was then measured. Remarkably, the extent of activated Ub-deficient PAR1 degradation was comparable with that observed with wild-type receptor (), suggesting that PAR1 ubiquitination is not required for lysosomal degradation. To confirm Ub-independent degradation of PAR1, we examined receptor degradation in Rat1 fibroblasts. PAR1 wild-type and 0K mutant exhibited a comparable, ∼60% loss of receptor protein after prolonged agonist exposure (), which is consistent with the extent of PAR1 degradation previously reported in these cell types (; ). These findings suggest a Ub-independent pathway for agonist-induced PAR1 lysosomal sorting and degradation. In this study, we have defined a novel function for ubiquitination in the negative regulation of PAR1 constitutive internalization and in specifying a distinct clathrin adaptor requirement for activated receptor internalization. We demonstrate that PAR1 is ubiquitinated under basal conditions and deubiquitinated after activation. The ubiquitination of PAR1 appears to negatively regulate constitutive internalization. This is supported by our findings that the constitutive internalization of PAR1 is markedly increased in the absence of ubiquitination and is inhibited by the fusion of Ub to the C tail of Ub-deficient PAR1. Several GPCRs, including the yeast Ste2 and Ste3 receptors, have been reported to be basally ubiquitinated at the plasma membrane, similar to PAR1. However, in contrast to PAR1, the ubiquitination of Ste2 and Ste3 receptors promotes constitutive and agonist-induced internalization (; ). The mammalian platelet-activating receptor GPCR has also been shown to be basally ubiquitinated, but the role of ubiquitination in receptor internalization was not examined (). Our studies also suggest that the ubiquitination of PAR1 specifies a distinct clathrin adaptor requirement for activated receptor internalization that is not critically dependent on AP2. We found that AP2 function was required for the constitutive internalization of both wild-type and Ub-deficient PAR1. Strikingly, AP2 function was also critical for the internalization of activated Ub-deficient PAR1, whereas the internalization of activated wild-type receptor occurred in the absence of AP2. We also demonstrate that ubiquitination is not essential for agonist-promoted PAR1 lysosomal degradation. Thus, our findings with PAR1 suggest a novel function for ubiquitination in the regulation of GPCR trafficking in mammalian cells. The ubiquitination of PAR1 is likely a highly dynamic and reversible process, and, under basal conditions, the receptor probably exists as a ubiquitinated and deubiquitinated species. Our findings raise the intriguing possibility that the ubiquitination of PAR1 might affect the ability of AP2 to regulate constitutive internalization. We previously reported that a PAR1 tyrosine-based motif (YKKL) localized at the extreme C terminus directly binds to the μ2 subunit of AP2 using surface plasmon resonance (). Moreover, both the μ2 subunit and the tyrosine-based motif are essential for promoting PAR1 constitutive internalization in multiple cell types. In this study, we show that the highly conserved K and K residues located within the PAR1 C-tail tyrosine-based motif are the major sites of ubiquitination and negatively regulate constitutive internalization, suggesting that receptor ubiquitination at these sites might affect AP2 binding. In addition, the fusion of Ub to the C tail of Ub-deficient PAR1 0K mutant places the Ub moiety within five residues of the tyrosine-based motif, which could also affect AP2 binding and, thereby, diminish constitutive internalization. However, the low micromolar affinity binding of the PAR1 C tail with the μ2 subunit is typical of μ2 subunit weak interactions with proteins bearing tyrosine-based motifs and has precluded our ability to directly test the role of ubiquitination in PAR1 and AP2 interaction in cells using coimmunoprecipitations or pull downs. Thus, we cannot exclude the possibility that other ubiquitination sites or regulatory domains could also contribute to the regulation of PAR1 constitutive internalization. It is also possible that the ubiquitinated PAR1 conformation is simply not compatible with AP2 interaction or that ubiquitinated PAR1 is bound to another protein important for localization at the plasma membrane. Regardless, our findings suggest that PAR1 ubiquitination provides a mechanism to retain the majority of the receptor at the cell surface so that it is readily available for proteolytic activation by extracellular proteases. PAR1 ubiquitination also appears to have a critical role in specifying a distinct clathrin adaptor requirement for activated receptor internalization. Several clathrin adaptors, including epsins and eps15, contain UBDs that recognize ubiquitinated cargo and facilitate clathrin-dependent internalization. Interestingly, the yeast homologues of the mammalian epsins Ent1 and Ent2 contain UBDs and facilitate the endocytosis of ubiquitinated Ste2 receptor (; ). In mammalian cells, epsin is ubiquitinated under basal conditions, which may prevent its interaction with AP2, clathrin, and membrane lipids, and the deubiquitination of epsin appears to enhance its endocytic activity (). Ub may also negatively regulate epsin function by binding to its UBDs intramolecularly, similar to other endocytic adaptor proteins (). In , the deubiquitinating enzyme Fat facets/USP9X regulates Delta/Notch receptor internalization by deubiquitinating Liquid facets, a homologue of epsin, which is consistent with a function for epsin deubiquitination in the regulation of receptor endocytosis (; ). In contrast, the ubiquitination of arrestins is critical for the internalization of certain GPCRs (). However, arrestins are not essential for PAR1 internalization (), suggesting that activated PAR1 internalization may require epsins similar to the Ste2 receptor. However, whether epsin and/or Ub regulation of epsin is important for PAR1 internalization remains to be determined. A role for ubiquitination in mammalian GPCR lysosomal sorting and degradation has been demonstrated. Agonist-induced ubiquitination of βAR and CXCR4 is critical for lysosomal degradation but is not required for internalization (; ). The Ub moiety on CXCR4 is thought to interact with UBDs in some endocytic adaptor proteins, such as hepatocyte growth factor–regulated kinase substrate (Hrs), to be efficiently degraded in the lysosome (). Hrs interacts with Tsg101 (tumor suppressor gene product 101) and promotes the assembly of a multiprotein ESCRT (endosomal sorting complex required for transport) complex that binds and sorts ubiquitinated cargo into the involuting membrane of multivesicular endosomes in a highly coordinated manner (). In contrast to βAR and CXCR4, the ubiquitination of PAR1 is not required for agonist-induced lysosomal degradation because Ub-deficient PAR1 is degraded comparably with wild-type receptor in HeLa cells and Rat1 fibroblasts. A δ opioid receptor mutant lacking all intracytosolic lysines was also shown to undergo efficient agonist-induced degradation, indicating that lysosomal degradation of certain GPCRs occurs independently of ubiquitination (). We show that after activation, PAR1 is deubiquitinated, suggesting that deubiquitinated rather than ubiquitinated receptor transits through the endocytic sorting pathway to lysosomes for degradation (). Moreover, we recently reported that agonist-induced PAR1 lysosomal degradation is independent of Hrs and Tsg101 but requires sorting nexin 1 (), which is consistent with a Ub-independent PAR1 lysosomal sorting pathway. These data, in conjunction with the ability of the Ub-deficient PAR1 mutant to be efficiently degraded, strongly suggest that the conventional Ub-dependent ESCRT-mediated pathway is not required for agonist-induced PAR1 lysosomal sorting and degradation. However, we cannot exclude the possibility that the ubiquitination of PAR1 has a role in basal turnover of the receptor. The efficient trafficking of proteolytically activated PAR1 to lysosomes is essential for the termination of receptor signaling (); thus, further delineation of the lysosomal sorting pathway of activated PAR1 is important. Our studies reveal a novel function for ubiquitination in the negative regulation of PAR1 constitutive internalization and in specifying a distinct clathrin adaptor requirement for activated receptor internalization. PAR1 is uniquely activated by proteolytic cleavage that results in irreversible activation, unlike normal ligand-activated GPCRs. Thus, rapid desensitization and receptor trafficking tightly regulate PAR1 signaling. We have shown that PAR1 trafficking does not require arrestins and is essential for the disposal of irreversibly activated receptor and for replenishing the cell surface with uncleaved receptor after protease exposure. The novel regulation of PAR1 internalization by ubiquitination has a critical role in these distinct endocytic pathways. Interestingly, the regulation of PAR1 internalization by ubiquitination is not observed with all PARs because the ubiquitination of PAR2, a second protease-activated GPCR, functions in lysosomal degradation but not in receptor internalization (). Unlike PAR1, arrestins are required for PAR2 internalization (). However, whether other GPCRs that do not require arrestins for endocytic sorting are similarly regulated by ubiquitination remains to be determined. Our studies provide new insight into novel mechanisms by which ubiquitination functions in the endocytic sorting of GPCRs in mammalian cells. The challenge is to now identify the physiologically relevant Ub ligases and deubiquitinating enzymes that function in PAR1 trafficking. Monoclonal M1 and M2 anti-FLAG and polyclonal anti-FLAG antibodies and the anti–β-actin antibody were purchased from Sigma-Aldrich. Monoclonal anti-HA antibody conjugated to HRP was obtained from Roche, and polyclonal anti-HA was obtained from Covance. Anti-PAR1 rabbit polyclonal antibody was previously described (). The anti-AP50 (μ), anti–β2 adaptin, anti–early endosomal antigen-1 (EEA1), and anti-CHC monoclonal antibodies were purchased from BD Biosciences. Antidynamin monoclonal antibody was obtained from Santa Cruz Biotechnology, Inc. HRP-conjugated goat anti–mouse and goat anti–rabbit antibodies were purchased from Bio-Rad Laboratories. AlexaFluor488 and -594-conjugated anti–mouse and anti–rabbit antibodies were obtained from Invitrogen. A human PAR1 cDNA containing an N-terminal FLAG or HA epitope was used to generate mutants. Mutations were introduced by site-directed mutagenesis using the QuikChange Mutagenesis kit (Stratagene) and confirmed by dideoxy sequencing. PAR1 0K and K/R mutants were generated by replacing intracytosolic lysine (K) residues with arginine (R). A PAR1 0K mutant with Ub fused in frame to the C tail was generated as follows. A Pm1I site was introduced at the 3′ end of the PAR1 0K C tail and positioned such that the PmlI sequence, CACGTG, coincided with the native stop codon. A SacII site was introduced in the 3′ untranslated region 21 bp from the PmlI site. PCR amplification was then used to generate the Ub cDNA fragment with the 5′ Pm1I site and 3′ SacII sites. A 222-bp PmlI–SacII fragment encoding Ub containing a K to R mutation and glycine deletions was then ligated in frame using PAR1 0K 3′ end compatible sites. PAR1 and Ub coding regions were separated by a spacer sequence, HVV. Insertion of Ub in frame after the PAR1 0K C tail was confirmed by dideoxy sequencing. HeLa cells and Rat1 fibroblasts stably expressing PAR1 wild type and mutants were generated and maintained as previously described (, ). HeLa cells were transiently transfected with a total plasmid amount of 0.4 μg per 24 wells, 0.8 μg per 12 wells, and 2 μg per six wells using LipofectAMINE reagent (Invitrogen) according to the manufacturer's instructions and were assayed 48 h after transfection. HEK293 cells plated at ∼1 × 10 cells per 10-cm dish were transiently transfected with a total of 7.4 μg of plasmids (consisting of 5 μg of PAR1, 0.4 μg of tagged Ub, and 2 μg of dynamin K44A) using FuGene-6 reagent or LipofectAMINE as described previously () and according to the manufacturer's instructions. Cells were then split into 6-cm plates and assayed 48 h after transfection. HeLa cells were transiently transfected with 50 nM of ns or μ2-specific siRNAs using LipofectAMINE 2000 according to the manufacturer's instructions. The μ-siRNA targeting the mRNA sequence 5′-GTGGATGCCTTTCGGGTCA-3′ and the ns-siRNA 5′-CTACGTCCAGGAGCGCACC-3′ were previously described (). The CHC siRNA targeting the mRNA sequence 5′-GCAATGAGCTGTTTGAAGA-3′ was used at 50 nM and was previously described (). All siRNAs were synthesized by Dharmacon. HeLa cells stably expressing PAR1 wild type or 0K mutant were labeled with 1.0 μCi/ml -[H]inositol (American Radiolabeled Chemicals), and accumulated [H]IPs were measured as previously described (). HeLa cells stably expressing PAR1 wild type or mutants were processed, fixed, permeabilized, and immunostained with species-specific secondary antibodies conjugated to AlexaFluor488 or -594 and were mounted in FluorSave reagent (Calbiochem) and imaged by confocal microscopy as we previously described (). Images were acquired using a laser-scanning confocal imaging system (FluoView 300; Olympus) configured with a fluorescence microscope (IX70; Olympus) fitted with a planApo 60× NA 1.4 oil objective (Olympus). Confocal images (x-y section at 0.28 μm) were collected sequentially at 800 × 600 resolution with 2× optical zoom using FluoView software at room temperature. The final composite image was created using Photoshop 7.0 (Adobe). Constitutive and agonist-induced PAR1 internalization were assessed using our previously described assays for loss of surface receptor and receptor-bound antibody uptake (, ). PAR1 recycling was measured as we previously described (). HEK293 cells transiently transfected with FLAG-tagged PAR1 wild type or mutants, HA- or FLAG-tagged Ub, and dynamin K44A were incubated with or without agonists and lysed in buffer containing 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 2 mM EDTA, pH 8.0, 0.5% (wt/vol) sodium deoxycholate, 0.1% (vol/vol) NP-40, 0.1% (wt/vol) SDS, 100 μM sodium orthovanadate, 20 mM -ethylmaleimide, and protease inhibitor tablet (Roche). Equivalent amounts of protein lysates were then immunoprecipitated with M2 anti-FLAG or anti-HA antibody. Immunoprecipitates were resolved by SDS-PAGE, transferred to nitrocellulose membranes, and probed with an anti-HA or M2 anti-FLAG antibody conjugated to HRP. Blots were then stripped and probed with a polyclonal anti-FLAG or -HA antibody to detect PAR1. Immunoblots were developed with ECL (GE Healthcare) and imaged by autoradiography. PAR1 degradation was determined in HeLa cells and Rat1 fibroblast cells stably expressing the PAR1 wild type or mutants as previously described (; ). Data were analyzed using Prism 4.0 software (GraphPad), and statistical significance was determined using InStat 3.0 (GraphPad). Group comparisons were made using an unpaired test.

The venue for phototransduction, the process whereby a light stimulus is translated biochemically into an electrical signal, is the outer segment of rod and cone photoreceptor cells. Rods are highly differentiated cells with a cylindrical rod outer segment (ROS; ). The ROS contains all components necessary for phototransduction and presents a remarkably well-ordered system of membranes. A ROS is composed of stacks of up to 2,000 discs surrounded by a plasma membrane (). The discs are shaped like flattened sacks that are made up of two lamellar membranes circumscribed by a hairpin rim region. Phototransduction is initiated in the discs by the absorption of a photon by rhodopsin and culminates in the closure of cyclic guanosine monophosphate–dependent ion channels located in the plasma membrane (; ). Whereas the integrity of the highly ordered structure of the ROS is critical to the fidelity of this signaling process, little is known about how this organization is maintained. Changes in the structure of the ROS can lead to retinal dystrophies. Mouse models have provided many insights into understanding phototransduction and retinal dystrophies. Many mouse strains are available with naturally occurring mutations that lead to retinal dystrophies like and mice (for reviews see ; ). Models of human retinal dystrophies have been created by genetic engineering in the form of transgenics, knockout, and knock-in methodologies (). Moreover, mouse models of retinal dystrophies provide a method for testing various genetic and pharmacological therapies to combat diseases leading to blindness (). Detailed morphological information from the native ROS, particularly from mice, will be important to understand mechanisms underlying phototransduction and retinal dystrophies. EM studies have contributed substantially to our current understanding of the ROS structure. One of the earliest EM studies on the ROS was performed by Sjöstrand on samples from guinea pig. The most extensively studied ROS system comes from amphibian sources. EM studies have revealed conserved features of the ROS among vertebrates (; ), which are likely to share similar mechanisms for structure maintenance. EM studies have also revealed differences among vertebrate ROS structures. The diameter of discs, number of stacked discs, and length of the ROS is much greater in amphibians (e.g., frogs and mudpuppies) compared with mammals. The number and depth of incisures, which are invaginations at the periphery of discs, is also variable among species. Rodent and bovine discs display only a single incisure, whereas amphibian discs can have as many as 30 deeply penetrating incisures (; ; ). Dimensions of the interior space of the ROS have been determined from EM studies for many different species (). Reports from the frog are by far the most abundant. Measurements for murine photoreceptors are not present in the literature. Variations are observed among measurements obtained for the same species, which may reflect differences that can arise from preparatory demands inherent to EM methods. Accurate measurements for the dimension of the intricate membrane system within the ROS are necessary to provide a proper framework in which to carry out theoretical studies on phototransduction (; ; ). In this study, cryoelectron tomography (ET [cryo-ET]) was used to visualize the detailed morphology of the ROS from the retina of mice in three dimensions. Cryo-ET combines optimal structure preservation with the advantage of three-dimensional imaging (; ; ). Tomograms of vitrified samples of the ROS revealed the existence of spacer structures connecting adjacent discs at locations not observed previously in the ROS of other species. The distances between the membranes comprising the ROS were determined from micrographs to provide an accurate framework for the space within which phototransduction occurs. Highly enriched and structurally preserved ROS preparations were obtained from murine retina as described previously (). Samples of purified ROS were vitrified by rapid freezing in liquid ethane. Micrographs of vitrified samples displayed characteristic features of the ROS (). The organization of discs exhibited a high degree of order (). Other organelles such as the cilium that connects the ROS to the rod inner segment and mitochondria that reside in the rod inner segment also were detected in some micrographs (). Multiple single-axis tilt series were recorded of vitrified samples of ROS (Video 1, available at ) to compute three-dimensional reconstructions (representative tomograms of a ROS are shown in and S1). An isosurface representation of the ROS was created from the reconstructed tomogram to provide a three-dimensional view of the structure ( and S2). Distance measurements between different membrane structures in the ROS recorded from cryoelectron tomograms are illustrated in and reported in . Mean distance measurements were obtained from three datasets and were consistent among the different datasets. The volume of the ROS cytoplasm and a single disc was calculated based on measurements determined from cryoelectron tomograms (). The mean length and diameter of murine ROS have been estimated to be 23.8 μm and 1.32 μm, respectively, and each ROS contains a mean of 810 discs (). The interior volume of a ROS was calculated to be 32 × 10 ml. This interior space will be occupied by the cytoplasm and discs. The volumes of the intradisc space and disc membranes were calculated to be 5 × 10 ml and 22 × 10 ml, respectively. Each disc occupies 27 × 10 ml of space, and all discs together will occupy 22 × 10 ml of space in the ROS. The cytoplasm occupies the remaining 10 × 10 ml of space in the ROS. The cytoplasm in murine ROS represents 31% of the space inside a ROS. This value is lower than the estimated 50% occupancy of amphibian ROS by the cytoplasm (). The density of disc membranes was not uniform in tomograms (), exhibiting areas of high density (, dark yellow) and areas of low density (, light yellow). The high density regions represented 71% of the disc volume, whereas the low density regions represented the remaining 29% of the disc volume (). The difference in density likely arises from the nonuniform distribution of rhodopsin, which accounts for 90% of all proteins in discs (; ) in disc membranes. Each murine retina has been estimated to contain 6.4 × 10 rod photoreceptor cells () and 527 pmol of rhodopsin (). Thus, the total calculated volumes of disc membranes and the ROS cytoplasm in the retina will be 114 × 10 ml and 64 × 10 ml, respectively. Rhodopsin is then present at a concentration of 4.62 mM in disc membranes and at a concentration of 8.23 mM with respect to the ROS cytoplasm. The density of rhodopsin in the disc membrane is estimated to be 24,102 molecules/μm on average or up to ∼34,000 molecules/μm in high density patches. The isosurface representation of tomograms also revealed spacer structures that connect discs to the plasma membrane and to adjacent discs (). The spacers between adjacent discs were seen not only at the rim region but also at the lamellar regions of the discs. Such spacers were distributed throughout the discs at a mean density of 492 ± 134 molecules/μm ( = 18). The size of spacers was heterogeneous, and they had a mean estimated mass of 500 ± 100 kD. This mass was calculated by determining the mean volume of 125 spacers and assuming a protein density of 1.3 g/cm. The missing information in Fourier space as the result of a limited tilt range of specimens (i.e., missing wedge; ) was not taken into account in the calculation of the mass of spacers. Spacers were observed independent of the orientation of the missing wedge. The challenge in studying the ultrastructure of the ROS is the preservation of the native state of the membrane structure and of macromolecules. This challenge is reflected in the variations observed in distance measurements between membrane structures of the ROS obtained from conventional EM studies (). The reported measurements were variable even for the same species, which highlights the risk of artifacts associated with conventionally prepared samples that can compromise the reliability of quantifications. In transmission EM (TEM) studies, samples are often chemically fixed, dehydrated, embedded, sectioned, and heavy metal stained, which can introduce artifacts (). The osmolarity of buffers used at various stages of sample preparation as well as the dehydration required for plastic embedding can have considerable effects on distances between disc membranes (; ; ; ). Also, the choice of fixatives and stains can alter the structure and dimensions of membranes in the ROS (; ; ). Freeze fracture or freeze etch studies avoid some of the problems associated with TEM sample preparations. However, samples still undergo some processing, such as fracturing and metal shadowing, which can also result in artifacts and difficulties in obtaining accurate quantitative data (). Cryo-ET avoids most of the potential pitfalls associated with conventional EM preparatory steps because blotting and vitrification are the only processing steps that samples are subjected to. Thus, preparatory steps in cryo-ET are minimally invasive. Vitrification aids in the preservation of the native state of macromolecules because vitrified water is amorphous in character like liquid water and keeps samples hydrated (). Despite these advances over conventional EM methods, cryo-ET is not free from all potential sources of the native state disruption of samples. Cells and lipid vesicles studied by cryo-ET undergo varying degrees of compression that may arise from the capillary forces introduced during the blotting procedure (; , ; ; ). The compression of samples reduces the thickness of samples and can result in the collapse of the cell structure in some instances (). The murine ROS represents one of the largest structures to be studied by cryo-ET so far. ROS from mice have a diameter of 0.85–1.4 μm (, ), which places this structure at the size limit for study by cryo-ET without sectioning. The ROS appears to undergo some compression, as indicated in , where the left half of the ROS is obscured as a result of the thickness of the sample in that area. Compression of discs may lead to breakage at one edge in some instances (Fig. S3, available at ). The high degree of order exhibited by discs, the consistency of quantitative measurements, and the conservation of features highlighted in this study among all datasets analyzed suggest that the native state of the ROS in analyzed datasets has been preserved in large part. Measurements of distances between membranes of the ROS obtained in this study are similar to some of the reported values obtained from other species, including amphibians (). Distances between adjacent discs and between disc rims and the plasma membrane may be conserved across all species. The thickness of a single disc membrane measured from cryo-ET micrographs is the same as that determined from atomic force microscopy (AFM) studies on murine ROS discs (; ) and bovine ROS discs (Sapra et al., 2006). Similar to cryo-ET, AFM imaging in buffer solution allows for studies on samples in a near-native state. The thickness of a single disc membrane was 8 nm from cryo-ET images. This thickness likely derives from the length of rhodopsin molecules rather than lipids because a lipid bilayer devoid of proteins is typically ∼4 nm in thickness (; Sapra et al., 2006). The agreement of cryo-ET data with AFM data demonstrates the reliability of measurements made from cryo-ET micrographs. Tomographic reconstructions provide three-dimensional information about the ROS structure ( and ). The interplay between electron dose and the level of noise that accompanies the signal sets the resolution limit in cryo-ET (; ). Radiation damage of the specimen limits the electron dose, which leads to low contrast and a low signal to noise ratio in tomograms. The tomograms analyzed contain information up to the resolution limit of ∼3–5 nm, which is determined by the instrumentation and reconstruction procedure. However, the resolution in ROS tomograms will be lower than the resolution limit because of the noise level. The resolution of tomograms was not sufficient to detect individual rhodopsin molecules in the disc membrane. The nonuniform density of the disc membranes suggests that rhodopsin is heterogeneously distributed in the membrane (). Regions that display no density are also present in the disc membrane. These areas are indistinguishable from the density of the surrounding buffer solution and may represent holes in the membrane. The nature of these density-free regions requires further investigation. Rhodopsin, like other G protein–coupled receptors, likely form dimers or higher order oligomers (; ; ). AFM measurements and power spectra recorded from TEM micrographs of negatively stained isolated disc membranes suggest that rhodopsin oligomers can exist in paracrystalline arrays (; ). In contrast, power spectra measured from discs in tomograms of the ROS gave no indication of a paracrystalline arrangement of rhodopsin. The propensity of rhodopsin to form paracrystalline arrays under native conditions and the proportion of discs within a ROS that display this arrangement requires further investigation. The rim region of discs can maintain contact with the plasma membrane and proper spacing along the length of the ROS even in the absence of the lamellar region of discs destroyed either by osmotic pressure or treatment with osmium tetroxide and Tris (; ). Sporadic reports have been made of spacer structures connecting adjacent discs and connecting the rim region of discs to the plasma membrane in EM micrographs of ROS from species including the frog, toad, eel, rabbit, cow, and rat (; ; ; ; ; ; ). Spacers reported in these studies were localized mainly to the rim region and incisures of discs and may play a role in anchoring the rim region of discs in their position. Such spacers are sensitive to the fixation procedures, and their presence seems to depend on the preparatory procedures used (; ). Spacers appear to be flexible and can be stretched up to 30 nm before breaking (). Spacers connecting the rim region of discs to the plasma membrane appear to have different characteristics compared with those connecting adjacent discs (; ; ). Lamellar regions of discs are flexible and prone to deformations. Reports of spacer structures in this region are largely absent. In the few instances in which such structures were observed, they occurred less frequently than those found at the rim region (; ). In contrast, our tomograms of the ROS display spacer structures linking adjacent discs to be distributed throughout the disc rather than concentrated at the periphery of the disc (Fig. S4, available at ). Part of the difference between our results and those reported by others may relate to variability in the number and depth of incisures observed among different species. Most studies have been performed in frogs, in which the discs have many incisures that constitute a large surface area. On the other hand, discs from mice have only a single incisure (). If spacers were localized exclusively in the rim region and in the incisures of discs in mice, the large lamellar area would be without the proper support to maintain the precise stacking of the discs. Spacers may be distributed throughout the disc surface from murine ROS to facilitate the absence of a large number of deeply penetrating incisures. Alternatively, spacers connecting the lamellar regions of adjacent discs may be more susceptible to disruption by conventional EM preparatory steps and, therefore, have gone undetected, for the most part, until now because of the preservation afforded by vitrification. The molecular identity of proteins comprising spacer structures is unknown. The spacers are required to span a distance of 14 nm between adjacent discs and 17 nm between the rim region of discs and the plasma membrane. A fairly large macromolecular complex would be needed to provide the bridge between adjacent discs in the lamellar regions with a mass in the range of ∼500 kD. The spacers are likely composed of a complex of proteins, and spacers in different regions may have different molecular compositions. The protein or protein complexes that comprise the spacers will be present at a level that is ∼49 times less than that of rhodopsin. No structural proteins have been implicated yet in the lamellar region of discs. All characterized structural proteins to date are localized to the disc periphery or incisure regions. Proteins involved in spacer complexes may include peripherin-2 and ROM-1, which are members of the tetraspanin family of proteins. Both proteins have been implicated as structural components required for the maintenance of ROS morphology and disc stacking. The importance of these proteins in maintaining the regular structure of the ROS, peripherin in particular, is revealed in mice homozygous for the disrupted peripherin gene. Homozygous mice fail to develop ROS, and heterozygous mice produce highly disorganized disc structures (). Peripherin associates through intramolecular and intermolecular disulfide bonds with itself and with ROM-1 to form a mixture of homo- and heterotetramers (; ). Rod photoreceptors also contain a set of glutamic acid–rich proteins (GARPs) that may be another component of spacer complexes. EM images show that GARPs appear to be localized in the rim region and incisures of discs in close proximity to guanylate cyclase and ABCR. GARPs have been proposed to organize a dynamic protein complex between the cyclic guanosine monophosphate–gated channels in the plasma membrane and peripherin in the rim region of discs, thereby playing a role in maintaining the distance between the rim region of the disc and the plasma membrane (; ). Soluble GARPs have been proposed to form a complex with the peripherin-2–ROM-1 complex to bridge the rim region of adjacent discs (). GARPs have been shown to be unstructured and flexible, a feature that would be beneficial for spacer structures to accommodate minor changes in the structure of the discs. C57BL/6 mice were obtained from The Jackson Laboratory. All procedures were performed under dim red light. Mice were maintained in darkness overnight before being killed. Retinal tissue from 15 mice (∼8 wk old) were placed in 300 μl of 8% (vol/vol) OptiPrep (Nycomed) in Ringer's buffer (10 mM Hepes, 130 mM NaCl, 3.6 mM KCl, 2.4 mM MgCl, 1.2 mM CaCl, and 0.02 mM EDTA, pH 7.4). The solution was vortexed at maximum speed for 1 min, and the sample was then centrifuged at 238 for 1 min. The supernatant was removed and stored on ice. The pellet was resuspended in 300 μl of solution containing 8% OptiPrep in Ringer's buffer, and the vortexing and centrifugation procedure was repeated as described above. This sequence was repeated five times. Supernatant from each spin was pooled and layered on a 10–30% (vol/vol) continuous gradient of OptiPrep in 12 ml of Ringer's buffer. The gradient was centrifuged for 50 min at 26,500 and 4°C with no brakes. Intact ROS migrates as a second band about two thirds of the way from the top. Intact ROS was collected and diluted threefold in Ringer's buffer. This suspension was then centrifuged for 3 min at 627 . The supernatant was collected and centrifuged for 30 min at 26,500 . The resulting pellet contained intact ROS. The pellet was resuspended in 40 μl of Ringer's buffer, and the resulting suspension was used to prepare the EM grids. An aliquot of ROS was applied to grids covered with holey carbon film pretreated with a mixture of 5–25 nm of gold bead markers. Excess liquid was blotted with filter paper and immediately plunged into liquid ethane (). Grids were transferred into liquid nitrogen and mounted on single-tilt cryoholders. ET was performed using a CM300 FEG electron microscope (Philips) and a Tecnai Polara G electron microscope (FEI) working at an accelerating voltage of 300 kV and equipped with a postcolumn energy filter (GIF 2002; Gatan, Inc.) operated in the zero-loss mode. Single-axis tilt series were recorded ranging typically from −65 to 65° with 1–2° increments. Electron micrographs, each covering 2,048 × 2,048 pixels, were recorded at magnifications between 20,000 and 40,000× with a pixel size at the specimen level of 0.5–1.1 nm. The defocus was set to between −4 and −6 μm unless otherwise noted, resulting in a first zero of the phase-contrast transfer function between 3 and 5 nm. All tomographic data acquisition steps were performed by fully automated procedures under strict low-dose conditions. Accumulative doses for recording the tilt series were ∼40,000–100,000 e/nm. All image processing steps were performed using the TOM software toolbox (). Images of individual tilt series were aligned with each other using colloidal gold beads as marker points added before cryofixation. Markers were picked interactively on a graphics display, and their coordinates were used to compute the alignment parameters. Alignment and three-dimensional reconstruction was performed by weighted backprojection. A total of 10 tomographic datasets were analyzed. Tomograms shown in and S1 are representative of the datasets analyzed. Visualization and isosurface representations of the tomograms were generated with Amira software (Mercury Computer Systems, Inc.). Individual components found in the isosurface images were segmented manually. The isosurface threshold was determined for each component separately. A mean isosurface value was computed for each component by probing at 20 or more positions. This mean isosurface value was then used as the threshold value. Fig. S1 is an electron tomogram of vitrified ROS obtained from a different dataset than that represented in Fig. S2 is an isosurface representation of a selected area from the electron tomogram of a ROS shown in Fig. S1. Fig. S3 is a cryoelectron micrograph showing compressed ROS discs. Fig. S4 is a schematic showing the distribution of spacers in discs. Video 1 shows a single-axis tilt series for a ROS. Video 2 shows an electron tomogram and isosurface representation of a ROS. Online supplemental material is available at .

The urokinase plasminogen activator (uPA) and its receptor (uPAR) play important roles in physiological processes such as wound healing, inflammation, and stem cell mobilization, as well as in severe pathological conditions such as HIV-1 infection, tumor invasion, and metastasis (; ; ; ; ). Besides uPAR's well-established role in the regulation of pericellular proteolysis, it also modulates cell adhesion, migration, and proliferation through interactions with proteins present in the extracellular matrix, including vitronectin (Vn) (; ; ). The absence of transmembrane and intracellular domains renders uPAR signaling incompetent, and it is generally believed that signal transduction originating from this receptor must involve lateral interactions with transmembrane proteins. In accordance, uPAR-mediated processes have been proposed to require its interaction with membrane proteins including members of the integrin family (), chemokine receptors (), and receptor tyrosine kinases (). The important role of uPAR in tumor cell adhesion, migration, invasion, and proliferation makes this receptor an attractive drug target in cancer treatment; however, this is complicated by the extent of the published uPAR “interactome”. With such a goal in mind, the most important question becomes which of the many molecular interactions are really essential to mediate uPAR function. Recently, the crystal structures of uPAR in complex with a peptide antagonist () and with the N-terminal fragment of uPA (; ) were presented, providing the first rational basis toward understanding how uPAR may organize its multiple molecular interactions. Attempts to identify the regions in uPAR involved in the specific interaction with Vn and integrins have been published (; ; ; ); however, the results from these experiments are largely incongruent and remain to be confirmed in independent studies. In this study we have applied a genetic approach, based on exhaustive mutagenesis and complementation experiments, to determine the essential direct molecular interactions required for uPAR to induce changes in cell morphology and migration. This was achieved in a comprehensive and unbiased way through a complete functional alanine scan of human uPAR in human embryo kidney (HEK) 293 cells. Of the 255 alanine substitutions analyzed, 34 were found to completely, or partially, impair uPAR-induced changes in cell morphology. The molecular defect of all mutants was subsequently shown to be an impaired binding to the somatomedin-B (SMB) domain of Vn. Although the RGD motif in Vn was dispensable for uPAR-dependent cell binding to Vn, it was implicated in subsequent signal transduction and changes in cell morphology, underscoring the importance of integrin receptors downstream of uPAR in these processes. Although integrins are clearly involved in the cellular processes initiated by uPAR expression, we found strong evidence against any functional relevance of a direct molecular interaction between these molecules. First, all the mutants identified in the complete alanine scan have the same molecular defect (impaired Vn binding), suggesting that this is also the only required interaction. Second, alanine substitutions of the published integrin interaction sites in uPAR had no effect on receptor function. Finally, a recombinant GPI-anchored plasminogen activator inhibitor-1 (PAI-1) molecule (PAI-1/GPI), which also binds Vn, mimicked the cellular effects of uPAR expression. We conclude that a direct Vn interaction is both necessary and sufficient to initiate uPAR-induced changes in cell morphology, migration, and signaling independently of direct lateral protein–protein interactions. The importance of the uPAR–Vn interaction was not cell type–specific, as very similar data were obtained in CHO cells. It has been previously reported that expression of uPAR modulates the adhesion and motility of 293 cells through interactions with Vn, integrins, and G protein–coupled receptors (, , , ; ; ; ;). In accordance, we find that expression of human uPAR in HEK293 Flp-In T-REx cells induces changes in cell morphology, migration, and signaling (). The morphological changes include a general flattening of the cells, reduced cell–cell contact, disappearance of membrane ruffles, and formation of extensive lamellipodia (, left panels), as well as a complete reorganization of the matrix-proximal F-actin cytoskeleton (, right panels). The changes in cell morphology reflect the cell motility induced by uPAR expression in these cells ( and Video 1, available at ). Migratory () and proliferative () signaling downstream of uPAR involves activation of the Ras/MAPK signaling pathway, and in accordance we found that uPAR-expressing cells had approximately threefold increased ERK1/2 activation as compared with mock-transfected cells (). As previously described (, ; ), expression of uPAR in 293 cells resulted in a strong increase in cell adhesion to Vn, whereas adhesion to other ECM proteins including fibronectin, type-1 collagen, and laminin was unaffected (). The binding sites for uPAR and integrins on Vn are distinct, with uPAR recognizing the SMB domain (), and integrins the adjacent RGD motif. To address the relative importance of uPAR versus integrin binding to Vn, we performed adhesion assays using a recombinant N-terminal fragment of Vn (Vn(1–66)) that includes both the SMB domain and the RGD motif. In addition, we monitored adhesion to two variants of this fragment where either most of the SMB domain had been removed (Vn(40–66), representing essentially RGD alone), or the RGD sequence had been mutated into RAD (Vn(1–66)) (). Adhesion of 293/uPAR cells was supported as long as the SMB domain was present and was not affected by mutating the RGD motif. Mock-transfected 293 cells adhered poorly to all of these substrates and not at all to Vn(1–66) (). Profiling of Vn receptor expression by the 293 cells and antibody inhibition experiments (Fig. S1, A and B; available at ) indicated that the weak RGD-dependent adhesion of mock-transfected 293 cells to Vn was mediated by the αβ integrin. Collectively, the adhesion data demonstrate that uPAR promotes cell adhesion to Vn through a direct interaction with the SMB domain, and not by inducing integrin binding to the RGD motif. Although the RGD motif in Vn is dispensable for adhesion of 293/uPAR cells, it is at least partially required for the subsequent changes in cell morphology and signal transduction (). Although 293/uPAR cells seeded on Vn(1–66) undergo marked changes in morphology, including cell flattening and extensive lamellipodia formation, these changes are absent when seeded on Vn(1–66), where only a round adhesion patch can be observed under the cell body. Under the same conditions, mock-transfected cells retain a rounded cell body, fail to form lamellipodia, and only extend thin membrane protrusions with limited matrix contact as evidenced also by the weak adhesive strength of these cells (). Mock-transfected cells remained completely round on Vn(1–66) in accordance with the adhesion data. When plated on Fn, no differences in cell adhesion and morphology were observed comparing 293/mock and 293/uPAR cells (unpublished data). The level of ERK1/2 activation was higher () when 293/uPAR cells were seeded on Vn(1–66) as compared with Vn(1–66), suggesting that also uPAR-induced downstream signaling is at least partially integrin dependent. In summary, these data show that the ability of uPAR to induce changes in 293 cell morphology is a two-step process in which uPAR, through the direct interaction with the SMB domain of Vn, triggers cell attachment to the matrix. Subsequently, this initial adhesion is followed by engagement of integrins, possibly αβ, with matrix Vn, triggering changes in cell morphology, migration, and signal transduction. In addition to the direct interaction with matrix Vn, extensive data suggests that the ability of uPAR to modulate cell adhesion, migration, and proliferation requires a complex network of lateral interactions with a variety of membrane proteins including integrins, receptor tyrosine kinases, and chemokine receptors. To address the nature of these lateral uPAR interactions in a comprehensive and unbiased fashion, we next conducted a complete functional alanine scan of uPAR in 293 cells. Receptor-induced changes in 293 Flp-In T-REx cells were chosen for this purpose for two reasons: as shown in , uPAR expression in these cells causes very evident changes in cell morphology which can be scored easily by phase-contrast microscopy; and in addition, the Flp-In system generates pools of isogenic transfectants carrying a single copy of the expression cassette, thus eliminating potential artifacts caused by clonal differences or heterogeneous expression levels. We generated 255 single amino acid substitution mutants of uPAR by site-directed mutagenesis (all residues of mature uPAR excluding 28 cysteines) and expressed these in 293 Flp-In T-REx cells. As expected, the cell surface expression of all mutant receptors was similar and comparable to that of wild-type uPAR as evaluated by FACS analysis and immunoblotting (; Fig. S1, C and D; and Table S1 for a complete listing). Of the 255 receptor variants analyzed, 34 were found to fully or partially impair the ability of uPAR to induce changes in cell morphology (listed in ). None of these residues coincided with previously identified interaction sites for Vn or integrins (; ; ; ), but several are known to be involved in uPA binding (). As the 293 cells do not express uPA (), the molecular defect of these mutants could not be explained by available evidence. Five of the mutants were found to suffer from severe folding problems as evidenced by the presence of high molecular weight covalent aggregates in immunoblotting experiments, and were not analyzed further ( and Fig. S1 D). Because the initial step in uPAR-induced signaling and cell morphology changes is triggered by the direct binding to the SMB domain of Vn (), we next analyzed the adhesive properties of cells expressing the remaining 29 mutant receptors with impaired ability to change cell morphology (). Remarkably, all of the mutant receptors displayed an impaired ability to promote cell binding to Vn(1–66) (). There was a strong correlation between the ability to induce changes in cell morphology and the reduction in uPAR binding to the SMB domain of Vn (), demonstrating a direct link between these parameters. Receptor mutants that did not impair the ability of uPAR to induce changes in cell morphology (S56A) as well as receptors containing mutations in published integrin interaction sites (E134A/E135A, S245A, H249A, and D262A) all displayed normal adhesion to the same substrate. All cell lines adhered equally well to Fn (Fig. S2 A, available at ). Ligand binding has been demonstrated to enhance uPAR-dependent cell adhesion to Vn, and we therefore also performed the adhesion experiments in the presence of exogenously added pro-uPA (). Under these conditions the adhesion of most of the mutants was restored to that of wild-type uPAR, suggesting that ligand binding may compensate for the molecular defect of most of these mutants. The rescued Vn adhesion was also associated with restored uPAR cell morphology when cells were cultured in the presence of pro-uPA (Fig. S2 B). For some mutants (in particular W32A and R91A) cell adhesion and uPAR morphology was only partially recovered by the addition of pro-uPA, suggesting that these may be key residues for the uPAR–Vn interaction. The location of the mutated residues on the published crystal structure of uPAR combined with the relatively small size of the SMB domain suggest that only part of the identified residues actually engage SMB directly. To identify this subset we next generated and expressed soluble variants of the mutant receptors and assessed their binding to immobilized Vn (). As we were seeking residues in the direct molecular interaction interface, the analysis was restricted to the 19 mutant receptors where the substituted amino acid exhibits a surface exposure area on the published crystal structure above 15Å. In these binding assays, five mutants (W32A, R58A, I63A, R91A, and Y92A) were found to display greater than fivefold reduction in binding to immobilized Vn () while still having a normal binding to immobilized pro-uPA (unpublished data), suggesting that these residues are involved in the direct interaction with Vn. The impaired Vn binding of these mutants was observed also in the presence of excess pro-uPA, suggesting () that the defect of these mutants is not related to impaired uPAR dimerization (unpublished data). On a molecular surface representation of the uPAR structure (), W32 and R91 define a composite epitope with residues located in both domain 1 and 2 of uPAR. The residues affected by the R58A, I63A, and Y92A substitutions are all located close to these residues, suggesting that they may also engage Vn directly. The Vn-binding epitope is located distal to the membrane anchorage position (indicated in cyan) and lies on the “top-back” of the molecule with respect to the central uPA-binding cavity. Although we currently do not know the precise molecular explanation for the failure of the remaining mutant receptors to bind to Vn in the absence of pro-uPA, it should be stressed that from a functional point of view their defect is identical to that of W32A and R91A, i.e., impaired Vn binding. The fact that all the mutants with impaired ability to induce changes in cell morphology share the same molecular defect in RGD-independent Vn binding suggests that this may also be the only important uPAR interaction required to trigger changes in cell morphology, migration, and signaling. This is particularly intriguing because the ability of uPAR to modulate cell adhesion and migration has been shown previously to require direct uPAR–integrin interactions (, ; ; ; ). Individual (not depicted) or combined alanine substitutions (E134A/E135A/S245A/H249A/D262A, termed uPAR/Int) in the published integrin binding sites in uPAR had no effect on the ability of uPAR to induce changes in cell morphology and F-actin cytoskeleton (), Vn-adhesion (), signaling (), and cell migration (Fig. S3 A, available at ). On the other hand, a single alanine substitution in the Vn-binding epitope (W32A) completely abolished all these cellular effects of uPAR expression. Although these data argue strongly against any functional relevance of direct integrin interaction(s) for the biological activity of uPAR in our cell system, the single amino acid substitution strategy does not allow us to exclude the existence of additional and functionally redundant binding sites for integrins in uPAR. To directly address the possibility that no lateral interactions are required for the observed biological effects of uPAR expression, we conducted a genetic complementation experiment. PAI-1 shares no sequence or structural homology with uPAR; however, both molecules bind the SMB domain of Vn (). The Vn-binding site in PAI-1 is located distal from the C terminus to which we attached the GPI-anchoring signal of uPAR (Fig. S3 B). The resulting membrane-tethered PAI-1 (termed PAI-1/GPI) was efficiently expressed on the cell surface (Fig. S3 C) and induced strong cell binding to Vn(1–66) (). Remarkably, the expression of PAI-1/GPI in 293 cells induced changes in cell morphology, adhesion, and ERK1/2 activation comparable to those induced by uPAR (). The ability of PAI-1/GPI to induce these changes in cell morphology strictly required its interaction with Vn, as the introduction of a triple alanine substitution that efficiently disrupts its interaction with Vn (R103A/M112A/Q125A, termed PAI-1/GPI/Vn) () failed to induce any of these changes (). The central importance of the Vn interaction in the biology of uPAR is not cell type specific. When seeded at low density, CHO cells form colonies with an epithelial-like morphology characterized by tight cell–cell contact and a highly defined colony contour (, left panels). In sharp contrast, CHO cells expressing high levels of uPAR form sparse colonies with no cell–cell contact. The expression of uPAR is also associated with marked changes in the organization of matrix-proximal F-actin cytoskeleton (, right panels). In analogy to the 293 cell line, uPAR expression was associated with a strong and specific increase in RGD-independent cell adhesion to Vn () and caused increased ERK1/2 activation (). Despite the obvious differences between the morphological features of CHO and 293 cells, the direct Vn interaction was found to be required for the cellular effects of uPAR also in this cell line (). Alanine substitution in the Vn binding site (W32A) abolished the ability of uPAR to induce scattering of cells, changes in F-actin cytoskeleton, and ERK1/2 activation. As observed with the 293 cells, the exogenous addition of pro-uPA rescued the uPAR cell morphology in a mutant-dependent manner (Fig. S4, available at ). Furthermore, single alanine substitutions (not depicted) or combinations of substitutions in the published integrin interaction sites had no effect on uPAR function. Finally, the PAI-1/GPI receptor replicated uPAR-induced changes in a Vn-dependent manner also in this cell line. All the mutants identified in this study fail completely or partially to induce cell adhesion to Vn in the absence of pro-uPA (), yet their adhesive properties are in most cases restored to normal levels in the presence of exogenous pro-uPA (). As most of these mutants bind pro-uPA normally (unpublished data), they provide an important tool to dissect the importance of the direct Vn interaction in signal transduction induced by uPA binding as well as for the analysis of the cellular processes occurring subsequent to Vn engagement of the receptor. In the absence of exogenous pro-uPA expression of W32A and T54A mutant receptors both fail to induce cell binding to Vn () and both display a low level of ERK1/2 activation similar to that of mock-transfected cells (). In the presence of the pro-uPA, cells expressing the T54A receptor, but not the W32A receptor, display a fully restored Vn binding () and ERK1/2 activation state (). For comparison the Vn binding and ERK1/2 activation is constitutively high in cells expressing wild-type uPAR and only moderately affected by the addition of uPA. As the wild-type, W32A and T54A receptors all bind pro-uPA equally well these data document that uPA signaling to ERK1/2 occurs through the induction of uPAR binding to Vn. The recovery of Vn binding and ERK1/2 activation induced by pro-uPA on cells expressing uPAR/T54A is paralleled by rapid changes in cell morphology and focal contact turnover as illustrated by the translocation of paxillin-GFP into newly formed lamellipodia ( and Video 2, available at ) and by the scattering of preformed CHO colonies ( and Video 3), which resembles a bona fide epithelial–mesenchymal transition. None of these uPA-induced changes were observed in cells expressing uPAR/W32A, underscoring the fundamental requirement for a direct uPAR–Vn interaction in these cellular processes. Using a systematic and unbiased approach, we demonstrate that a direct interaction between cell surface uPAR and the extracellular protein Vn is required for the ability of uPAR to modulate changes in cell morphology, migration, and signal transduction in two different cell lines. This conclusion is based on two central observations. All of the alanine substitutions that affect the ability of uPAR to induce changes in cell morphology and migration directly affect uPAR-mediated cell binding to Vn independently of integrin binding to the RGD motif of Vn. Furthermore, a GPI-anchored PAI-1 molecule, which shares only the binding site in the SMB domain of Vn with uPAR, induces cellular changes that are virtually identical to those induced by uPAR. It has been extensively documented that the ability of uPAR to modulate cell adhesion, migration, and proliferation occurs through the regulation of integrin signaling (; ; ; ). Our data are fully in accordance with this view as we observe that the RGD motif in Vn is involved in uPAR-mediated downstream signal transduction and changes in cell morphology. Nevertheless, our data argue firmly against a functionally relevant direct interaction between uPAR and integrins. Alanine substitutions in the previously published integrin interaction sites, alone or in combination, have no effect on uPAR function in both cell lines analyzed. We found no mutants with normal RGD-independent Vn binding and impaired ability to induce changes in cell morphology. Finally, an engineered PAI-1/GPI molecule, which recapitulated the membrane–ECM interaction induced by uPAR, was sufficient to induce similar changes. Our data do not rule out direct uPAR–integrin interactions; however, they do allow us to conclude that if these interactions occur they are of little or no functional importance with respect to uPAR-induced changes in cell morphology, migration, and signaling, at least in 293 and CHO cells. Based on binding assays using purified proteins and the presence of pro-uPA, we were able to identify five residues (W32, R58, I63, R91, and Y92) that, when mutated into alanine, impair the ability of uPAR to interact with Vn. Noticeably, when these five mutants were tested for their ability to mediate cell attachment to Vn in the presence of uPA, only two of them (W32 and R91) were still significantly impaired. These two residues are therefore likely to represent the most important direct interaction site for Vn in uPAR. That we have identified the bona fide Vn-binding epitope in uPAR is supported by several structural, functional, and evolutionary observations: the binding epitope is located distal to the GPI anchor, as would be expected for a membrane receptor that interacts with a component of the ECM. The location of the binding site is different from the location of the uPA-binding site, allowing for the simultaneous binding of these two molecules. The composite epitope contains residues both in domain 1 (W32, R58, and I63) and domain 2 (R91 and Y92), explaining why proteolytic cleavage between these two domains abolishes binding (; ). All the residues in the identified epitope are exposed on the crystal structure of uPAR. Finally, all the residues are highly conserved between different species. How do we explain the mutants with impaired cell binding to Vn but normal binding when purified proteins are used? In contrast to cell binding assays, the binding of soluble uPAR to immobilized Vn cannot be measured in the absence of uPA (). Several of the mutants (R53, L55, Y57, L113, E120, L123, R145, and G146) are located in the area of contact between domain 1 and 2 of uPAR, suggesting that the correct “docking” of these two domains is critical for Vn binding. Binding of uPA, which engages extensive areas of both domain 1 and 2 of uPAR (; ), may compensate for this defect by enforcing the correct Vn binding competent alignment of these two domains. In addition, the N52 and T54 mutants suggest that in the absence of uPA, glycosylation of uPAR at N52 is required for Vn binding. Particularly intriguing is the finding that several of the mutant residues (L31, R53, L55, Y57, T64, L66, and E68) have their side chains exposed inside the uPA-binding cavity of uPAR. The fact that these residues were identified in a screen where uPA is absent suggests that something else, which is required for the ability of uPAR to bind Vn, is engaging this cavity in the absence of uPA and that this engagement is required for uPAR binding to Vn. The identity of this interaction has yet to be established, but based on the finding that uPAR binding to Vn involves receptor dimerization (; ), it is tempting to speculate that this interaction partner may be uPAR itself. Interestingly, part of the Vn binding site identified in this study (R91 and Y92) maps to the previously identified chemotactic epitope SRS located in the linker region connecting domain 1 and 2 of uPAR (). However, it seems unlikely that this chemotactic activity is important in our cell systems, as most of the critical residues identified in this study do not map to this region, yet they do share the same molecular defect. Furthermore, PAI-1 does not contain an SRSRY sequence, yet does it mimic uPAR function. Interestingly, peptides covering this region (uPAR aa 84–95) have been shown to inhibit uPAR-dependent cell adhesion to Vn (). Although PAI-1/GPI closely mimic uPAR function with respect to the changes in cell adhesion, morphology, and signal transduction presented in this work, this molecule should be used as a uPAR mimic with caution, as it has cellular effects which are not seen with uPAR. These effects include a pronounced tendency of the CHO cells to “lose pieces”, possibly because of an impaired lamellipodia retraction, as well as an impaired cytokinesis in the 293 cells (unpublished data). These effects are likely due to the fact that PAI-1 binds with much higher affinity to Vn than uPAR (). In line with several other studies (; ; ; ), we observe that overexpression of uPAR leads to increased ERK1/2 activation in both cell lines analyzed. A direct uPAR–Vn interaction is required for this activation, as Vn binding–deficient uPAR mutants display levels of active ERK1/2 comparable to those of mock-transfected cells. Interestingly, uPA binding to uPAR also leads to ERK1/2 activation in different experimental systems (; ), suggesting that both overexpression of the receptor and ligand binding induces the same signal transduction pathway(s), possibly through a common molecular mechanism. In line with other studies we have used systems with high-level expression of uPAR (∼3 × 10 binding sites per cell in both cell lines as determined by binding assays using Eu-labeled pro-uPA; unpublished data), and under these conditions cell adhesion to Vn is independent of pro-uPA binding to uPAR (, ; ). At physiological expression levels, uPAR-dependent cell adhesion to Vn requires uPA binding (; ; ). Consequently, it appears likely that uPA binding may actually induce “Vn signaling” by stimulating uPAR binding to matrix Vn. In support of this possibility we show that there is a strict correlation between the ability of pro-uPA to promote Vn binding and to induce ERK1/2-activation and changes in cell morphology (, , Fig. S2 B, and Fig. S4). What is the molecular mechanism of uPAR-induced “Vn signaling”? The TIR-FM images and adhesion assays clearly demonstrate that expression of uPAR dramatically increases the contact between the cell and the ECM. Although uPAR does so passively by a physical interaction with matrix Vn, it will invariably bring all matrix receptors present in the plasma membrane in closer contact with their extracellular ligands. In the 293 cell line the uPAR-induced increase in cell/matrix contact triggers subsequent changes in cell morphology and signaling that are at least partially mediated by an RGD-dependent integrin, possibly αβ. Although it is possible that uPAR and integrins bind contemporarily to the same Vn molecule, this does not seem to be required because PAI-1, which is known to block integrin binding to Vn (), induces similar changes when tethered to the cell membrane. Because the Vn-binding form of uPAR is associated with membrane structures known as lipid rafts (), the engagement of Vn by uPAR will bring these membrane domains in contact with the matrix. Lipid rafts are enriched in signal transduction molecules () and the uPAR–Vn interaction will thus promote the tight encounter between the ECM, its different cell surface receptors, and the signal transduction machinery. As different cells have different repertoires of membrane receptors and different matrices have different repertoires of ligands, our data suggest that the single interaction between uPAR and Vn may be responsible for many of the proteolysis-independent biological effects initiated by uPAR. It is important to underscore that the conclusions of this work cannot be directly extrapolated to uPAR functions different from its ability to induce changes in cell morphology, basal cell migration, and ERK1/2 activation. For example, we have not assayed the ability of the different uPAR mutants to modulate directional cell migration and cell proliferation. Consequently, our data does not exclude nor confirm the requirement for additional direct uPAR interactions in these processes. Nevertheless, the identification of the Vn binding site presented here provides the first tool to directly address the importance of the Vn interaction in any experimental system where uPAR is expressed by transfection and/or infection. The single alanine substitution W32A efficiently disrupts the interaction with Vn while maintaining normal pro-uPA binding. CHO Flp-In and HEK293 Flp-In T-REx cells, expression vectors pcDNA5/FRT/TO and pOG44, zeocin, blasticidin S HCl, and Ham's F12 medium were from Invitrogen. Dulbecco's modified eagle medium (DME) and αMEM were from BioWhittaker. PBS, trypsin, glutamine, penicillin, and streptomycin were obtained from EuroClone, and fetal bovine serum (FBS) was from HyClone. Non-tissue culture plates were from Falcon Becton Dickinson. Tetracycline, poly--lysine, laminin (from Engelbreth-Holm-Swarm murine sarcoma), phalloidin-FITC, and CHO protein-free culture medium were from Sigma-Aldrich. FuGENE 6, fibronectin, and Hygromycin B were from Roche. Type-1 collagen from rat tail was from BD Biosciences. Urea-purified Vn was obtained from Promega. Pro-uPA was provided by Dr. Jack Henkin (Abbott Laboratories, Abbott Park, IL). Antibodies against total and phosphorylated ERK1/2 were from Cell Signaling Technology. Blocking antibodies against αβ (LM609) and αβ (P1F6) integrins were from Immunological Sciences. Monoclonal antibody against β integrin (mAb 13) was from BD Biosciences. Monoclonal antibodies against human uPAR and the PAI-1 polyclonal antibody were provided by Dr. Gunilla Høyer-Hansen (Finsen Laboratory, Copenhagen, Denmark) and Dr. Peter Andreasen (Århus University, Århus, Denmark), respectively. Goat anti–mouse and goat anti–rabbit Ig F(ab′)-FITC were from Jackson ImmunoResearch Laboratories. The expression vector encoding full-length uPAR was generated by amplifying (oligos uPARu and uPARd) the human uPAR cDNA () and cloned in the pcDNA5/FRT/TO vector using Kpn1/Not1. The expression vector encoding a GPI-anchored PAI-1 molecule was generated by uniting the complete PAI-1 coding region (aa −22 to 380), a short linker sequence (AlaGlyAlaGlyAlaGlyLys), and the GPI-anchoring signal sequence of uPAR (aa 276–313). The PAI-1 coding region and the linker region were generated by amplifying a HT1080 cDNA with oligos PAIu and PAId, and the GPI-anchor signal by amplifying the uPAR cDNA with oligos GPIu and uPARd. The PAI-1/linker fragment (Kpn1–HindIII) and the GPI fragment (HindIII–Not1) were assembled into Kpn1–Not1 digested pcDNA5/FRT/TO. The Vn(1–66)/Fc fusion protein was generated by attaching the first 66 amino acids of mature human Vn to a constant region of a human IgG. Primers and templates were as follows: a vector containing part of the Vn cDNA (pTrx-Vn(1–97) []) amplified with oligonucleotides Vn(1–66)u and Vn(1–66)d and a vector containing an Fc region of human IgG (pIG-1 []) amplified with primers FcU and FcD. The Vn(1–66) (Kpn1–Xho1) and Fc (Xho1–Not) fragments were assembled in Kpn1–Not1 digested pcDNA5/FRT-TO. Variants of this construct where the SMB domain was deleted (Vn(40–66)/Fc) or where only the signal peptide was retained (Fc) were generated as above, substituting the oligo Vn(1–66)u with oligos Vn(40–66)u and SigU, respectively. The correct sequences of the complete coding regions of all constructs were verified by sequencing. All oligonuclotide sequences can be found in Table S1 (available at ). The expression vector encoding paxillin-GFP was provided by Dr. Alan F. Horwitz (University of Virginia, Charlottesville, VA; ). Alanine substitutions were generated by site-directed mutagenesis according to the QuikChange protocol (Stratagene). Alanine residues present in the uPAR sequence were substituted with glycine. Multiple rounds of mutagenesis were used to generate constructs carrying multiple substitutions. Expression vectors encoding soluble uPAR variants were generated by a second round of site-directed mutagenesis changing codon 284 of uPAR into a stop-codon (oligo pair A284Stop). Parental HEK293 Flp-In T-REx cells were cultured in DME supplemented with 10% FBS, 100 U/ml penicillin, 100 U/ml streptomycin, 5 mM glutamine, 15 μg/ml blasticidin, and 100 μg/ml zeocin at 37°C in 5% CO. Parental CHO Flp-In cells were cultured in Ham's F12 supplemented as above, but without blasticidin. Transfections were performed with a 1:10 ratio of pcDNA5/FRT/TO-based expression vector and pOG44 using FuGENE 6, and stable HEK293 Flp-In T-Rex and CHO Flp-In transfectants were selected in medium lacking zeocin using 150 μg/ml and 300 μg/ml hygromycin B, respectively. Expression in HEK293 Flp-In T-REx was induced by adding 1 μg/ml tetracycline to the medium overnight. For the production of soluble uPAR mutants, semi-confluent CHO Flp-In stably transfected with the relevant expression vectors was washed with PBS and incubated for 7–10 d in CHO protein-free medium. The concentration of suPAR in the conditioned media was typically higher than 100 nM as determined by ELISA. The conditioned medium was used for in vitro binding assays without further purification. Recombinant Fc-fusion proteins were expressed in the same way, but were purified by standard Protein A affinity chromatography. Cell surface expression of integrins, uPAR, and PAI-1/GPI were analyzed by flow cytometry. αβ, αβ, and β-integrins were detected using primary antibodies (10 μg/ml). PAI-1/GPI was detected using a polyclonal antibody (2 μg/ml) and uPAR was detected using the monoclonal antibodies R2 and R4, as well as a polyclonal rabbit anti-uPAR antibody (all 2 μg/ml). Cells were stained with appropriate secondary FITC-labeled antibody (diluted 1:100) and analyzed by flow cytometry (FACSCalibur; BD Biosciences). For the scoring of cell morphology, changes induced by the different mutant receptors in semi-confluent cells were inspected by phase-contrast microscopy by three to five independent, trained observers. The observers were asked to score the morphology of the cells in comparison to uPAR and mock-transfected cells analyzed in parallel. A cumulative score of “uPAR” or “mock” morphology was assigned to clones where unanimous scoring by the individual observers was achieved. A cumulative score of “intermediate” morphology was assigned to clones where the majority (two or more) of the observers gave an “intermediate” score. Adhesion assays were performed as described previously (). In brief, cells were harvested, counted, and allowed to adhere in the presence or absence of 10 nM pro-uPA for 30 min at 37°C to 96-well plates (3 × 10 cells/well) coated with the different substrates as indicated. After washing, the adherent cells were fixed, stained with crystal violet, and quantified by measuring the absorbance at 540 nm. Coatings were as follows: 100 μg/ml poly--lysine, 10 μg/ml fibronectin, 10 μg/ml type-1 collagen, 20 μg/ml laminin, and 1 μg/ml vitronectin (Vn). Vn(1–66), Vn(1–66), Vn(40–66), and Fc were all coated at 1 μg/ml. Cells were plated on glass coverslips and allowed to adhere overnight in the presence (293 cells) or absence (CHO cells) of 1 μg/ml tetracycline. Cells were washed in PBS, fixed in 4% paraformaldehyde, quenched with 30 mM NHCl, permeabilized with 0.1% Triton X-100, 0.2% BSA in PBS, and blocked with 2% BSA in PBS. Labeling of F-actin was performed using 0.2 μg/ml phalloidin-FITC, 1% BSA in PBS. TIR-FM imaging of cells was performed using a Biosystem TIR-FM workstation (Olympus) based on the Cell^R Imaging System (Olympus). A 488-nm argon laser was coupled in an inverted epifluorescence motorized microscope (IX81; Olympus) and focused at an off-axis position of the objective back focal plane. Cells plated on glass coverslips were viewed through a high-aperture 60× objective lens (UIS2 60× TIRFM PlanApo N, NA 1.45; Olympus) with an additional 1.6× lens. Images (16-bit depth) were acquired using an Orca-ER (C4742-80) Cooled CCD digital camera (Hamamatsu). Coverslips were directly inserted into an Attofluor Cell chamber (Invitrogen), rinsed with PBS (to maintain the requested difference of refractive indexes) and subjected to TIRF analysis. Time-lapse TIR-FM imaging was performed as above, with the exception that fixation/permeabilization/staining was omitted and that the cells were maintained at 37°C in normal growth medium throughout the recordings (every 15 s for a total of 15 min). Adjustment of brightness/contrast and smoothening of images was done using ImageJ 1.38i and always applied to the entire image. Phase-contrast and time-lapse live-cell imaging was performed at 37°C, 5% CO with an inverted microscope (IX70; Olympus) equipped with an incubation chamber (Solent Scientific). Cells were always plated in serum-containing growth medium unless otherwise stated, and viewed through 20× (LCPlanFl, NA 0.4 Ph1, Olympus) or 40× (UIS2 40× UPlanFLN, NA 0.75 Ph2; Olympus) objective lenses with an additional 1.5× lens. All time-lapse acquisitions were performed using the 20× objective. The acquisition system includes a digital camera (Sensys; Roper Scientific) and System Control Software Metamorph 7.0r4 (Universal Imaging). Adjustment of brightness/contrast and smoothening of images was done using ImageJ 1.38i and always applied to the entire image. Cell migration speed was quantified with ImageJ 1.38i using the plug-in “manual tracking”. In each experiment, 20 randomly chosen cells were tracked and their average migration speed throughout the experiment was calculated. In vitro binding assays were performed essentially as described previously () using Nunc Maxisorb black-well plates for the detection of bound suPAR with an Eu-labeled polyclonal anti-uPAR antibody (0.5 μg/ml) followed by the measurement of time-resolved fluorescence. Binding to immobilized Vn was measured using an excess of suPAR (80 nM) in the presence of a limiting concentration of pro-uPA (20 nM). All measurements were done in triplicate, and the specific binding calculated by subtraction of the nonspecific binding to BSA-coated wells. Cells were washed and lysed directly on the culture dish in ice-cold lysis buffer (25 mM Tris, pH 7.6, 150 mM NaCl, 1% Triton X-100, protease inhibitor cocktail [Complete-EDTA-free], 1 mM PMSF, 1 mM EDTA, 1 mM NaF, and 1 mM NaVO). After clarification by centrifugation (16,000 rcf, 15 min, 4°C) the total protein content was determined using the DC-Protein assay (Bio-Rad Laboratories) with BSA as standard. Equal amounts of total protein were separated by SDS-PAGE and probed as indicated. The significance of differences in cell adhesion and ERK1/2 activation states were established using test (paired, two-tailed) after log-transformation of the data from independent experiments. The surface-accessible area of the residues of uPAR was calculated based on the A-chain of pdb-entry 1YWH (). Calculations were performed using the program AREAIMOL from the CCP4 suite of programs () with a probe-radius of 1.4 Å. Fig. S1 (A and B) shows the profiling of Vn receptor expression and functionality in 293 cells. Fig. S1 (C and D) provides an example of comparable expression levels of the different mutants expressed in 293 cells. Fig. S2 A demonstrates that adhesion to Fn is unaffected by the expression of Vn-deficient uPAR mutants in 293 cells. Fig. S2 B and Fig. S4 show the rescue of morphological changes induced by exogenous pro-uPA for some uPAR mutants in 293 and CHO cells, respectively. Fig. S3 A indicates the migration speed of 293 cells expressing either Vn or integrin-deficient uPAR mutants. Fig. S3 B illustrates the crystal structure of the PAI-1–SMB complex and the position of the attached GPI anchor. Fig. S3 C illustrates the comparable cell surface expression level of uPAR and PAI-1. Table S1 lists information on all the mutants generated in this study, as well as all sequences of all oligonucleotides used in the study. Video 1 illustrates the increase in cell migration upon uPAR expression in 293 cells. Video 2 shows the redistribution of paxillin-GFP upon uPA-mediated rescue of uPAR/Vn binding in CHO cells. Video 3 shows the scattering effect upon uPA-mediated rescue of uPAR/Vn binding in preformed CHO colonies. Online supplemental material is available at .

sup xref #text Time-resolved confocal microscopy using fluorescent probes revealed localized Ca release events termed Ca sparks () that result from clustered openings of RyRs, probably under the participation of CICR. In frog muscle, their frequency was found to increase with fiber membrane depolarization, supporting the hypothesis of a voltage-dependent recruitment of Ca release quanta (). However, to the research community's disappointment, such quantal events were not detectable in skeletal muscle fibers of mammals, shedding doubts on a general physiological role of spark-based quantal Ca release in skeletal muscle. Only muscle fibers whose outer membranes were permeabilized by saponin (which leaves the SR membrane functional) showed sparklike events (). Sparks are also more readily seen in developing muscle cells but not in regions of these cells that respond with Ca release to membrane depolarization (). It has been proposed that the DHPR–RyR alignment prevents sparks from occurring (). The search for conditions to elicit Ca sparks in intact mammalian muscle finally led to success when demonstrated localized Ca fluctuations in isolated mouse muscle fibers challenged with external solutions of different osmolarities. An increase of osmolarity to ∼150%, which was synchronous with an increase of the external Ca concentration (by raising CaCl from 2.5 to 50 mM), elicited shrinkage and localized Ca transients in the periphery of the fibers. Such signals could also be observed on return to a normosmotic bath solution after a temporary decrease in osmolarity to 60%, which caused cell swelling. In normal mouse muscle fibers, the spark activity during recovery from swelling ceased within several minutes. In contrast, fibers from the mdx mouse, an animal model of human Duchenne muscular dystrophy resulting from lack of the cytoskeletal protein dystrophin, showed persistent, apparently irreversible spark activity. A defective Ca response has long been suggested to be responsible for the death of muscle fibers in this devastating muscle disease (). The results of provided further direct evidence for this notion. In the present issue of this journal, a new study from the Ma laboratory () reports experiments on muscle fibers from mice close to the end of their natural life span (∼2 yr). These fibers exhibit just the opposite effect to that seen in dystrophic muscle (i.e., a drastic shortening of the period of spark activity after brief hyposmotic stimulation). Weakening of skeletal muscle is one of the burdens accompanying the aging process, and part of the loss of muscle force in old age has been attributed to alterations in the Ca release machinery. For instance, it has been shown that the DHPR/RyR ratio decreases with age (). The now discovered drastic changes in osmotic stress-provoked local Ca response is a new facet of altered Ca regulation in aged muscle. Interestingly, observed a very similar phenomenon in young mice deficient in the protein MG29. This protein normally sits in the membrane area of the SR that faces the T system (i.e., in a region critical for Ca handling). It resembles the synaptophysins, which are found in the membrane of synaptic vesicles, and is one of several small proteins that are suspected to modulate the Ca release mechanism. MG29 has been reported to affect store-operated Ca entry and the susceptibility of muscle to fatigue (). Because the level of MG29 was found to be reduced by ∼50% in skeletal muscle of old mice, propose that this protein plays a role in the weaker Ca response to osmotic stimulation of aged muscle fibers. They suggest that the substantially altered spark activity might result from a reduced sensitivity of the RyRs to CICR. Another contributor to the rapid decline of spark frequency could be the reduced load of Ca in the SR that was found for muscle fibers of both aged and MG29-deficient mice. Evidence from reconstituted RyRs and from single cell experiments indicates that the concentration of Ca on the luminal side of the SR affects the opening probability of Ca release channels (; ). Calsequestrin, the major luminal Ca-binding protein of the SR, has been suggested to serve as a Ca sensor to modulate RyR under the participation of triadin and junctin, two further SR proteins of the T system–SR junction (). Other interesting parallels between old mice and MG29-null mice reported by include muscle atrophy and characteristic changes in ultrastructure, indicating a disruption of the alignment of the SR and T system and fragmentation of SR regions. The structural changes and the altered force response to depolarization and caffeine led to the suggestion that the T system loses control of sections of the SR, possibly as a result of the loss of MG29 as an essential component in maintaining SR–T system alignment. In view of the similarity of MG29 with synaptophysin, a membrane protein of still unknown function concentrated in the clear and dense core synaptic vesicles, the study of muscle might ultimately be useful for the understanding of regulated exocytosis in neurons and neurosecretory cells. There is no doubt that the age-related dysfunction of muscle is a complex process, and it remains to be established whether MG29 is indeed a key element in this process, as the new results seem to indicate. MG29's possible involvement in the generation of sparks is intriguing and may open a new avenue to understanding the mechanism that links osmotic changes to Ca signaling in muscle. Are the sparks observed under these conditions indeed comparable with the stereotyped signals found in other muscle preparations? They may simply be artifacts caused by a mechanical perturbation of the membranes leading to local leaks of Ca into the cytoplasm. Alternatively, are they part of a signal transduction chain involving specific mechanosensitive elements, perhaps even designed to modulate the Ca signal during the mechanical stress of muscle contraction? If so, is MG29 part of the mechanism? There is much work ahead. Sparks may get old, but not old fashioned.

xref #text Although they are not the only source of oxidants, the NADPH oxidase (Nox) family members are the principal complexes that function solely to redox-couple NADPH and molecular oxygen to generate O and, thence, HO. Thus, the examination of Nox biology reveals much about the cellular logic behind regulated oxidant production. The seven known human Noxs include Nox1–5 and Duox1–2, with Nox2 (gp91) being the founding member. As a family, these oxidases participate in a variety of adaptive functions, ranging from mitogenesis to immune cell signaling (; ). Reflecting these varied biological roles, the Nox proteins have been implicated in several cell-fate pathways, such as the Ras mitogenic pathway (), the MAP kinases (; ), the JAK–STAT pathways (), and NF-κB (; ). Nox-dependent signaling has been a biologically successful device by all accounts, having appeared early and persisted throughout evolution on the aerobic earth. Orthologous Nox genes arose in concert with multicellular organization (), and so are found as early as the slime mold and the filamentous fungus (; ). During starvation conditions, free Dictyostelial amoebae aggregate into a slug that behaves as a single organism, differentiating a distinct organ, the spore-bearing fruiting body. Although single deletions of any of the three genes or fail to produce a phenotype in unicellular amoebae, starvation of these knockout mutants interrupts fruiting body morphogenesis (). Similarly, deletion of either of the two Nox genes results in failed fruiting body differentiation. (). Thus, Noxs control developmental signaling in the most primitive multicellular organisms, an ancestral function that foreshadowed their later involvement in basic mammalian cell fate pathways. One might fairly ask why the utilization of reactive oxidants has been so evolutionarily durable and how oxidants can manage to selectively relay a diverse array of signaling cassettes, especially because the different Noxs presumably produce the same oxidant species perceived by the cell as an oxidative threat. #text Different cells, when imaged with different oxidant-detection methods, display subcellular restriction of oxidant activity around regions of cytoskeletal rearrangement (). Again, considering the Nox proteins as archetypal signaling oxidases, molecular links between the oxidase and the cytoskeleton have been described. Activation of the phagocyte oxidase, for instance, causes translocation of the adaptor p47 and the activator p67 to the cytoskeletal fraction such that the functioning oxidase is quantitatively cytoskeleton bound (; ). More recent studies have demonstrated constitutive cytoskeletal targeting of oxidase subunits in nonprofessional phagocytes such as endothelial cells (; ). In these cells, cytoskeletal disruption interrupts oxidant-mediated JNK signaling, suggesting a connection between cytoskeletal targeting, oxidant production, and the relay of signaling information (). A specific example of oxidant-dependent cytoskeletal function is the dependence of endothelial cell migration on Nox proteins (; ). In these cells, oxidants concentrate within membrane ruffles, mirroring the distribution seen in stimulated adherent neutrophils (; ). Interestingly, p47, which is the principal kinase target during oxidase activation, directly binds two proteins enriched within leading edge lamellipodia–moesin and WAVE1 (; ). The latter protein catalyzes dendritic actin nucleation responsible for lamellar structure in a Rac1-dependent fashion; accordingly, p47–WAVE1 complexes contain Rac1 and the Rac1 effector PAK1, and antioxidants or truncations that disrupt p47–WAVE1 interactions diminish ruffle formation (). p47 also associates with cortactin, which is a protein involved in lamellipodial persistence, although it is unclear whether the proteins associate directly or indirectly through larger cytoskeletal complexes (). Another targeting device for p47 within specific lamellar structures appears to be TRAF4, an orphan that, unlike its paralogues, has not been demonstrated to actively function in innate immune signaling. Knockout models suggest that mouse TRAF4 and the orthologue dTRAF1 instead control ontogenic migration during respective dorsal closure events (; ). In the fly, dTRAF1 operates within a Rho-GTPase/JNK cassette during cell migration, and a parallel situation in human endothelial cells may require a direct interaction between TRAF4 and p47. This interaction governs oxidant-dependent JNK activation, and endothelial cell migration involves TRAF4-dependent activation of the NADPH oxidase through the Rho-GTPases and PAK1 (; ). TRAF4 and p47 target focal integrin complexes within the lamellipodia of motile endothelial cells, tethered by the focal contact scaffold Hic-5. Thus, TRAF4 appears to function in this regard by focusing the activation of the oxidase to a specific cytoskeletal structure. Besides p47 phosphorylation, Rac1 activation is also required to activate many Noxs; thus, sites of Rac1 activation may also be expected to specify the subcellular location of Nox-dependent signaling complexes. Active Rac1, for instance, concentrates within ruffling lamellae, suggesting spatial coordination of Rac's cytoskeletal and prooxidant effects (). One potential mechanism for Rac1 targeting is through the actin-binding scaffold IQGAP, which targets leading edge actin structures and mediates cell migration (). IQGAP not only binds and, therefore, localizes the active forms of Rac1 and Cdc42 but also associates with VEGFR2 and Nox2 at leading edge structures, mediating VEGF-dependent oxidant production (). Another tactic cells use to spatially restrict Rac1 function is local exclusion of Rho-GDI. A striking example of how Rho-GDI specifies Nox activation sites was recently demonstrated in the plant . Focal cytoskeletal rearrangements within the specialized trichoblast cell cause a single root hair to extend from each cell. A mutation resulting in the root hair–defective phenotype localizes to the gene for a plant Nox, RHD2/AtrbohC (). Although wild-type plants produced oxidants confined to the tip of extending root hairs (), mutants neither produced oxidants nor formed root hairs. Conversely, diffuse exposure to exogenous oxidants caused loss of spatial control, with the resultant formation of numerous aberrant root hairs. Two subsequent mutants causing a similar phenotype of multiple aborted growth bulges () were found to encode SCN1, which is a Rho-GDI (). Whereas wild-type plants demonstrated a single focus of oxidant production at the growing root hair tip, mutants displayed multiple foci of oxidants corresponding to abnormal growth bulges. Therefore, the plant Rho-GDI SCN1 functions to restrict oxidant production exclusively to a single root tip. A third method of localizing Rac1 activation is through targeting of Rho guanine nucleotide exchange factors (GEFs) with Rac1 activity. Recruitment of a Rac1 GEF is suggested by the association of human Nox1 with the Rac GEF βPIX (). Thus, βPIX, which is known to modulate EGFR function, activates Rac1, causing EGF-dependent oxidant production. In addition, Rap1a, which associates with the Nox2 complex, targets membrane protrusions and locally activates the Rac GEFs Vav2 and Tiam1, and thus Rac1 itself, at the lamellipodial edge (). Membrane rafts are known to facilitate the congregation of several signaling proteins, including Nox subunits. In suspended myeloid cells, for example, the Nox2 cytochrome subunits constitutively sequester in raft fractions, with translocation of the soluble proteins p47 and p67 into rafts after stimulation (). Raft association of the mitogenic Nox1 has also been noted in smooth muscle cells (), and angiotensin II stimulation, which proceeds through Nox1, promotes Rac1 trafficking into rafts, whereas raft disruption blocks angiotensin II–dependent oxidative signaling (). Similarly, rafts contain the focal complex–associated TRAF4, and raft disruption blocks TRAF4-dependent oxidative signaling (). The association of Nox proteins with raft microdomains may explain, in part, why oxidant production by Noxs, which is presumed to be directed outside the cell, can affect intracellular targets. Plasma membrane rafts containing Rac1 are known to be internalized in response to integrin signals (), and caveola-derived signaling endosomes, which are a type of membrane-derived “signalosome,” continue to transduce growth factor receptor signals after internalization. Indeed, small caveolin-containing vesicles termed cavicles are thought to be transported, possibly as microtubular cargo, between the plasma membrane and pericentrosomal caveosomes (). Although it is as yet unclear whether functioning Nox complexes are transported within similar internalized structures, Nox2, p47, p67, and p22 clearly exist in discrete, detergent-insoluble complexes within the cytosol of endothelial cells in association with microtubules (; ). More recently, IL-1β has been shown to activate Nox2 within early endosomes containing IL-1R (). The possible functioning of Nox complexes within these or other intracellular membranous structures warrants further investigation. Mitochondria have long been known to represent focal sources of reactive oxidants, and more recently, have been appreciated as important signaling organelles. Mitochondria, for instance, regulate several facets of cellular energetics beyond ATP production, at least some through local oxidant production. AMP-activated protein kinase (AMPK), which is believed to serve as an energy gauge, is activated by mitochondrial oxidants, perhaps through mitochondrial c-Src (). AMPK controls several energy-related pathways, including the inhibition of acetyl CoA carboxylase with suppression of fatty acid synthesis and the activation of glycolysis and β-oxidation. Under hypoxic conditions, mitochondrially derived oxidants cause activation of AMPK; the compound metformin, which is commonly used to treat diabetes, activates AMPK, again, through mitochondrial oxidant production (; ). Another mediator of cellular energetics is pyruvate, a watershed metabolite that drives mitochondrial respiration. Pyruvate-induced mitochondrial oxidants appear to activate JNK, leading to inhibition of GSK-3β and activation of glycogen synthase, thus, sequestering glucose and lowering pyruvate in a negative feedback cycle (). It is not clear whether this oxidative signaling is restricted to the local mitochondrial environment or what the proximal oxidant target is; however, both JNK and JNK scaffolds, as well as the putative downstream target GSK-3β, associate with mitochondria, allowing the possibility of a locally confined circuit (; ). Mitochondrial redox signaling in response to nutrient availability is likely to have ancient roots. The simple colonial hydroid , for instance, responds to changes in its food supply by adopting either dense feeding or runner-like searching colony morphologies. Interestingly, these morphologic changes appear to be controlled by changes in mitochondrial redox states (). More generally, across many phyla several connections between cellular energetics, mitochondrial oxidants, and aging phenotypes have been noted (). Perhaps at some level related to energy management, mitochondria also play a central role in programmed cell death; mitochondrial oxidants are well known to mediate this form of death. Less clear are the exact mechanisms by which mitochondria are stimulated to produce increased oxidants, and what the proximate targets of such oxidants are. Recently, the proapoptotic protein p66 was shown to localize to the mitochondrial intermembrane space and redox cycle with cytochrome to produce oxidants that induce the permeability transition (). Such oxidants are thought to locally target the inner mitochondrial membrane, causing both depolarization and cytochrome release (), although other targets may be important. O produced outside of purified mitochondria, for instance, causes massive cytochrome release without inner membrane damage in a process targeting the outer membrane voltage-dependent anion channel (VDAC; ). It is unclear whether VDAC itself is an oxidant target or, perhaps more likely, is required for ingress of O into the intermembrane space, but these data nevertheless support the notion that cytochrome release proceeds as a result of local effects of mitochondrial oxidants. xref #text To the extent that homeostatic signaling requires spatial confinement of oxidants, the cell may in many instances recognize oxidative stress through the detection of diffuse cytosolic oxidants that have escaped their usual designated locations. The appearance of oxidants out of an appropriate spatial context may represent reasonable cause to alert the cell to a dangerous excess of exogenous or pathologically controlled endogenous oxidants, to activate either defense or fail-safe death programs. The model whereby subcellular localization of oxidants (or lack thereof) discriminates homeostatic from stress signaling allows several predictions. First, oxidative stress pathways should be triggered in response to the delocalized appearance of cytosolic oxidants. Typically, for instance, oxidative stress pathways are activated by suffusing the cell or organ with membrane-permeant oxidants such as HO or by irradiation with UVB, both of which would be expected to blanket the cell with oxidants. Hypoxia reoxygenation also induces oxidative stress; it does so in many tissues via xanthine oxidase, which is a cytosolic protein. Mitochondria serve as a principal source of oxidants in several forms of oxidative stress, including reoxygenation and hyperoxia states. Both of the latter conditions increase global indices of cellular oxidative effects, indicating significant escape of oxidants into the cytoplasm. Indeed, mitochondrial release of O into the cytosol is controlled by VDACs, and mice with heterozygous deficiency of mitochondrial SOD sustain oxidative damage to nuclear, as well as mitochondrial, DNA (; ). Even in yeast, senescence accompanies the accumulation of oxidatively modified proteins, more than half of which are cytosolic (with the remainder being mitochondrial; ). Second, one would expect to find oxidative stress reporters free within the cytosol. The early prototypes OxyR and Yap1p are found within the bacterial protoplasm and yeast cytosol, respectively, and become activated in response to exogenous HO through disulfide bond formation and transcriptional activation (; ). Redox-sensitive cytosolic reporters that translocate into the nucleus persist in mammals. Redox factor-1, for instance, translocates from a diffuse cytosolic location into the nucleus with oxidative stress to facilitate the DNA-binding activity of NF-κB (). Thioredoxin also functions as an oxidative stress reporter, moving into the nucleus to activate NF-κB, AP-1, and p53 (; ). Third, the cell may be expected to deploy the oxidative stress mechanism in response to other forms of cellular stress through a secondary increase in cytosolic oxidants. Heat shock, for instance, increases mitochondrial oxidant production, thereby activating HSF-1 through oxidant intermediates, whereas antioxidants diminish the heat shock response in (; ; ; ). Heavy metals induce a large burst of HO that activates the cytosolic factor HSF-1 (). Notably, this oxidant production is suppressed by Rac1(N17), suggesting specific oxidant regulation. Finally, p53 not only responds to oxidative stress within the cytosol but also activates stress pathways by increasing mitochondrial oxidant production. The induction of oxidative stress simply through p53 overexpression highlights the broad utility of this mechanism as a general response device, even to nonoxidative genomic stress (). If, indeed, oxidants require a high degree of spatial ordering to confer signal fidelity, one might wonder why organisms did not evolve a more robust cytosolic antioxidant defense to completely suppress stray redox signals and minimize oxidant stress. One possible answer may be that an excessively high level of cytosolic antioxidants would be expected to dampen local redox signals. A second answer may lie in the speculation that oxidative stress pathways may have evolved before localized oxidant signaling, meaning that pathways related to the latter had to be retrofit into an organism that already used oxidant production and sensing in its alarm system. As mentioned earlier, in we find enhanced mitochondrial oxidant production after heat shock, leading to the activation of cytosolic reporters and transactivation of stress response genes; therefore, exuberant scavenging of cytosolic oxidants may gainsay what, in this case, would be a protective stress response. Both answers reveal the redox tightrope the cell is required to walk to spatially discriminate homeostatic from stress signaling. This issue becomes particularly vexing in regard to the mitochondrion, which, despite its ability to confine its oxidative effects locally, can alternatively flood the cell with oxidants and damage itself in the process, functioning as a principal loudspeaker for sounding oxidant stress alarms. This scenario highlights the exquisite control required for both oxidant production and its escape into the cytosol. The consequence of losing such control would appear to be inappropriate stress responses, such as unscheduled cell cycle arrest or apoptosis, or insensitivity to real stress with failure to activate these processes. Not surprisingly, human states that reflect these same cellular signaling defects result in either degenerative or neoplastic diseases that arise in the context of either excessive oxidative stress or insensitivity to such stress. The ubiquity of SOD in aerobic cells indeed reflects the dire consequences of poor oxidant regulation. The cellular strategy of subsequently adopting these oxidants for signaling purposes appears to have required the evolution of spatial control, incorporation into other general signaling devices, and the preservation of a global oxidative distress pathway. When Emperor Joseph II complained about the commissioned opera that there were “too many notes, my dear Mozart,” Mozart is said to have responded: “(There are) exactly the right number, your Majesty.” This comment appears to apply to reactive oxidants as well, with the further caveat that they should be in exactly the right places.

The centrosome nucleates and organizes microtubules in animal cells. It consists of a pair of cylindrically shaped centrioles surrounded by fibrous pericentriolar material. Before or during DNA replication in S phase, the centrioles split, and each cylinder serves as a template for the assembly of a new “daughter” centriole. Before mitosis, when cells contain two pairs of centrioles, each pair serves as a nucleation center for microtubules of the spindle apparatus. Defects in centrosome assembly or in centrosome separation can result in defective nucleation of spindle microtubules, and in several cases, in the formation of monopolar spindles and mitotic arrest (; ). Several years ago, evidence was published that defective centrosome assembly can prevent cells from entering S phase. In particular, removal of the centrosome by microsurgery or by laser ablation resulted in a cell cycle arrest, as did inhibition or silencing of several centrosome-associated proteins, such as dynactin, PARP-3, centriolin, or AKAP450 (; ; ; ; ; ). The mechanism leading to this centrosome-dependent cell cycle arrest in G1 phase has been unclear; it was proposed that a checkpoint control would prevent those cells with imperfect centrosomes from continuing the cell cycle, to prevent the assembly of defective spindles later in mitosis (). In this study, we followed cell cycle progress after inhibition of centrosome assembly by depleting the pericentriolar proteins pericentriolar material 1 (PCM-1) and pericentrin. These proteins have been shown to be necessary for the assembly of other centrosomal constituents (; ; ). We found that depletion of PCM-1 or pericentrin activates the p38-dependent stress pathway and the p53-dependent cell cycle checkpoint. We have previously shown that depletion of the protein PCM-1 leads to defects in the assembly of the centrosomal components centrin, ninein, and pericentrin, and to an altered organization of the microtubule network in interphase cells (). To investigate the consequences of PCM-1 depletion on the cell cycle, we performed RNA silencing experiments in primary human fibroblasts, MRC-5. After 72 h, PCM-1 depletion was monitored by immunofluorescence () and Western blotting (). Depleted cells were tested for incorporation of BrdU into the nucleus, as an indicator of DNA synthesis (). We determined that in PCM-1–depleted cells only 15 ± 4% incorporated BrdU, as compared with 35 ± 3% in controls, as expected for a normal cycling population (). This is consistent with previous reports on microinjection of PCM-1–inhibiting antibodies () and on centrosome removal by microsurgery or laser ablation, which prevent cells from entering S phase (; ). Several years ago, experiments on cells treated with the microtubule drugs colcemid, nocodazole, and taxol indicated that untransformed cells are arrested in G1 phase, when microtubules are depolymerized or when microtubule dynamics are altered (; ; ). This raises the question of whether DNA replication in PCM-1–depleted cells is inhibited because of an altered microtubule network, or whether defects at the centrosome itself suffice to induce a cell cycle arrest. Therefore, we depleted a second centrosome protein, pericentrin (), which in contrast to PCM-1, only slightly reduces microtubule density but seems to have no significant effect on microtubule anchoring at the centrosome (). Consistently, depletion of pericentrin also led to a reduction of BrdU incorporation (). Because these data indicated that cells failed to undergo S phase–dependent DNA replication when missing the full complement of centrosome proteins, we wanted to test in more detail at what stage cells are arrested. Immunoblotting of cell extracts of PCM-1–depleted cultures indicated drastic reduction of cyclin A (), which is normally found expressed in cells during late G1, S, and G2 phases. Furthermore, we detected that PCM-1–depleted cells also showed reduced amounts of the proteins minichromosome maintenance deficient 3 (MCM3) and proliferating cell nuclear antigen (PCNA) acting in licensing and DNA replication () and reduced percentages of cells expressing the cell proliferation marker Ki-67 (). Equivalent data were also obtained by depletion of pericentrin (). Altogether, these data indicated that depletion of centrosome proteins reduces the number of cells entering S phase. We verified this hypothesis by comparing profiles of cell cultures analyzed by flow cytometry (): consistent with our data on BrdU incorporation, PCM-1–depleted cells showed a reduction of S phases from 28 to 15%. We then wanted to determine whether S phase entry was blocked because of checkpoint activation in cells depleted of PCM-1 or pericentrin. We found that the overall levels of the retinoblastoma protein (pRb) were reduced to 38 ± 17% and that most of the remaining pRb, normally hyperphosphorylated during late G1 and S phase, was present in its faster migrating, hypophosphorylated form (77 ± 13% in depleted vs. 38 ± 13% in control cells; ). The checkpoint protein p53, however, was found up-regulated, especially after prolonged depletion of PCM-1 or pericentrin (). In depleted cells, the Cdk2 inhibitor p21 was found equally up-regulated (). These data suggested that depletion of the two centrosome-associated proteins PCM-1 and pericentrin leads to the activation of the p53-dependent checkpoint. In the next step, we wanted to determine whether cell cycle progress would be affected if the p53-dependent checkpoint control was abrogated. For this purpose, we attempted simultaneous depletion of p53 and PCM-1 by cotransfecting siRNA oligomers against both. shows that p53 levels could not be reduced when PCM-1 was missing. We tried to refine this experiment by sequential depletion of MRC-5 cells, by first depleting >90% of p53 after 72 h, followed by simultaneous siRNA treatment against p53 and PCM-1. However, we observed that under these conditions, p53 levels increased back to 40–50% in three different experiments (unpublished data). We concluded that p53 turnover is altered and that the residual p53 protein might be stabilized in the absence of an intact centrosome. We therefore changed our experimental strategy and compared cell cycle progress after PCM-1 depletion in several cell lines lacking p53. We used mouse embryonic fibroblasts from p53 knockout mice (unpublished data) as well as the human lung carcinoma cell line H1299. Because of the loss of p53 checkpoint control, both lines displayed a relatively high basic rate of DNA synthesis (). We found that in both p53−/− cell lines, PCM-1 depletion did not inhibit cell cycle progress. The levels of PCNA, MCM3, and hyperphosphorylated pRb remained high (, H1299). In addition, we also tested the effect of PCM-1 depletion in the p53+/+ and p53−/− lines of HCT116 cells. Unfortunately, only 30% of these cells showed lower PCM-1 levels. BrdU incorporation in these was reduced from 42% in controls to 33% in partially depleted p53+ cells, whereas p53− cells showed BrdU incorporation in 49% after partial PCM-1 depletion. Consistently, HeLa cells with functionally suppressed p53 checkpoint control do not arrest in the absence of centrosomes (). In contrast to p53, the removal of pRb did not cause resumption of the cell cycle in PCM-1–depleted MRC-5 cells, because the percentages of BrdU-incorporating cells and Ki-67–expressing cells remained low, as did the expression of MCM3 protein (). On the other hand, the amounts of cyclin A and PCNA were restored to nearly control levels. A possible scenario would be that centrosome defects activate both pRb and p53 in parallel, with p53 having feedback effects on pRb phosphorylation and pRb protein levels but not vice versa. This would be consistent with our observation that pRb levels drop after PCM-1 or pericentrin depletion and that depletion of pRb itself does not fully restore S phase activity, which is probably blocked because of checkpoint control mechanisms directly dependent on p53. Eventually, loss of pRb might be compensated by Rb-related “pocket proteins,” such as p107 or p130. Finally, we addressed the question of how defects in centrosome assembly could activate the p53-dependent checkpoint. It has been reported that p38 MAPK is involved in the cell's response to a range of stress factors, such as UV irradiation, osmotic shock, heat shock, starvation, and cytokine treatment (). The response to these stress factors is mediated partly through p53 phosphorylation, which is believed to stabilize p53 and to increase the expression of the Cdk inhibitor p21, thereby blocking the cell cycle (). To assess the involvement of stress-activated p38 MAPK in response to centrosome defects, we have treated PCM-depleted MRC-5 cells with the p38 MAPK–specific inhibitor SB203580. This inhibitor blocks p38 activity but affects neither p38 protein levels nor p38 phosphorylation. We found that the inhibitor prevented cell cycle arrest, as indicated by resumption of DNA synthesis (). p38 inhibition in PCM-1–depleted cells also restored regular levels of cyclin A and hyperphosphorylated pRb, as well as expression of Ki-67 and MCM3 (). Further, the inhibition of p38 led to decreased levels of p21 but only slightly decreased p53 (). This could be explained by a stabilization of p53 after centrosome inactivation, as discussed in the previous paragraph, which might occur independently of p38-dependent p53 activation. Alternatively, the centrosome-dependent cell cycle arrest might not simply be a linear consequence of p53 activation via p38. For example, p53 might be activated by other kinases in addition to p38. Moreover, although p38 has been shown in multiple experiments to mediate cell cycle arrest by phosphorylating various sites of p53 (; ), p38 is also known to phosphorylate and thereby inactivate Cdc25A and cyclin D, which in turn arrests the cell cycle (; ; ; ). However, p38 MAPK activation in the absence of p53 seems to be insufficient for centrosome-dependent cell cycle arrest because PCM-1 depletion in p53−/− cells did not stop the cell cycle. In our PCM-1–depletion experiments, we observed a significant decrease in the expression of proteins associated with cell proliferation, such as MCM3, PCNA, or Ki67. However, because about half of the depleted cells were still entering S phase, we wanted to test whether centrosome defects only slowed down cell cycle activity, or whether depleted cells were predisposed to a stable arrest that did not affect all cells simultaneously. Immunofluorescence of p53 after PCM-1 depletion revealed that p53 levels varied between individual cells, with some cells showing a very high increase but a small increase or no obvious response in others (). We then tested whether the cells' response to centrosome inhibition induces exit from the cell cycle via senescence. Senescence is marked by the acquisition of a permanent G0 state and an increase of cellular β-galactosidase activity (). A cytochemical analysis of PCM-1–depleted cultures revealed a 10-fold increase of β-galactosidase–positive cells (), compared with a basal level of 3% in controls. Consistently, we previously reported that cultures of U2OS cells died after silencing of PCM-1 for long periods (), although the experimental conditions varied from the protocol used here. Senescence has been characterized as a cellular program activated as a result of physiological stresses preventing further cell proliferation (). Instead of a specific centrosome-dependent cell cycle control as proposed by , we propose a stress-driven response to centrosome defects. The idea of the centrosome as a center for stress-related signaling seems plausible, considering its central location in the focus of the microtubule network and its ability to bind various molecules of signaling pathways and of cell cycle regulation. For example, significant amounts of protein kinase A, polo-like kinase, and protein phosphatases 1 and 2A, as well as cyclin E and p53, have all been localized to the centrosome (; ; ). Their localization and anchoring is mediated by large coiled-coil proteins such as pericentrin, AKAP450, or AKAP220 (; ; ). Absence or inhibition of anchoring proteins could therefore disrupt cellular signaling pathways and elicit a stress response. Further research will be necessary to investigate the multiple interactions between centrosome proteins, cell signaling, and the complex regulation of the cell cycle. Human fetal lung fibroblasts MRC-5 (European Collection of Cell Cultures) and H1299 cells (a gift from B. Vojtesek, Masaryk Memorial Cancer Institute, Brno, Czech Republic) were cultured in DME (Invitrogen), supplemented with 10% fetal bovine serum (Perbio Science), 2 mM -glutamine, and antibiotics. HCT116 cells (a gift from B. Vogelstein, Johns Hopkins University, Baltimore, MD) were cultured in McCoy's medium containing the same supplements. The cells were plated to reach 80% confluency at the time of transfection. About 3 × 10 cells from early passages of MRC-5 and 0.5 × 10 cells of H1299 or HCT116 were transfected with 9 or 18 μg of siRNA oligonucleotides, respectively, in a Nucleofector electroporation device, using the appropriate transfection kit (Amaxa). The cells were then plated onto two 10-cm Petri dishes to obtain a confluency of ∼30–40% on the next day. Routinely, siRNA treatment was for a total of 72 h before fixation or preparation of cell extracts, unless indicated otherwise. P38 MAPK inhibition was performed in cells that were transfected with oligos and cultured overnight before the specific inhibitor SB203580 (Calbiochem) was added at a concentration of 10 μM. Inhibitor-treated cells were left in culture for another 48 h and then fixed in methanol or extracted for SDS-PAGE. Cells were harvested using trypsin-EDTA, counted, and washed in ice-cold PBS. Cells were extracted in SDS-PAGE sample buffer and boiled for 6 min. Amounts of extract equal to 200,000 cells were loaded and run on 7.5, 10, or 12.5% SDS-PAGE gels and blotted onto nitrocellulose. The following antibodies were used for Western blot analysis or immunofluorescence: affinity-purified rabbit anti–PCM-1 (), anti-MCM3 mAb clone 3A2 (MBL International Corporation), anti-pRB mAb clone 4H1 (Cell Signaling), anti-PCNA mAb clone PC-10 (Sigma-Aldrich), anti–cyclin E mAb clone HE12 (Zymed Laboratories), anti–cyclin A mAb clone Cy-A1 (Sigma-Aldrich), anti p53 mAb clone DO-1 (Novocastra), anti-p21 mAb clone SX118 (BD Biosciences), rabbit anti-ACTIVEp38MAPK polyclonal antibody (Promega), anti–α-tubulin mAb clone DM1A (Sigma-Aldrich), goat anti–mouse polyclonal antibody HRP (Promega), donkey anti–rabbit IgG polyclonal antibody HRP (GE Healthcare), rat anti-BrdU mAb (Harlan), anti–Ki-67 mAb clone MM1 (Novocastra), rabbit anti-pericentrin polyclonal antibody (Covance), donkey anti–rabbit or anti–mouse IgG conjugated with Alexa 488 or Alexa 594 (Invitrogen). Quantification of protein levels was performed by scanning immunoblots and analyzed using the Photoshop (Adobe) histogram tool. Quantification of cells expressing specific proteins was performed by counting siRNA-treated cells that were double stained with PCM-1 antibodies to verify depletion. The following siRNA oligomers with dTdT overhangs (QIAGEN) were used: PCM-1.2, corresponding to human PCM-1 (UCAGCUUCGUGAUUCUCAG); peric, corresponding to human pericentrin (GCAGCUGAGCUGAAGGAGA; ); pRB (GCCCUUACAAGUUUCCUAG); and p53 (CUACUUCCUGAAAACAACG). All cells in a culture dish were harvested by trypsinization, washed in ice-cold PBS, and fixed in 80% ice-cold ethanol in PBS. Before staining, the cells were spun down in a cooled centrifuge and resuspended in the cold. Bovine pancreatic RNase (Sigma-Aldrich) was added at a final concentration of 2 μg/ml, and cells were incubated at 37°C for 30 min, followed by an incubation in 20 μg/ml of propidium iodide (Sigma-Aldrich) for 20 min at room temperature. 10,000 cells were analyzed on a flow cytometer (FACSCalibur; BD Biosciences). β-Galactosidase staining at pH 6.0 was performed as described in . Cells grown on coverslips were fixed in ice-cold methanol and stored at −20°C until use. Antibody staining was performed using the reagents listed in the previous paragraphs, according to standard protocols. The percentage of cells in S phase was assessed by adding 100 μM BrdU (Sigma-Aldrich) to the cultures 30 min before fixation. Double labeling of BrdU and PCM-1 was performed by probing first for PCM-1, using rabbit anti–PCM-1 and fluorescent anti-rabbit antibody, and coverslips were postfixed in PBS containing 3.7% paraformaldehyde, treated with 2 M HCl for 30 min, and stained with rat anti-BrdU and fluorescent secondary anti-rat antibody. Cells were viewed with a fluorescence microscope (Axioskop 2; Carl Zeiss MicroImaging, Inc.) equipped with a camera (Axiocam; Carl Zeiss MicroImaging, Inc.) and software (Axiovision; Carl Zeiss MicroImaging, Inc.). The images were imported into Photoshop for presentation.

Eukaryotic cells are divided into many membrane bounded organelles that have unique protein compositions to perform a variety of specialized functions. Mitochondria are such organelles that consist of four compartments, the outer membrane, intermembrane space (IMS), inner membrane, and matrix. Because most mitochondrial proteins are synthesized in the cytosol, they are imported into mitochondria with the aid of translocator complexes in the outer and inner mitochondrial membranes (; ; ; ). More than 30 proteins have been identified as translocator components, indicating that pathways of import and sorting of mitochondrial proteins are much more complex than previously envisaged. The TIM23 complex in the mitochondrial inner membrane, which mediates protein translocation across the inner membrane and protein release into the inner membrane, consists of several different subunits (; ). Tim23 and -17 constitute the protein-conducting channel through which precursor proteins, usually with an N-terminal cleavable presequence, cross the hydrophobic barrier of the inner membrane in an unfolded state. Tim50 facilitates protein transfer from the TOM40 complex in the outer membrane to the TIM23 complex, and Tim21 is proposed to promote the coupling of the two translocator complexes. Mitochondrial Hsp70 (mtHsp70) in the matrix functions as an import motor to drive vectorial translocation and unfolding of the substrate precursor proteins in cooperation with its partner proteins, mitochondrial Hsp70–associated motor and chaperone (MMC) proteins. Tim44 provides an anchor for mtHsp70 to bind to the translocating polypeptide that emerges from the outlet of the TIM23 channel. Pam18/Tim14 (and Mdj2p) functions as a J protein for mtHsp70, and Pam16/Tim16 mediates association of Pam18 to Tim44. Pam17 is also proposed to facilitate coupling of Pam18 and -16 with Tim44. Yge1/Mge1 and Zim17/Tim15/Hep1 bind to the nucleotide-free form of mtHsp70 to promote its function. We report the identification and characterization of the gene product of , Tam41 (translocator assembly and maintenance 41), which is a peripheral inner mitochondrial membrane protein facing the matrix. The cells without Tam41 are defective in the import of precursor proteins destined for the matrix and the inner membrane via the TIM23 complex both in vitro and in vivo. Tam41 likely mediates maintenance of the functional integrity of the TIM23 complex from the matrix side of the inner membrane. On the basis of our analyses of the mitochondrial localization of yeast proteins that are indicated in the database to have essential but unknown functions (; ; ), we found Tam41, the gene product of , to be a candidate for possible new mitochondrial translocator proteins. The gene product of is reported as an essential mitochondrial protein in yeast (; ). However, when we deleted the gene in diploid cells and subjected them to tetrad analysis, all of the four spores grew normally on YPD at 23°C. The strain with chromosomal deletion exhibited slow growth at an elevated temperature (37°C) as compared with that at 23°C, and the temperature-sensitive growth was more prominent on nonfermentable (SCLac) media than on fermentable (SCD) media (Fig. S1, A and B, available at ). Tam41 comprises 385 amino acid residues with a calculated molecular weight of 44,199 and is predicted to possess a mitochondrial targeting signal at the N terminus. We thus analyzed the in vitro import of Tam41 into isolated yeast mitochondria (). When we translated Tam41 with reticulocyte lysate in vitro, a radiolabeled 41-kD protein was synthesized. Upon incubation with isolated yeast mitochondria, it was converted to a 39-kD form in a ΔΨ (membrane potential across the inner membrane)–dependent manner. The 39-kD form was resistant to proteinase K (PK) in mitochondria and in mitoplasts with ruptured outer membrane by osmotic swelling but was digested in mitochondria solubilized with Triton X-100, indicating that the 39-kD form is Tam41 imported into the matrix. We also confirmed that the 41-kD Tam41 precursor is converted to the 39-kD mature form in vivo and that the N-terminal 34 residues of the Tam41 precursor are sufficient to direct nonmitochondrial protein to mitochondria in vitro (Fig. S1, C and D). A search of the database revealed that Tam41 has homologues in a wide range of eukaryotic organisms from yeast to human (Fig. S2, available at ). We then analyzed the properties of endogenous Tam41 in mitochondrial association by immunoblotting with anti-Tam41 antibodies (). Tam41 was inaccessible to protease added to intact mitochondria (−SW) or to mitoplasts (+SW) but was accessible to protease added to mitochondria solubilized with Triton X-100. This behavior resembled that of Tim44, an inner membrane protein exposing a domain to the matrix, but is different from those of Tom70, a surface-exposed outer membrane protein, or Tim23, an inner membrane protein with the IMS domain (). Tam41, like the peripheral membrane proteins Mdj1p and Tim44, was extracted by alkaline (NaCO) treatment of mitochondria but was not released to the supernatant by sonication followed by ultracentrifugation (). These results indicate that Tam41 is a peripheral membrane protein of the inner membrane facing the matrix. To assess the function of Tam41, we analyzed the effects of Tam41 depletion on mitochondrial protein import in vivo and in vitro. Total lysates were prepared from wild-type (WT) or cells grown in YPD at 23 or 37°C and subjected to immunoblotting for various mitochondrial proteins (). The amounts of the components of the TOM40 complex (Tom40), the TIM23 complex (Tim23, -17, -50, and -44), and the TIM22 complex (Tim22), as well as MMC proteins (Pam18 and -16), were not affected by depletion of Tam41 at 23 or 37°C. However, we could observe accumulation of uncleaved precursor forms of mtHsp60, Mdj1p, and Zim17/Tim15 for cells, and the accumulation was more pronounced at 37 than at 23°C. These results suggest the role of Tam41 in mitochondrial protein import in vivo. Next, we tested the ability of mitochondria isolated from cells to import various radiolabeled precursor proteins in vitro. Mitochondria isolated from cells did not exhibit a decrease in ΔΨ, which is essential for protein import via the TIM23 and -22 complexes (unpublished data). The steady-state levels of translocator and MMC proteins were similar between WT and mitochondria (Fig. S1 E). We first analyzed the import of matrix-targeting precursor proteins with an N-terminal presequence, which depends on both the TIM23 complex and the import motor machinery comprising mtHsp70 and MMC proteins. The import rates of the precursors to mtHsp60 and Zim17/Tim15 into mitochondria decreased as compared with those into WT mitochondria (). On the other hand, import of presequence-less polytopic inner membrane proteins, ADP/ATP carrier, phosphate carrier, and Tim23, via the TIM22 complex was not retarded in the absence of Tam41 (). These results suggest that Tam41 is involved in the protein import pathway via the TIM23 complex. We then tested whether Tam41 facing the matrix plays a role, like Zim17/Tim15 and other MMC proteins (), in the import motor function of mtHsp70. Presequence-containing precursors that possess additional stop-transfer sorting signals are inserted into the inner membrane by the TIM23 complex and ΔΨ but do not require the motor function of mtHsp70. precursor and mouse DHFR, and the precursor to cytochrome do not require the ATP-dependent motor function of mtHsp70. Indeed, import of pb-DHFR into the IMS was not affected by depletion of Tam41 (). precursor and pb-DHFR was strongly retarded by the Tam41 depletion (). These results suggest that the observed defects of the import via the TIM23 complex in the absence of Tam41 cannot be simply ascribed to the defects in the motor functions of mtHsp70 and MMC proteins. How does the depletion of Tam41 in the matrix affect the protein import via the TIM23 complex? We compared the TIM23 complex structure by glycerol density gradient centrifugation between mitochondria with and without Tam41 (). Interestingly, although Tim23 and -17, the channel subunits of the TIM23 complex, were recovered in the fractions corresponding to ∼250 kD in the presence of Tam41, they shifted to the ∼200 kD fractions in the absence of Tam41. Tim44 associating with Tim23 and -17 also shifted to fractions of smaller molecular mass in the absence of Tam41, although the molecular mass of its peak was slightly smaller than that for Tim23 and -17 (∼200 kD), probably reflecting the dynamic association of Tim44 with -23 and -17. On the other hand, Tim50, the peripheral subunits of the TIM23 complex, exhibited a peak at 100–150 kD and a shoulder at ∼250 kD, and they were not affected by the Tam41 depletion. Tim54 in the TIM22 complex and Tom40 in the TOM40 complex were not affected by depletion of Tam41. Although the above results suggest the possibility that Tam41 constitutes the TIM23 core complex with Tim23 and -17, Tam41 was recovered in the fractions corresponding to ∼140 kD after glycerol density gradient centrifugation. When we analyzed the subunit interactions involving Tam41 in the solubilized TIM23 complex containing the FLAG-tagged version of Tim23 by coimmunoprecipitation with the anti-FLAG antibody or anti-Tam41 antibodies, we failed to observe interactions between Tam41 and Tim23, although Tim17, Tim21, Tim44, Tim50, Pam18, and Pam16 were coimmunoprecipitated with Tim23 but not with Tam41 (). Therefore, Tam41 is not stably associated with the TIM23 complex, ruling out a trivial possibility that the shifts of Tim23, -17, and -44 toward fractions of lower molecular weight are due to the lack of Tam41 in the TIM23 complex upon depletion of Tam41. After glycerol density gradient centrifugation, Pam18 and -16, peripheral MMC subunits of the TIM23 complex, were recovered in the fractions corresponding to ∼250 kD in WT mitochondria. Interestingly, however, the amounts of Pam18 and -16 in the ∼250-kD fractions decreased but instead shifted to the bottom upon depletion of Tam41. This suggests that Pam18 and -16 form aggregates in the absence of Tam41. Indeed, when solubilized WT mitochondria were subjected to immunoprecipitation with the anti-FLAG antibody, the amounts of Pam18 and -16 coimmunoprecipitated with the TIM23 complex via FLAG-tagged Tim23 were significantly reduced (). These results suggest that Tim23, -17, and -44 shifted to the smaller molecular weight fractions partly because Pam18 and -16 were not properly assembled into the TIM23 complex in the absence of Tam41. When we analyzed the aggregate formation of imported Pam18 and -16 by centrifugation of the mitochondria solubilized with digitonin, the amount of imported Pam16 in the pellet was larger for mitochondria than for WT mitochondria (), indicating that Pam16 forms aggregates inside the mitochondria in the absence of Tam41. Therefore, Tam41 is apparently involved in the proper assembly of Pam16 into the TIM23 complex. Recently, it was proposed that Tim21 and the Pam18–Pam16–Tim44 set are mutually exclusive as constituents of the TIM23 complex; i.e., the TIM23 complex is present in two different forms, one with Tim21 but not with Pam18–Pam16–Tim44, and the other with Pam18–Pam16–Tim44 but not with Tim21 (). However, this model is not consistent with our observation that Tim21 at ∼350 kD also shifted to ∼300 kD upon depletion of Tam41, which leads to dissociation of Pam18 and -16 from the TIM23 complex (). We thus examined whether Tim21 is in association with Pam18 and -16 by immunoprecipitation of solubilized mitochondria containing the FLAG-tagged Tim21 or Pam16 with anti-FLAG antibodies, and we indeed found that Tim21 was coimmunoprecipitated with Pam16FLAG (, lane 9) and Tim44, Pam18, and Pam16 with Tim21FLAG (, lane 6). Finally, we analyzed the assembly states of the TIM23 complex in details in the presence or absence of Tam41 by blue-native PAGE (BN-PAGE) separation of digitonin-solubilized mitochondria. Tim23 and -17 were found mainly in a 100-kD core complex (, Tim23–17), but additionally in larger TIM23 complexes containing Tim21 (, Tim23–17–21) or Tim50 and -21 (, Tim23–17–21–50). Constituents of the different forms of the TIM23 complex were identified by immunoblotting to detect the FLAG-tagged Tim23, -17, -21, and -50. Imported radiolabeled Tim23 was found in similar forms of the TIM23 complexes, and antibodies against Tim17FLAG, Tim21, and Tim50 caused shifts of all or a part of those TIM23 complexes (). All of the TIM23 complexes became smaller and the amount of the core Tim23–Tim17 complex decreased in mitochondria lacking Tam41 (, lane 2), whereas the major fractions of Tim50 and -44 were not affected by Tam41 depletion (, lanes 7, 8, and 13–16). Pam16 and -18 were found in the Pam16–Pam18 complex, whose amount was also reduced in the mitochondria without Tam41 (, lanes 9–12). These results suggest that depletion of Tam41 leads to partial dissociation of the TIM23 complex as well as destabilization of the Pam16–Pam18 complex. In conclusion, we identified Tam41 in yeast mitochondria, whose depletion affects protein transport via the TIM23 complex to the matrix and to the inner membrane. Although most TIM23 pathway–associating proteins that reside in the matrix are members of MMC proteins and cooperate with the motor protein mtHsp70, Tam41 facing the matrix does not promote the motor function of mtHsp70 directly but is instead involved in the maintenance of the integrity of the TIM23 complex. In particular, Tam41 affects integration of Pam18 and -16 into the TIM23 complex, and the stability of the TIM23 complex consisting of Tim23, -17, -21, and -50. Elucidation of the detailed mechanisms for Tam41 to maintain the functional integrity of the TIM23 complex from the matrix side of the inner membrane will enhance our understanding of the dynamic nature of the TIM23 complex responsible for its function. During the process of revision of the manuscript, reported online a study on the same protein with a similar conclusion. Yeast strains used in this study are listed in Table S1 (available at ). Cells were grown in YPGal (1% yeast extract, 2% polypeptone, and 2% galactose), YPD (1% yeast extract, 2% polypeptone, and 2% glucose), SD (0.67% yeast nitrogen base without amino acids and 2% glucose), SCD (0.67% yeast nitrogen base without amino acids, 0.5% casamino acid, and 2% glucose), or SCLac (0.67% yeast nitrogen base without amino acids, 0.5% casamino acid, and 2% lactate) media with appropriate supplements. was cloned from the yeast genomic DNA by PCR using primers 5′-GCCGGCTCGAGTGCACTCATAATGCTACTCG-3′ and 5′-CCGCCGGATCCACTGGAACGTATGATCCCCC-3′. The amplified DNA fragment was digested with BamHI and XhoI and introduced into the BamHI and the XhoI sites of pRS316 to produce pRS316-Tam41. The plasmid used for in vitro translation of Tam41 was constructed as follows. A DNA fragment corresponding to the gene was amplified from pRS316-Tam41 by PCR using primers 5′-CCGGATCCATAATTTGAATTAATAGGAGCTGCTTT-3′ and 5′-GCGCGTCGACGATACACTAGCTTCTCCTCATCGAT-3′. The amplified DNA fragment was digested with BamHI and SalI and introduced into the BamHI and the SalI sites of pGEM-4Z (Promega) to produce pGEM-Tam41. A DNA fragment corresponding to full-length ORF of was amplified from the yeast genomic DNA by PCR using primers 5′-CCGCCGGATCCATGTTACGAGTTTCTGAAAA-3′ and 5′-GCCGGCTCGAGTGCTTCTCCTCATCGATTTTA-3′. The amplified DNA fragment was digested with BamHI and XhoI and introduced into the BamHI and the XhoI sites of pET-21a (Novagen) to produce pET-21a-Tam41. The fusion protein expressed in the strain BL21(DE3)/pLysS was recovered in the inclusion body fraction. The protein was solubilized in 8 M urea, 20 mM Tris-HCl, pH 7.5, and 150 mM NaCl and was purified by Ni-NTA agarose chromatography (QIAGEN). The purified protein was used to immunize rabbits with complete Freund's adjuvant followed by four booster immunizations with incomplete Freund's adjuvant. Mitochondria were isolated from D273-10B grown in lactate medium at 30°C and W303-1A and strains grown in YPD at 23°C. Radiolabeled precursor proteins were synthesized with rabbit reticulocyte lysate by coupled transcription/translation in the presence of [S]methionine. Mitochondria isolated from W303-1A and cells were incubated with radiolabeled precursor proteins in import buffer (250 mM sucrose, 10 mM MOPS-KOH, pH 7.2, 80 mM KCl, 2 mM KPi, 2 mM methionine, 5 mM DTT, 5 mM MgCl, 2 mM ATP, 2 mM NADH, and 1% BSA) at 25°C. The mitochondria were isolated by centrifugation, and proteins were analyzed by SDS-PAGE and radioimaging. Treatment of mitochondria with PK, sodium carbonate, and Triton X-100 and preparation of mitoplasts were performed as described previously (). Mitochondria were solubilized at 2 mg/ml in 1% digitonin buffer (1% digitonin, 20 mM Tris-HCl, pH 7.4, 50 mM NaCl, 0.1 mM EDTA, 10% glycerol, and 1 mM PMSF) for 20 min on ice and were centrifuged at 20,000 for 15 min. The supernatant was layered onto linear glycerol gradient (20–40%) in 20 mM Tris-HCl, pH 7.4, 50 mM NaCl, 50 mM 6-aminohexanoic acid, 0.1 mM EDTA, 0.1% digitonin, and complete protease inhibitor cocktail (Roche) and centrifuged at 166,000 for 15 h. Mitochondria were solubilized at 2 mg/ml in 1% digitonin buffer. After centrifugation at 20,000 for 15 min, the solubilized proteins were added to the anti-FLAG antibody bound to agarose (Sigma-Aldrich) or anti-Tam41 antibodies bound to protein A–Sepharose (GE Healthcare). The samples were gently rotated for 2 h at 4°C. The resins were washed with wash buffer (0.2% digitonin, 20 mM Tris-HCl, pH 7.4, 50 mM NaCl, 0.1 mM EDTA, 10% glycerol, and protease inhibitor cocktail [Sigma-Aldrich]) three times before bound proteins were eluted with 100 mM glycine-HCl, pH 2.5, or SDS-PAGE sample buffer. The eluted proteins were analyzed by SDS-PAGE and immunoblotting. Mitochondria containing FLAG-tagged Tim23, Tim17, Tim50, or Pam16 were used for BN-PAGE analyses with or without Tam41. The mitochondria (2 mg of protein per milliliter) were solubilized with 1% digitonin buffer for 20 min at 4°C, cleared by centrifugation at 20,000 for 15 min, diluted 20-fold in 0.2% (wt/vol) CBB G-250 and 5 mM 6-aminohexanoic acid, and subjected to BN-PAGE (). Fig. S1 shows characterization of Tam41. Fig. S2 compares amino acid sequences of Tam41 from various organisms. Table S1 shows the yeast strains used in this study. Online supplemental material is available at .

Effective muscle contractile performance is contingent upon the maintenance of Ca homeostasis and signaling, which requires that intracellular Ca ([Ca]) be readily available for release. In skeletal muscle, excitation–contraction (E–C) coupling is primarily mediated by conformational coupling between voltage sensors of the sarcolemmal membrane and ryanodine receptor (RyR) Ca release channel of the sarcoplasmic reticulum (SR; ; ; ). A secondary process, called Ca-induced Ca release (CICR), amplifies [Ca] release in skeletal muscle, particularly under stress conditions in muscle fatigue and dystrophy (; ; ). Aging effects on muscle function have been associated with muscle fiber denervation, loss of motor units, and motor unit remodeling. Because functional alterations occur before significant muscle wasting becomes evident, changes in the E–C coupling machinery and [Ca] homeostasis may act as causative factors for, or adaptive responses to, muscle aging (; ; ). We show that stress-induced Ca sparks, which are the elemental events of CICR in striated muscles (; ), are severely compromised in aged skeletal muscle. In addition, we find that muscle aging is associated with the development of a segregated SR Ca pool that uncouples from the normal E–C coupling machinery. We present evidence to suggest that mitsugumin-29 (MG29) may act as a sentinel against the effects of age on skeletal muscle Ca homeostasis. To delineate the contribution of Ca sparks to the aging phenotype in skeletal muscle, intact flexor digitorum brevis (FDB) fibers isolated from young (2–4 mo) and aged (26–27 mo) mice were treated with osmotic shock to induce Ca sparks. Exposure of the muscle fiber to a hypotonic solution leads to swelling of the fiber. Upon return to isotonic solution, the recovery of cell volume to normal is accompanied by a robust, peripherally localized Ca spark response (; ). Young muscle fibers display a dynamic Ca spark response to repeated stress cycles, with each round of osmotic shock generating Ca sparks that continue for several minutes (). This Ca spark response is located in the periphery of both young and aged muscle fibers. The dynamic nature of this Ca spark response is significantly reduced in aged muscle (). Relative to young muscle (), aged muscle appears to contain a diminished capacity for the generation of dynamic Ca sparks with repeated osmotic stresses (). The diminished Ca spark response in aged muscle could result from changes in resting [Ca] levels or altered Ca storage inside the SR. We measured [Ca] levels and SR Ca storage in FDB muscle fibers obtained from young and aged mice. As shown in , the resting [Ca] levels appear to be similar between young and aged skeletal muscle fibers, whereas the caffeine/ryanodine-mobilized SR Ca pool is significantly less in aged skeletal muscle. Thus, the reduced caffeine/ryanodine-mobilized SR Ca store may represent one potential factor for the compromised Ca spark signaling associated with muscle aging. Additional factors that may contribute to this defective Ca spark signaling include changes in membrane ultrastructure or altered expression of Ca regulatory proteins in skeletal muscle. We conducted a survey of triad junction proteins and found that the expression level of MG29 (), a synaptophysin-related membrane protein, is significantly down-regulated during muscle aging (). To determine the extent that decreased MG29 levels contribute to age-related alterations in muscle Ca homeostasis, muscle fibers obtained from young (3–5 mo) mice () were stressed by osmotic shock. As with aged () muscle, there is an initial Ca spark response to the first osmotic shock and subsequent osmotic shocks produce little to no Ca spark response in young muscle fibers (). Using fura-2 Ca measurements, we found that the resting [Ca] level and SR Ca storage are similar between aged and young muscle fibers (). These results point to a role for MG29 in maintaining normal Ca homeostasis that is lost with its diminished expression during aging. Our previous studies have shown that the mice display contractile alterations and muscle atrophy at ages of 6 mo or younger (; ) that resemble the atrophic phenotype of aged mice. Electron microscopy studies reveal similar ultrastructural alterations to triad junction membrane structures of aged and young skeletal muscle. Although organized alignment of SR and transverse-tubule (TT) membranes is present in young muscle, fragmented SR is frequently observed in slow (soleus) and fast (extensor digitorum longus [EDL]) twitch muscles from both young and aged skeletal muscles (). Aged soleus muscle also displays swelling of the TT that is very similar to that seen in muscle (). The development of these defects appears to be progressive during aging, as EDL fibers display a continuum of damage ranging from minor SR fragmentation () to formation of large aggregations of SR ( and Table S1, available at ; ; ). We suspect that aggregation of SR may result from the accumulation of subtle defects in Ca signaling and membrane recycling, leading to progressive sarcopenia (). The similar structural alterations seen in aged and young muscle suggests that further defects in Ca homeostasis beyond their Ca spark response should be present. Previous studies by revealed that repetitive Ca release using KCl stimulation on intact muscle in the absence of extracellular Ca ([Ca]) did not lead to complete depletion of the SR Ca pool. We now present evidence to show the presence of a segregated SR Ca pool that uncouples from the normal voltage-induced Ca release (VICR) machinery in young muscle and aged muscle (). When isolated, intact muscle fibers are exposed to 0 [Ca] for 90 min, intermittent nonfatigue voltage stimulation leads to passive depletion of the SR Ca pool, followed by minimal caffeine-induced Ca release in young muscle, whereas this response is larger in aged and young muscle (). In the absence of [Ca], fatiguing stimulation using repetitive VICR leads to rapid depletion of the voltage-sensitive pool of [Ca] releasable, revealing striking differences among the three muscle preparations in their subsequent response to caffeine treatment. As shown in , after this fatiguing stimulation, caffeine induces significantly larger Ca release in young and aged muscles compared with the young wt muscle. Data from multiple experiments are summarized in . These findings suggest the presence of a VICR-uncoupled, RyR-sensitive Ca pool in aged and young muscles; however, the relative sizes of these pools in soleus and EDL muscle fibers are difficult to evaluate because of increased sensitivity of the contractile apparatus of the soleus muscle to caffeine and Ca (; ; ; ). The segregated SR Ca pool in aged and young muscle is not likely to result from major changes in the VICR machinery itself, as the initial VICR responses in young and aged and young muscle are generally similar (). The maximal specific tetanic force produced by aged EDL muscle (250 ± 15 kN/m) is lower than that produced by young muscle (300 ± 18 kN/m), which is consistent with previous studies (; ). The maximal specific tetanic force in young EDL muscles is 262 ± 17 kN/m, which is similar to that measured in aged muscle. To test the possibility that changes in contractile machinery during muscle aging might contribute to our measurement of Ca release, the force versus pCa relationship was examined in all three types of muscle preparations. As shown in Table S2 (available at ), the isometric contractile properties of Triton X-100–skinned muscle fibers are similar in young and aged , as well as young , mice. Therefore, under these conditions, the force output is an authentic measurement of Ca release from the SR, and a segregated SR Ca pool must exist in aged and young muscle to account for the elevated, caffeine-induced Ca release. We have established that aged muscle fibers have a disrupted Ca spark response and a segregated Ca store that cannot be mobilized by VICR, which are associated with ultrastructural disruption of triad junctions and the SR network (). One possible explanation for the development of a segregated SR Ca pool is that subtle disruption of SR and TT alignment at the triad junction could result in uncoupling of RyR1 and DHPR, which could also lead to the compromised Ca spark signaling observed in aged skeletal muscle. An inhibitory role for DHPR on RyR1 function has been proposed by other investigators (; ; ). If the Ca spark response associated with membrane deformation and the segregation of [Ca] release was solely caused by disruption of the inhibitory effects of DHPR on RyR1 function, one would expect to see an elevated Ca spark response in aged skeletal muscle, as it has been established that DHPR expression and the ratio of DHPR to RyR1 is decreased in aged skeletal muscle (). Therefore, the reduced Ca spark response observed in aged skeletal muscle suggests that changes in other cellular factors, such as MG29 expression, may play a role in regulation of Ca signaling in skeletal muscle at different developmental stages. MG29 contains a high degree of homology with synaptophysin, a protein thought to be involved with membrane fusion during exocytosis (). Decreased MG29 expression may lead to improper lipid membrane formation or fusion, altering the dynamic process of membrane recycling and SR network formation. In aging or young mice, the lack of this synaptophysin family member would suppress the efficient maintenance of triad junction structure, while also generating a fragmented SR network. Recent results from our laboratory have begun to shed light on the physiological function of MG29. We have demonstrated that MG29 increases sensitivity of the RyR channel to CICR when expressed in a heterologous cell system and when reconstituted with RyR in lipid bilayer single-channel studies (). The lack of MG29 in aged and muscle should decrease the sensitivity of RyR to CICR, in agreement with our current results. Segregation of Ca pools in aged muscle may have a physiological role in maintaining muscle integrity in the face of decreasing homeostatic capabilities. The resulting dampened Ca mobilization in aged muscle may be a compensatory mechanism that protects aged fibers from Ca-induced injury. It is also possible that the presence of a segregated Ca reserve isolated from VICR responses contributes to cellular stress and decreased homeostatic capacity. Although the mechanism of active shuttling of Ca from the VICR-responsive to the VICR-nonresponsive pool is not known, selective regulation of this Ca shuttling to modulate the VICR-responsive pool would allow for the enhancement of aging skeletal muscle performance and/or protect skeletal muscle during aging. The mouse represents a model system in which these mechanisms can be examined and these hypotheses can be tested. Studies were conducted with C57Bl6/J male mice (Aged Rodent Colony, maintained by the National Institute on Aging), which were either 2–4 or 26–27 mo old. Male mice (2–5 mo old), and animals of the same genetic background (129Sv/J backcrossed to C57Bl6/J) were maintained in local facilities and handled in a manner approved by local regulations. FDB fibers were isolated by enzymatic disassociation in 0.2% type I collagenase (Sigma-Aldrich) for 55 min at 37°C and loaded with 10 μM Fluo-4–AM for 60 min at room temperature. Mean FDB fiber size was 1 mm × 20 μM. Measurements of Ca release were performed on a confocal microscope (Radiance 2100; Bio-Rad Laboratories) equipped with an argon laser (488 nm) and a 60×, 1.3 NA, oil immersion objective. For Ca spark measurements, fibers were perfused with a 170-mosM hypotonic solution containing (in mM) 70 NaCl, 5 KCl, 10 Hepes, 2.5 CaCl, 2 MgCl pH 7.2, for 60–180 s to induce swelling before perfusion was switched back to the initial Tyrode solution (in mM) 140 NaCl, 5 KCl, 10 Hepes, 2.5 CaCl, 2 MgCl, pH 7.2, with an osmolarity of 290 mosM as measured by a Micro Osmometer 3300 (Advanced Instruments). Image analysis was performed using custom routines on IDL software (Research Systems, Inc.; ; ). For determination of resting cytosolic Ca levels and total SR Ca store, individual FDB fibers were loaded with 10 μM fura-2 AM for 45 min at room temperature in Tyrode solution. 20 μM N-benzyl-p-toluene sulphonamide, a myosin II inhibitor, was applied for 15 min to prevent motion artifact from muscle contraction (; ). Fibers were also embedded into silicone grease to maintain their position in the culture dish (). The ratio of fura-2 fluorescence at excitation wavelength of 350 and 380 nm was measured on a PTI spectrofluorometer (Photon Technology International) to assess the resting [Ca] level. The SR Ca store was measured by addition of 20 mM caffeine plus 5 μM ryanodine in the presence of 0 [Ca]. Electron microscopy studies were performed following our previously published protocols (). In brief, skeletal muscles were fixed in 3% paraformaldehyde, 2.5% glutaraldehyde, and 0.1 M cacodylate buffer, pH 7.4, and later postfixed in 1% OsO4 and 0.1 M cacodylate buffer, pH 7.4. Microthin sections were double stained with uranyl acetate and lead citrate. These sections were examined under a transmission electron microscope (JEM-1010; JEOL). Intact EDL and soleus muscles were dissected from mice and maintained in modified Ringer's solution containing the following (in mM): 142 NaCl, 4.0 KCl, 2.5 CaCl, 2.0 MgCl, 10 glucose, and 10 Hepes, pH 7.4 ± 0.1, continuously bubbled with 100% O. EDL muscles had a mean length of 12 mm and a mean mass of 80 mg, whereas soleus muscles had a mean length of 10 mm and a mean mass of 10 mg. Muscles were mounted vertically on a glass-stimulating apparatus (Radnoti) with platinum electrodes and attached to a movable isometric force transducer and to a stationary anchor, which allowed muscles to be stretched until both maximal forces for a given frequency and the frequency producing T were obtained. After T was determined, the intact muscles were allowed to equilibrate for 20 min in the Ringer's solution. During equilibration, muscle strips were stimulated with ∼100–120 Hz (EDL) or ∼60–80 Hz (SOL), 330 mA, 500 ms electrical pulse–trains administered with a periodicity of 1 min to generate T. After equilibration, the muscles from the passive-depletion group were washed five times in Ringer's solution with the same composition as described in the previous section, except that no CaCl was added while 0.1 mM EGTA was added, to create a nominal 0 [Ca] solution. Muscles were stimulated with one T every minute in 0 [Ca] solution until force declined to nondetectable levels; the muscles were then exposed to 30 mM caffeine. Force produced in response to caffeine application was recorded until a stable plateau was obtained. After the passive depletion protocol, muscles were subjected to extensive washes in normal Ringer's solution containing 2.5 Ca and then electrically stimulated until force returned to initial equilibration values. In 80% of our preparations, this was achieved. After forces were stable and comparable to the initial levels before the onset of the passive depletion, muscles were returned to the 0 [Ca] solution for 5 min and subsequently subjected to a 1–5 min fatiguing protocol consisting of the same stimulatory pattern administered at a 1-s periodicity (i.e., 50% duty cycle). In between fatigue runs, muscles were washed in 2.5 Ca solution and force was allowed to recover to prefatigue levels before the onset of the next fatigue run. At the end of each fatiguing protocol, muscles were treated with 30 mM caffeine and maximal response to caffeine was recorded. Caffeine was mixed in a small volume of the Ringer's solution and added to the chambers to produce a final concentration of 30 mM in the bathing chamber. Whenever possible, paired experiments were performed with young wild-type animals and aged or young animals. Experiments were also conducted with fibers only exposed to passive depletion or fatigue in 0 [Ca] to confirm that effects of each treatment can be observed independently. The integrity of the fiber contractile apparatus and Ca-handling machinery was tested at the conclusion of the protocol by exposure to 100 mM KCl. Only fibers with at least 85% of T were included for statistical analysis. All force data were normalized to the last tetanic contraction at the end of the equilibration period and just before the start of the fatiguing protocol (this T = 100%). Absolute force, normalized per cross sectional area (i.e., in kN/m) was determined at the end of the equilibration period by the following relationship: Force (in Kg) = ( of force) × (muscle length in cm) × 1.06/muscle weight (), where 1.06 represents the density of the muscle strips. Triton X-100–skinned muscle fiber experiments followed the protocols as previously described (; ). All statistical analysis in this study was conducted using ANOVA, and data is presented as the mean ± SEM. Table S1 describes the parallel disruption of triad junctions in aged and young skeletal muscle. Table S2 is an assessment of muscle contractile function in young , aged , and young muscle. Online supplemental material is available at .

Cell migration involves membrane polarization and cytoskeletal dynamics, both of which are regulated by Rho family GTPases (). Among these molecules, Rac is crucial for generating the actin-rich lamellipodial protrusion, which is a principal part of the driving force for movement. Rac is composed of three isoforms, Rac1, Rac2, and Rac3. Rac1 is ubiquitously expressed and Rac3 is highly expressed in the brain, whereas Rac2 expression is largely restricted to hematopoietic cells. The role of Rac in neutrophil functions has been extensively analyzed with knockout mice lacking Rac1 and/or Rac2, and in a human patient with a point mutation in the conserved GTP-binding domain of Rac2. These studies clearly indicate that Rac2 is a major Rac isoform that regulates chemoattractant-induced neutrophil functions, such as chemotaxis and superoxide production (; ; ). However, the defects in chemotaxis and superoxide production of Rac2-deficient neutrophils are significantly augmented by additional loss of Rac1 (). In addition, it has been reported that Rac1 deficiency alone results in an inability of neutrophils to detect and to orient in a chemotactic gradient (), suggesting that Rac1 is also involved in the chemotactic response of murine neutrophils. Like other Rho family GTPases, Rac cycles between GDP-bound inactive and GTP-bound active states. Because the GTP loading is mediated by guanine nucleotide exchange factors (GEFs), significant efforts have been made to identify a Rac GEF that functions downstream of chemoattractant receptors in neutrophils. P-Rex1 is a phosphatidylinositol 3,4,5-triphosphate (PIP)– and Gβγ-regulated Rac GEF that has been purified from neutrophils (). It had been thought that P-Rex1 would be a major Rac activator that regulates neutrophil chemotaxis. Unexpectedly, however, it was recently found that neutrophil chemotaxis is only mildly affected in P-Rex1–deficient (P-Rex1) neutrophils (; ). In addition, chemoattractant-induced Rac activation was also reported to occur normally in neutrophils lacking both Vav1 and Vav3 (). Thus, the Rac activator that is critical for neutrophil chemotaxis remains to be determined. CDM family proteins ( CED-5, mammalian DOCK180, and Myoblast City) are known to regulate the actin cytoskeleton by functioning upstream of Rac (). DOCK2 is a new member of the CDM family proteins, and is expressed predominantly in hematopoietic cells (). Although DOCK2 does not contain the Dbl homology domain and the pleckstrin homology (PH) domain that are typically found in GEFs, DOCK2 binds to nucleotide-free Rac and catalyzes GTP loading through its Docker (also known as DHR-2) domain (; ). We had previously reported that DOCK2 regulates lymphocyte migration and immunological synapse formation through Rac activation (; ). However, the role of DOCK2 in neutrophils remains unknown, an issue that was addressed in this study. To examine whether DOCK2 functions in neutrophils, we first compared Rac activation between C57BL/6 (B6) and DOCK2 mice. When bone marrow (BM) neutrophils from B6 mice were stimulated with -formyl-methionyl-leucyl-phenylalanine (fMLP), activated Rac1, Rac2, and Cdc42 were readily detected at 5 s after stimulation (). This activation was rapid and transient, and the levels of the GTP-bound Rac1, Rac2, and Cdc42 were substantially decreased at 15 s. In DOCK2 neutrophils, Cdc42 was activated to the same extent and at the same kinetics as in B6 neutrophils (). However, fMLP-induced activation of both Rac1 and Rac2 were reduced by 70% in DOCK2 neutrophils at 5 s after stimulation (). Similar defects were observed when DOCK2 neutrophils were stimulated with PMA (Fig. S1, available at ). These results indicate that DOCK2 plays a major role in chemoattractant- and PMA-induced Rac activation in neutrophils. We next examined how DOCK2 deficiency affects neutrophil functions. In the transwell chemotaxis assay, 8–11% of B6 BM neutrophils migrated into the lower chamber in response to 8 μM fMLP (, left). In the case of DOCK2 neutrophils, however, the percentage of migrating cells was <2% under the same conditions (, left). A similar defect was observed when complement factor 5a (C5a) was used as a chemoattractant (, right). When B6 BM neutrophils were stimulated with fMLP, they produced superoxides in a superoxide dismutase (SOD)–inhibitable manner (, left). However, the total amount of superoxides produced by DOCK2 neutrophils was reduced to <20% of the wild-type level (, left). In addition, unlike P-Rex1 neutrophils (; ), DOCK2 neutrophils exhibited a defect in PMA-induced superoxide production (, right). These results indicate that DOCK2 is required in neutrophils for chemotaxis and superoxide production. To determine more precisely the role of DOCK2 in neutrophil chemotaxis, we analyzed BM neutrophils undergoing chemotaxis in a Zigmond chamber containing the fMLP gradient. At first glance, DOCK2 neutrophils were less motile than B6 neutrophils (Video 1, available at ). Detailed analysis revealed that the average speed of DOCK2 neutrophils was reduced to 55% of the wild-type level (). When the final location relative to the initial position was analyzed at 12.5 min, >85% of neutrophils had migrated toward the fMLP source, irrespective of DOCK2 expression (). However, although B6 neutrophils moved in relatively straight paths up the fMLP gradient, DOCK2 neutrophils often showed indecisive wandering behavior and took a turn in a short period (Video 2). Supporting this, DOCK2 deficiency was found to affect the directional change and the straightness during neutrophil chemotaxis (). To understand the mechanism by which DOCK2 regulates neutrophil chemotaxis, we first examined actin polymerization in BM neutrophils stimulated in suspension with fMLP. Although fMLP-induced actin polymerization is almost totally abolished in neutrophils lacking Rac1 and Rac2 (), such a drastic effect was not observed in DOCK2 neutrophils (). This might result from the difference in Cdc42 activation because, unlike DOCK2 deficiency, Rac2 deficiency has been reported to impair fMLP-induced Cdc42 activation in neutrophils (). Microscopic analysis revealed that both B6 and DOCK2 neutrophils uniformly accumulated F-actin at 15 s after stimulation (). However, although B6 neutrophils exhibited a localized accumulation of F-actin at 30 s, such polarization was scarcely found in DOCK2 neutrophils, although there was partial recovery of F-actin polarity at 60 s (). Consistent with this finding, the majority of DOCK2 neutrophils undergoing chemotaxis exhibited aberrant morphology with poorly focused distribution of F-actin (). These results indicate that DOCK2 is required for polarized accumulation of F-actin at the leading edge. In response to chemoattractants, neutrophils accumulate PIP, which is a lipid product of phosphatidylinositol 3-kinases (PI3Ks), at the leading edge (). Because this process requires Rac activation and actin polymerization (; ), we examined how DOCK2 deficiency affects PIP accumulation by transiently expressing a GFP-tagged PH domain of Akt (PH-Akt), which is a widely used probe for detecting the spatial distribution of PIP. When BM neutrophils were exposed to a uniform concentration of fMLP, PH-Akt rapidly translocated to the plasma membrane in both B6 and DOCK2 neutrophils (). However, while B6 neutrophils accumulated PH-Akt at the plasma membrane in a highly asymmetric manner at 30 and 60 s after stimulation, the membrane accumulation of PH-Akt was impaired in DOCK2 neutrophils at these time points (). Similar, but more profound, effects of DOCK2 deficiency were observed when BM neutrophils were stimulated with fMLP supplied by a micropipette (Fig. S2, available at ). Consistent with this defect, fMLP-induced Akt phosphorylation was reduced in DOCK2 neutrophils (). Thus far, Akt phosphorylation has been used as an indirect assay for PIP synthesis. Surprisingly, however, BM neutrophils from B6 and DOCK2 mice comparably generated PIP in response to fMLP and C5a ( and Fig. S3). These results suggest that DOCK2-mediated Rac activation regulates the persistent accumulation of PIP at the leading edge, independently of PI3K activities. We then asked whether DOCK2 itself is recruited to the leading edge in response to chemoattractants. For this purpose, we developed knock-in mice, where the gene encoding GFP is inserted immediately after the last exon of with a modification of the stop codon. When BM neutrophils of DOCK2-GFP mice were exposed to a uniform concentration of fMLP, DOCK2 rapidly translocated to the plasma membrane (, left). However, such translocation was totally abolished by pretreating the cells with a PI3K inhibitor wortmannin (, right). Although BM neutrophils polarized and accumulated DOCK2 at the leading edge in response to a point source of fMLP, the intracellular DOCK2 dynamics and morphological changes were severely impaired in cells pretreated with the PI3K inhibitors LY294002 and wortmannin ( and not depicted). These results indicate that DOCK2 translocates to the leading edge in a PI3K-dependent manner. To elucidate the mechanism for PI3K-dependent intracellular DOCK2 dynamics, we examined whether DOCK2 binds to PIP by expressing DOCK2 in human embryonic kidney (HEK) 293T cells with or without ELMO1, which is known to function cooperatively with DOCK2 in lymphocytes (). When cell extracts expressing DOCK2 alone were precipitated with PIP-coated beads, only a weak association was found (). However, the association of DOCK2 with PIP was significantly augmented by coexpression with ELMO1 (). DOCK2 was associated with PIP, but not PIP and PIP (). This association seemed to be specific because the PIP binding was inhibited when cell extracts were preincubated with PIP-containing liposomes (). Recently, DHR-1, which is an evolutionarily conserved domain among CDM family proteins, has been shown to play an important role in PIP binding of DOCK180 (). When a DOCK2 mutant lacking DHR-1 (DHR1del) was expressed in HEK293T cells with ELMO1, PIP binding was hardly detected (). This does not result from an inability to bind to ELMO1 because the deletion of DHR-1 does not affect DOCK2 binding to ELMO1 (unpublished data). Collectively, these results indicate that DOCK2 associates with PIP through DHR-1, and that this association is indirectly regulated by ELMO1. Several lines of evidence indicate that a PIP- and Rac-mediated positive-feedback loop is required for neutrophil chemotaxis by amplifying chemoattractant signals at the leading edge (; ; ). We have shown that whereas DOCK2 translocates to the leading edge in a PI3K-dependent manner, DOCK2 activates Rac and stabilizes the accumulation of PIP at the leading edge. These results suggest that DOCK2, in some sense, regulates leading edge formation by functioning in a PIP- and Rac-mediated feedback loop. However, it is unlikely that DOCK2-Rac signaling affects the catalytic activities of PI3Ks because BM neutrophils from B6 and DOCK2 mice comparably generated PIP in response to chemoattractants. Thus far, it has been reported that an inhibitor of actin polymerization reduces insulin-mediated Akt phosphorylation and PH-Akt translocation, without affecting PI3K activity, in cells other than neutrophils (). Similar findings have been reported in neutrophil-like HL-60 cells stimulated with fMLP (). On the other hand, we found that the persistence and polarized distribution of PIP is correlated with the localized F-actin assembly in fMLP-treated BM neutrophils. Although the precise meaning of this correlation remains unknown at this stage, the localized F-actin assembly might prevent PIP diffusion away from the leading edge and regulate activation of PIP-binding proteins by facilitating the effective protein–protein or protein–lipid interaction. DOCK2 mice were backcrossed with B6 mice for more than eight generations before use. For development of DOCK2-GFP mice, a targeting vector was designed to insert the gene encoding GFP and a floxed-neomycin–resistant cassette () immediately after the last exon of and was introduced into embryonic stem cells by electroporation. Correctly targeted embryonic stem clones were microinjected into B6 blastocysts, and the male chimeras obtained were crossed with female B6 mice. Heterozygous mutant mice were crossed with Ella Cre mice to remove , and -deleted heterozygous mutant mice were intercrossed to develop homozygous mutants expressing the DOCK2-GFP chimeric molecule. All experiments were done in accordance with the guidelines of the committee of Ethics of Animal Experiments, Faculty of Medical Sciences, Kyushu University. BM cells were isolated from femurs and tibias of mice and layered onto the discontinuous Percoll (Sigma-Aldrich) gradient. After centrifugation, cells at the 81/62% interface were recovered and incubated with anti-B220–coated magnetic beads (Miltenyi Biotec) to remove B cells. More than 90% of the remaining cells were Gr-1Mac-1 mature neutrophils. Transwell chemotaxis assays were performed as previously described (), using fMLP (Nacalai Tesque) or C5a (Sigma-Aldrich) as chemoattractants. After a 3-h incubation at 37°C, cells migrating to the lower chamber were collected and stained with anti–Gr-1 (BD Biosciences) and anti-F4/80 (Invitrogen) mAbs. The percentage of migrating neutrophils was calculated by dividing the number of Gr-1F4/80 cells in the lower chamber by that of the input cells. Analysis was performed with a FACSCalibur flow cytometer (Becton Dickinson). BM neutrophils (1 × 10) suspended in Hepes-buffered saline containing 0.03% BSA were stimulated at 37°C with 8 μM fMLP or 200 ng/ml PMA, and the reaction was terminated by addition of 50 μg/ml SOD. The chemiluminescence was counted with an enhancer-containing, luminol-based detection system (National Diagnostics) using a luminometer (Auto Lumat LB953; Berthold). BM neutrophils (5 × 10) were electroporated with 2.5 μg DNA construct by using the Human Monocyte Nucleofector kit (Amaxa Biosystems) following protocol Y-001, and then placed on a glass-bottom microwell dish. At 2 h after transfection, BM neutrophils were exposed to a uniform concentration of fMLP or stimulated on a heated stage (37°C) with a micropipette containing 10 μM fMLP. Images were taken from either a laser scanning confocal microscope (FLUOVIEW FV500; Olympus) or a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with differential interference contrast and fluorescence image capability. The intracellular DOCK2 dynamics were similarly analyzed using freshly isolated BM neutrophils from DOCK2-GFP mice with or without pretreatment of PI3K inhibitors. Neutrophils (6–7.5 × 10) were labeled with [P]orthophosphate in a labeling buffer (136 mM NaCl, 4.9 mM KCl, 5.5 mM glucose, 0.1% BSA [fatty acid–free], and 10 mM Hepes-NaOH, pH 7.4) for 1.5 h at 37°C. Cells were stimulated with 8 μM fMLP or 25 nM C5a for the specified times. Treatments were quenched by the addition of chloroform/methanol/8% HClO (5:10:4). After vigorous vortexing, chloroform/HClO (1:1) was added to isolate the organic phase, which was washed twice in chloroform-saturated 1% HClO before drying. Dried lipids were resolved in chloroform/methanol (95:5) and separated by thin-layer chromatography, or they were deacylated and analyzed by HPLC as previously described (), using a Partisphere SAX column (Whatman). For lipid-binding assays, HEK293T cells were transfected with the specified plasmid DNAs and suspended in 450 μl of lipid-binding buffer (20 mM Tris-HCl, 150 mM NaCl, and 1 mM EDTA, pH 7.5). The cells were passed 10 times through a G25 needle and sonicated on ice. After insoluble debris was removed by ultracentrifugation at 100,000 for 30 min at 4°C, a 20-μl slurry of PIP beads (Echelon Biosciences) and 1% NP-40 solution (at a final concentration 0.25%) were added to the tube and incubated for 3 h at 4°C under rotary agitation. The beads were washed with lipid-binding buffer supplemented with 0.25% NP-40, and the bound proteins were subjected to immunoblotting. For competitive inhibition with liposomes, cell extracts were incubated on ice with 200 μM phosphatidylserine/PIP liposomes for 2 h before addition of PIP beads and NP-40. Liposome was prepared as follows: phospholipids (Avanti Polar Lipids) dissolved in chloroform were mixed and dried down under nitrogen. The lipid film was then resuspended in lipid-binding buffer and sonicated on ice. Fig. S1 shows that DOCK2 deficiency impairs PMA-induced Rac activation. Fig. S2 shows that lamellipod formation and membrane accumulation of PH-Akt are impaired in DOCK2 neutrophils stimulated by a point source of fMLP. Fig. S3 shows that BM neutrophils from B6 and DOCK2 mice comparably generate PIP in response to C5a. Video 1 shows a low magnification image for B6 and DOCK2 neutrophils chemotaxing under the fMLP gradient. Video 2 shows the migratory behavior of B6 and DOCK2 neutrophils chemotaxing under the fMLP gradient. Online supplemental material is available at .

Sumoylation plays a key role in many cellular processes, including nuclear transport, signal transduction, transcriptional regulation, and maintenance of genome integrity (; ; ). Sumoylation is a process by which a small ubiquitin-like modifier (SUMO) protein (Smt3 in yeast) is conjugated to a target protein at a lysine residue. The consequences of sumoylation are substrate specific and can involve altering a substrate's interaction with other macromolecules (e.g., proteins and DNA) or blocking lysine residues on target proteins from being modified by other lysine-targeted modifications such as ubiquitin (; ). The addition of SUMO to a protein is reversible through the action of ubiquitin-like proteases (Ulps) that are able to cleave SUMO from its substrate (, ). Ulps are also required for the posttranslational maturation of SUMO through cleavage at its C terminus to reveal the diglycine motif used to form the isopeptide bond with its substrate. In the budding yeast , sumoylation is essential for viability and is required for proper chromosome segregation, with sumoylation-deficient cells arresting in G2/M with short spindles and replicated DNA (; , ). In mammalian cells, defects in sumoylation cause abnormal nuclear architecture, chromosome missegregation, and embryonic lethality (). The critical protein targets that lead to these phenotypes when SUMO conjugation is disrupted remain unknown. In , proteomic approaches have identified >400 potential sumoylation targets; thus, pinpointing the biologically relevant sumoylation events is a challenge (; ; ; ; ; ). The phenotypes associated with sumoylation-defective mutants point to proteins involved in chromosome segregation as critical targets of SUMO modification. Indeed, many proteins involved in chromosome segregation, including kinetochore proteins that mediate the attachment of chromosomes to microtubules, have been identified as sumoylation substrates (; ; ; ; ; ; ; ; ). However, the functional roles of sumoylation on these proteins are not understood. The kinetochore is a supramolecular protein complex that assembles on centromere DNA to perform two functions that are critical for genome stability. First, the kinetochore functions in a structural manner by facilitating the connection between sister chromatids and spindle microtubules, allowing microtubule-dependent forces to separate duplicated DNA during mitosis and meiosis (; ). In budding yeast, this connection is mediated by ∼60 proteins that bridge sister chromatids to the microtubule interface of the mitotic spindle (). Second, the kinetochore functions in a regulatory manner by providing a framework for the generation of signals used by cell cycle checkpoints to gauge the progress of mitosis, through which the timing of mitotic events are coordinated. The kinetochore does this by producing a wait anaphase signal until the proper bipolar attachment of sister chromatids to opposite spindle poles and the generation of tension has occurred (; ). The Ndc80 kinetochore complex is required to localize spindle checkpoint proteins to the kinetochore, which is critical for the generation of the wait anaphase signal (; ). Correct attachment and subsequent silencing of the checkpoint ultimately leads to the progression of cells into anaphase and separation of sister chromatids to daughter cells (). A subset of kinetochore proteins are relocalized to the mitotic spindle in anaphase and to the spindle midzone before cytokinesis, reflecting the localization of chromosome passenger proteins in higher eukaryotes (; ). In yeast, the set of chromosome passenger-like proteins includes the Aurora kinase Ipl1, the inhibitor of apoptosis (IAP) protein Bir1, and the inner kinetochore proteins Cep3 and Ndc10 (; ; ; ; ; ). The localization of these proteins to the mitotic spindle is thought to be important for regulating mitotic spindle dynamics and for cytokinesis (; ). Indeed, temperature-sensitive alleles of or cause cytokinesis defects at the restrictive temperature (; ). However, mutations in Bir1 that result in the loss of Ndc10 from the mitotic spindle cause defects in proper spindle elongation, not in cytokinesis (). To gain insight into the function of Ndc10 on the mitotic spindle, we began by identifying protein-interacting partners of Ndc10 with the goal of discovering proteins that are required for Ndc10's spindle localization. Using a genome-wide two-hybrid screen, we identified multiple interactions between Ndc10 and the sumoylation machinery of budding yeast, and subsequent analysis demonstrated that Ndc10 is a target for sumoylation in vivo, as are other kinetochore proteins (Bir1, Cep3, and Ndc80). We also found that sumoylation of these proteins is differentially regulated in response to checkpoint activation, suggesting that sumoylation has distinct roles in modulating the function of these kinetochore proteins. Importantly, lysine residues required for Ndc10's sumoylation are necessary for Ndc10's proper localization to the mitotic spindle, suggesting that sumoylation plays a direct role in facilitating Ndc10's interaction with the spindle apparatus. As is the case when Bir1 is mutated, the mislocalization of Ndc10 is not associated with cytokinesis defects, suggesting that the spindle-bound form of Ndc10 is not responsible for Ndc10's cytokinesis-related functions. Furthermore, the loss of Ndc10's mitotic spindle association results in anaphase spindles of abnormal length, highlighting a role for Ndc10 in controlling mitotic spindle dynamics. In two independent genome-wide two-hybrid screens using Ndc10 as bait with the Gal4 DNA-binding domain fused to either the N or C terminus, 10 proteins were identified as putative Ndc10 protein interactors (). Three of these interactions occurred with only one of the baits, suggesting that the presence of the Gal4 DNA-binding domain on the N or C terminus may be interfering with specific protein–protein interactions, as has been previously reported (). Of these 10 interacting proteins, Bir1 and Ubc9 were previously identified in two-hybrid screens with Ndc10 (; ); Bir1 has also recently been shown to play a role in the localization of Ndc10 to the mitotic spindle (; ). Within this set of two-hybrid interacting proteins, three components of the sumoylation machinery were also identified (Ubc9, Smt3, and Nfi1). , which was originally identified as a high copy suppressor of a mutation in the kinetochore protein Mif2 (), encodes the ubiquitin-like protein SUMO. Given 's genetic interaction with the kinetochore and the number of two-hybrid interactions between Ndc10 and the sumoylation machinery, we hypothesized that these interactions occur because Ndc10 is a target for sumoylation. To address this possibility, we tested whether bacterially expressed GST-Ndc10 is a substrate for sumoylation in an in vitro sumoylation reaction. Western blot analysis of the reaction products revealed two SUMO-modified forms of Ndc10 that were generated in an ATP-dependent manner, indicating that Ndc10 is sumoylated in vitro (). To determine whether Ndc10 is modified by SUMO in vivo, we immunoprecipitated Ndc10 from a yeast cell lysate and detected SUMO-modified proteins by Western blot analysis using an antibody that recognizes the yeast SUMO protein. Two signals were apparent by Western blotting that corresponded to the correct molecular weight to be SUMO-modified forms of Ndc10 (). Both of these signals changed electrophoretic mobility upon switching the tag on Ndc10 from 13 to 3 copies of myc, demonstrating that both of these signals represent in vivo SUMO-modified forms of Ndc10 and not a coprecipitated protein (). The presence of two modified forms of Ndc10 both in vitro and in vivo suggests that Ndc10 may be sumoylated on at least two lysine residues. However, other posttranslational modifications and/or polysumoylation on a single lysine residue cannot be ruled out as reasons for the appearance of multiple modified forms of Ndc10 from these data alone. We estimate that the fraction of SUMO-conjugated Ndc10 is ∼1% or less, which is consistent with the level of modification of most known SUMO substrates. The addition of SUMO is often facilitated by a protein ligase (E3; ; ), and the observed two-hybrid interaction between the E3 protein Nfi1 and Ndc10 suggests that Nfi1 may act as an E3 for Ndc10 sumoylation. We found that strains carrying deletions ( and ) or point mutations () in known E3 proteins individually had no effect on Ndc10 SUMO modification levels (). However, in the absence of both and , Ndc10 sumoylation was reduced (), indicating a functional redundancy between Nfi1 and Siz1 in targeting Ndc10 for sumoylation. It is worth noting that Ndc10 has also been implicated as a substrate for ubiquitination (; ), raising the possibility of alternative regulation through both sumoylation and ubiquitination. However, in preliminary experiments, we have not been able to identify ubiquitinated forms of Ndc10. During the course of this study, Ndc10 was also identified as a sumoylated protein in a proteomic analysis of sumoylated proteins in yeast (). To begin investigating the possible functions of Ndc10 sumoylation, we examined the modification state of Ndc10 at discrete arrest points within the mitotic cell cycle. The mating pheromone α factor (αF) and the microtubule-destabilizing drug nocodazole (NZ) were used to arrest cells in G1 and G2/M, respectively. The terminal arrest of each population after treatment was verified by flow cytometry (unpublished data). Immunoprecipitations (IPs)/Western blotting showed that the sumoylation of Ndc10 was decreased to almost undetectable levels in those cells treated with NZ, whereas it was maintained in αF-treated cells (). The loss of sumoylation is not a general consequence of NZ treatment because other sumoylated proteins were not affected (, Ndc80) or exhibited increased levels of sumoylation upon NZ treatment (). Moreover, the loss of sumoylation is not a result of the G2/M arrest because Ndc10 does not become desumoylated in cells that are arrested in G2/M with temperature-sensitive alleles in the anaphase-promoting complex/cyclosome (). To further analyze Ndc10 sumoylation during the cell cycle, we assayed the modification state of Ndc10 in a synchronized population of cells as they progressed through mitosis after release from αF arrest (). Timing of the progression of cells through the cell cycle was monitored by flow cytometry (). The modification state of Ndc10 remained relatively constant throughout the synchronized cell cycle, and, in replicate experiments, the fluctuation observed between individual time points was not consistent, suggesting that Ndc10 sumoylation does not change dramatically over the cell cycle. These results suggest that the loss of Ndc10 sumoylation in NZ-arrested cultures may be a direct consequence of NZ addition relating to checkpoint activation and/or loss of microtubules. Specific cysteine proteases are responsible for cleaving SUMO from modified substrates and for maturation of the SUMO protein itself (). Of the two SUMO proteases in yeast, Ulp2 localizes to the nucleus, placing it in the proper location to mediate the removal of SUMO from Ndc10 during NZ exposure. We treated cells with NZ and found that the sumoylated forms of Ndc10 were still present (); therefore, Ulp2 is required for the loss of Ndc10 sumoylation in response to NZ treatment. To understand whether sumoylation is a common modification on kinetochore proteins, we tested a panel of 14 proteins comprised of inner (Cep3 and Mif2), central (Chl4, Cnn1, Ctf3, Ctf19, Iml3, Ndc80, Nuf2, and Spc24), and outer kinetochore components (Dam1 and Spc34) for modification by SUMO. We also tested the spindle checkpoint protein Bub1 and the kinetochore-associated microtubule-binding protein Bik1. Of all the proteins tested, Cep3 and Ndc80 were found to be sumoylated by IP/Western blotting, with the detection of a doublet signal for Cep3 and a ladder of sumoylated forms of Ndc80 (). Like Ndc10, the fraction of SUMO-conjugated Cep3 is 1% or less, whereas we estimate Ndc80 to be modified at slightly higher levels (∼1–5%). Ndc80 and Cep3 were also found to be sumoylated throughout the cell cycle, and, in replicate experiments, the fluctuations observed between individual time points was not consistent, suggesting that Ndc80 and Cep3 sumoylation, as observed for Ndc10, is not altered dramatically over the cell cycle (). Ndc80 was recently identified as a sumoylated protein in a proteomic analysis of sumoylated proteins in yeast by mass spectrometry along with the kinetochore proteins Bir1, Sli15, and Mcm21 (; ; ; ; ; ). Our analysis revealed that Cep3, which is a member of the CBF3 complex with Ndc10 that also localizes to the mitotic spindle (), showed decreased sumoylation in response to NZ, an effect similar to that seen for Ndc10 (). For Ndc80, which does not localize to the mitotic spindle, sumoylation does not change in NZ-treated cells, suggesting that desumoylation during the NZ-induced spindle checkpoint arrest () may be associated only with those proteins interacting with the mitotic spindle. Furthermore, the differing response in NZ is indicative of a distinct role for Ndc80 sumoylation relative to that of Ndc10 and Cep3. Sumoylation often occurs on lysine residues found in the consensus motif ΨKxE, where Ψ is any large hydrophobic residue and x is any residue (). To identify the possible sites of sumoylation in Ndc10, Ndc80, and Cep3, we mutated select lysine residues in amino acid sequences that resemble the consensus sequence for sumoylation. In total, we individually mutated 11, 14, and 6 potential sumoylation sites in Ndc10, Ndc80, and Cep3, respectively (Table S1, available at ). Strains expressing each K→R mutant were assayed for growth at 37°C, chromosome instability (CIN), and changes in sumoylation by IP/Western blotting. None of the 31 K→R individual mutations in Ndc10, Cep3, or Ndc80 caused a Ts phenotype or CIN. However, lysine residues were identified that when mutated caused an overall decrease in the SUMO modification state of Ndc10 (K651, 652R, and K779R) or abolished one specific SUMO-modified form of Ndc10 (K556R; ). Given Ndc10's localization to the mitotic spindle, these three Ndc10 K→R mutants were also assayed for proper subcellular localization. In all three cases, the localization of Ndc10 was indistinguishable from that of wild type (WT; unpublished data). Of the 14 K→R mutations in Ndc80, one lysine mutation (K231R) abolished the majority of Ndc80 sumoylation (), whereas none of the six K→R mutations in Cep3 affected its sumoylation state (not depicted). The complete loss of sumoylation in Ndc80 through a single mutation suggests that K231 may be required for SUMO modification of other lysine residues, which leads to the ladder of modified species, or that this site is polysumoylated. To distinguish between these possibilities, we checked Ndc80 sumoylation in a strain that carries a form of SUMO with mutations at positions K11, 15, and 19 that eliminate the formation of polysumoylated chains (). In Ndc80 IPs/Western blots using this strain, the ladder of modified Ndc80 proteins recognized by the SUMO antibody remained present, indicating that the Ndc80 K231 site is not polysumoylated (unpublished data). Both in vitro and in vivo Ndc10 sumoylation data showed two modified forms of Ndc10, indicating that Ndc10 may be modified by SUMO on more than one site (). Moreover, lysine mutations in single SUMO consensus sites within Ndc10 did not completely block sumoylation, suggesting that there is more than one sumoylation site in Ndc10 (). If each of these lysine residues represents a potential sumoylation site, defects caused by the loss of one site may be masked as a result of sufficient levels of modification at one of the other two sites. To circumvent this problem, we combined all three potential sumoylation site mutations to create Ndc10-4xK→R, which carries four lysine to arginine mutations at residues 556, 651, 652, and 779. By IP/Western blotting, Ndc10-4xK→R lacked any detectable levels of sumoylation, indicating that these three sites are required for the majority of Ndc10 sumoylation (). Combining the mutations at residues 556, 651, 652, and 779 also caused a dramatic reduction in the amount of Ndc10 localized to the anaphase mitotic spindle both along the length of the spindle and at the spindle midzone (), which was not caused by a change in the overall protein level of Ndc10-4xK→R (). Using a allele to arrest → strains in anaphase with elongated spindles, we quantified the localization defects associated with these mutations and found that 57% of cells had undetectable levels of Ndc10-4xK→R on the spindle, whereas 43% had faint staining that was observable but well below WT levels ( = 200). This result is in contrast to WT cells, which showed robust spindle staining in 88% of cells and faint staining in only 12% of the population ( = 200). Ndc10 may be on the spindle as part of the CBF3 complex because Cep3 was recently shown to localize to the mitotic spindle (). We found that Cep3 was also mislocalized from mitotic spindles in → mutants, supporting the idea that Ndc10 and Cep3 are present on the spindle as a complex (). The chromosomal passenger protein Sli15 and the microtubule-associated protein Ase1 still localized normally in the → mutant (not depicted) as did Bir1 (see ), suggesting that the mislocalization of Ndc10 and Cep3 is not caused by a gross defect in spindle structure. We also found that → strains are not Ts and did not show changes in Ndc10's ability to dimerize (), suggesting that these mutations have not altered the structure of the protein (unpublished data). Beyond Ndc10's canonical function at the kinetochore in chromosome segregation, recent studies have described roles for the CBF3 complex and Ndc10 in cytokinesis (; ). For this reason, we tested → mutants for multibudding, proper septin ring maturation, and defects in the axial pattern of haploid budding. In these assays, we did not observe any differences between WT and → (Fig. S1, available at ; and not depicted). In addition to the spindle localization defects, → strains showed increased CIN in a color sector assay (). When quantified by half-sector analysis (; ), → strains had rates of chromosome loss and chromosome nondisjunction 160 and 60× greater, respectively, than WT. In comparison, the loss of checkpoint function in or strains causes a 50× increase in the rate of chromosome loss (). We found that these strains also lose endogenous chromosomes at an increased rate as determined by a diploid bimating assay (). In this assay, → homozygous diploid strains formed mating colonies at a rate 10× that of WT diploids, presumably as a result of the loss of chromosome III (2N-1; ). Although → strains showed increased rates of CIN, we have been unable to observe a delay in the cell cycle progression of a αF-synchronized cell culture (). → strains are also G2/M checkpoint proficient in both their ability to arrest and recovery from NZ exposure (unpublished data). These results suggest that the defect causing chromosome missegregation is not eliciting a checkpoint response and frequent repair but is more likely to be either a rare event that always leads to failure or an event that does not trigger a G2/M checkpoint response at all. A common phenotype associated with mutations affecting kinetochore and/or spindle function is increased sensitivity to the microtubule-destabilizing drug benomyl. → strains displayed benomyl sensitivity (), which, taken together with the mitotic spindle mislocalization phenotype, raises the possibility that the loss of Ndc10 sumoylation may be causing specific defects in spindle function. To investigate this possibility, we measured the lengths of mitotic spindles in anaphase-stage cells of an asynchronous culture. In the → mutant, we found that anaphase spindles are of abnormal length (representative images in Fig. S2, available at ), with spindles averaging 7.30 ± 0.08 μm ( = 483) compared with 6.62 ± 0.06 μm ( = 443) for WT cells. In comparing the distribution of spindle lengths graphically, we noted that the → mutants had abnormally long anaphase spindles ranging up to 10–12 μm, which are lengths that are never seen in WT cells (). The observed spindle defect and CIN, which is associated with no observable G2/M delay, suggests that the defects caused by the loss of Ndc10 sumoylation is related to events in anaphase after the G2/M checkpoint has been silenced (e.g., spindle elongation). In anaphase, Ndc10 and Bir1 colocalize on the mitotic spindle in a Bir1-dependent manner, suggesting that Ndc10 and Bir1 are part of a complex on the mitotic spindle (; ). In the → strain, Ndc10 no longer localized to the mitotic spindle (), whereas Bir1 remained spindle bound (). Therefore, we hypothesized that the spindle length defect and mislocalization of Ndc10 was caused by a disruption of the Ndc10–Bir1 protein interaction. However, co-IPs demonstrated that the Ndc10-4xK→R interaction with Bir1 was comparable with that seen in WT cells (). We also found that the interaction between Ndc10 and Bir1 was not disrupted in cells arrested with αF or NZ, demonstrating that this interaction is not specific to mitosis and is not microtubule dependent (). These results suggest that the observed interaction between Ndc10 and Bir1 occurs independently of the mitotic spindle and that the spindle length defect in the → strain is not caused by the failure of Ndc10-4xK→R to interact with Bir1. Bir1 is a phosphorylated protein that was also recently identified as a sumoylated protein in proteomic studies (; ; ). We were able to confirm that Bir1 is sumoylated by IP/Western blotting, with an estimated 1% of Bir1 being SUMO conjugated in an asynchronous culture. As seen for Ndc10, we also observed that the treatment of cells with NZ resulted in the loss of Bir1 sumoylation (). Given that the interaction between Ndc10 and Bir1 is not disrupted in an → strain, we wondered whether the loss of Ndc10 sumoylation and/or spindle localization resulted in a change in Bir1's modification state. Therefore, we assayed both the phosphorylation and sumoylation state of Bir1 in the → mutant strain. Western blots show that the overall appearance of Bir1 phosphorylation (as indicated by the diffuse Western signal representing phosphorylated forms of Bir1) is unchanged in the mutant as compared with WT (, α-myc blot) but that the sumoylation of Bir1 is reduced in the → mutant to undetectable levels (, α- SUMO blot). We tested Cep3 and Ndc80 for changes in the sumoylation state in the → mutant () but found no significant differences compared with WT; thus, the loss of sumoylation in the → mutant appears to be specific to the modification of Bir1. These data suggest that the sumoylation of Ndc10 is required for Bir1 sumoylation. Ndc10's effect on Bir1 sumoylation may be dependent on the activity of the CBF3 complex or may occur via a CBF3-independent mechanism. To distinguish these two possibilities, we tested Bir1 sumoylation in strains carrying Ts mutations in the CBF3 component Cep3 () or Ctf13 (). In both cases, when cells were arrested at the nonpermissive temperature, Bir1 sumoylation was detected; however, the loss of sumoylation was observed when the Ts allele was used to abrogate Ndc10's function. This result indicates that Ndc10's effect on Bir1 sumoylation is not caused by altered CBF3 activity (). In this study, we report that the kinetochore proteins Ndc10, Cep3, Ndc80, and Bir1 are substrates for sumoylation. In the case of Ndc10, the effect of lysine mutations on Ndc10's localization demonstrates that sumoylation demarcates a specific subset of Ndc10 that associates with the mitotic spindle. This demarcation most likely occurs through an alteration of Ndc10's binding interactions with microtubules or microtubule-associated proteins, leading to the establishment or maintenance of spindle localization. Our data also assigns a function to the fraction of Ndc10 protein on the mitotic spindle in controlling mitotic spindle dynamics. Several other kinetochore proteins also localize to the spindle during anaphase, including Bir1, Cep3, Cin8, Dam1, Duo1, Ipl1, Sli15, Slk19, and Stu2, most of which have also been shown to play a role in controlling spindle dynamics (; ; ; ; ; ; ; ). The localization of Ipl1, Sli15, and Ndc10 to the spindle requires the function of the phosphatase Cdc14 that is activated in anaphase to regulate mitotic exit (; ; ). Moreover, Ndc10, Bir1, Cep3, and Sli15 are SUMO substrates (; and this study), indicating that these spindle-associated kinetochore proteins may be regulated via common mechanisms to control mitotic spindle dynamics during anaphase. The regulation of spindle dynamics likely includes delivery of these kinetochore proteins to the spindle midzone, which may be used as a signal to coordinate the collapse of the mitotic spindle and to initiate other late-stage mitotic events once chromosome segregation has occurred. Given →'s long spindle phenotype, it is possible that the signal to commence spindle disassembly is delayed as a result of the absence of Ndc10 on the mitotic spindle, providing time for additional spindle elongation. It is currently unknown whether the altered spindle dynamics are also responsible for the increase in CIN observed in → mutants, but the lack of an observable cell cycle delay is suggestive of events that are invisible to the mitotic checkpoint machinery or that occur during stages of the cell cycle when the checkpoint has already been satisfied (e.g., anaphase). Intriguingly, Ndc10 also localizes to microtubules in telophase and into G of the next cell cycle, raising the possibility that microtubule-associated Ndc10 may be required in nonmitotic stages of the cell cycle (). This may include S phase when centromere DNA is replicated and kinetochore microtubule attachments are being established. Ndc10's localization to the mitotic spindle is dependent on sumoylation and Bir1 (; ; and this study). Our work demonstrates that the interaction between Ndc10 and Bir1 occurs in nonmitotic cells and is independent of microtubules, Ndc10 spindle localization, and sumoylation of Ndc10 or Bir1. How then does Bir1 facilitate Ndc10's spindle association if not by physically interacting with Ndc10 on the mitotic spindle? Based on localization data, the other structure at which these two proteins colocalize is the kinetochore (). At the kinetochore, Bir1 could direct Ndc10 to microtubules in a manner dependent on the function of the Bir1–Sli15–Ipl1 kinase complex. In support of this possibility, Ndc10 has a weak two-hybrid interaction with Sli15 (). Moreover, Ndc10 has been shown to be an Ipl1 substrate for phosphorylation in vitro, and mutants, like → mutants, contain spindles of abnormal length (; ). In experiments that use the temperature-sensitive mutant, Ndc10 shows only slightly lower levels of mitotic spindle localization (). However, it is known that at the restrictive temperature, retains a low level of kinase activity, and, thus, a phosphorylation-dependent mechanism is still possible (). In contrast to Ndc10, mutations that abolish Bir1 sumoylation () do not result in a loss of Bir1 mitotic spindle localization (), implying that Bir1's association with the mitotic spindle does not rely on sumoylation. A clue to the function of Bir1's sumoylation may come from a recent study on the mammalian homologue of Bir1, Survivin. In mammals, the modification of Bir1/Survivin with ubiquitin appears to regulate dynamic protein–protein interactions that are important for chromosome segregation at the centromere (). Intriguingly, these modifications map to the IAP region of mammalian Bir1/Survivin (), the deletion of which results in a loss of sumoylation in yeast Bir1 (). Thus, the IAP domain may be used to regulate both yeast and mammalian Bir1/Survivin's function by lysine-directed modifications. Once Ndc10 and Bir1 are on the mitotic spindle, it appears that these two proteins function to regulate spindle dynamics in an opposing manner given that the spindle length defect observed in → mutants is in contrast to that seen in Bir1 mutants in which spindles fail to fully elongate and are on average shorter than WT (; and this study). The interplay between these two proteins is further illustrated by the fact that in the → mutant, Bir1 sumoylation is undetectable even though the Ndc10 mutant protein and Bir1 still interact normally. This dependence is not reciprocal in that Ndc10 sumoylation is not affected by mutations that block Bir1's sumoylation or that disrupt the interaction between Bir1 and Ndc10 (). The reliance on Ndc10 being competent for sumoylation suggests that Ndc10 requires prior modification with SUMO before the sumoylation of Bir1 can occur and is indicative of a cascade of SUMO modification events in which Ndc10 functions in trans to facilitate the modification of Bir1 by SUMO. Although Ndc10 is not conserved in higher eukaryotes, Bir1 and Ndc80 are well conserved. An important question is whether the sumoylation of kinetochore proteins, including Bir1 and Ndc80, function in an analogous manner in these evolutionally diverse organisms. Although neither Bir1 nor Ndc80 homologues have been shown to be sumoylated in higher eukaryotes, there is evidence linking sumoylation to kinetochore and mitotic spindle functions. For example, SUMO-2–modified proteins are enriched at the inner centromere of chromatids in egg extracts, and the alteration of SUMO modification of chromosomal substrates by SUMO-2 causes a block in the segregation of sister chromatids in anaphase (, ). Moreover, RanGap1, the first identified sumoylated substrate, is targeted to mitotic spindles and kinetochores, and, like Ndc10, mutations that abolish sumoylation lead to RanGAP1 mislocalization from the spindle (; ; ; , ). Complexed with RanGAP1 is the SUMO E3 ligase RanBP2, which, when depleted in mitotic cells, results in the mislocalization of RanGAP1, the spindle checkpoint proteins Mad1 and Mad2, and the kinetochore proteins CENP-E and -F (). The phenotypic consequences of these effects include the accumulation of mitotic cells with multipolar spindles and unaligned chromosomes. The localization of RanBP2 to kinetochores places an E3 protein near the kinetochore, highlighting the possibility that there may be many proteins of the kinetochore targeted for sumoylation. The number of sumoylated substrates in yeast and higher eukaryotes continues to grow rapidly, but the biological functions of the majority of these modifications remains elusive. The data presented here provide evidence for the regulation of chromosomal passenger proteins and kinetochore proteins by SUMO modification, including the localization of Ndc10 to the mitotic spindle during anaphase. Findings in higher eukaryotes provide evidence that the mechanism for localizing proteins to microtubules via sumoylation may be conserved (e.g., RanGAP1). Future work will be required to discern the details and extent of this mechanism in eukaryotic cells and the importance of sumoylation in regulating the mitotic program. Yeast strains (S288C background) and plasmids used in this study are listed in Tables S2 and S3 (available at ). Mutant alleles were integrated into the yeast genome at the endogenous locus by cotransformation of a PCR product carrying the desired mutations and a PCR product containing a nutritional marker (). Integration of the desired mutation was confirmed by DNA sequencing. To arrest cell cultures, 1 mg/ml αF (in methanol) or 5 mg/ml NZ (in DMSO) was added, and cultures were incubated for 2 h at 30°C. To assay benomyl sensitivity, benomyl was added at the indicated concentration to YPD media, dimethyl sulfoxide was used in the control plate (0 μg/ml benomyl), and fivefold serial dilutions were spotted on the plates and grown at 30°C for 2 d. Flow cytometry analysis to monitor DNA content was performed as previously described (). NDC10 was cloned into pOBD2 and pBDC using standard techniques (; ). Fusion of the DNA-binding domain to the C terminus (pBDC) but not the N terminus of Ndc10 (pOBD2) resulted in a functional protein as judged by the ability to rescue an ndc10Δ strain. Two independent genome-wide two-hybrid screens were performed using an activation domain array () as described previously (). The two-hybrid positives from these genome-wide screens were reconfirmed by repeating the two-hybrid assay. The identities of the activation domain fusions were confirmed by rescuing plasmids and sequencing. GST-Ndc10 was expressed in strain BL21 from a pGex 4T-2 vector (GE Healthcare) and was purified using glutathione beads. In vitro sumoylation reactions were performed using bacterially expressed proteins (expression plasmids were provided by L. McIntosh, University of British Columbia, Vancouver, British Columbia, Canada) as previously described (), and Western blotting was performed with an Ndc10 antibody (). To detect sumoylation in vivo, yeast cells (100–200 OD600) were lysed by bead beating (10 times for 30 s) in 2.5 ml lysis buffer (50 mM Tris-HCl, pH 8.0, 5 mM EDTA, 150 mM NaCl, 0.2% Triton X-100, complete protease inhibitor [one tablet per 25 mL; Roche], 10 mM -ethylmaleimide, 2 mM PMSF, and 20 μg each of leupeptin, aprotinin, and pepstatin per milliliter) on ice. Lysates were cleared by centrifugation at 30,000 for 20 min, and soluble protein concentrations were determined by protein assay (Bio-Rad Laboratories). Equal amounts of protein were incubated with α-myc–conjugated beads for 3 h at 4°C and washed four times with cold lysis buffer for 2 min. Immunoprecipitated protein was then eluted with lysis buffer containing 2% SDS at 42°C for 15 min. 2.5 μL of the eluted protein was used for Western blotting with α-myc antibody to confirm pull down of the tagged protein; 25 μl was used for blotting with α-SUMO antibody to detect sumoylated protein. To detect the Ndc10–Bir1 interaction, 25 μl Bir1 IP was used for Western blotting with α-Ndc10 antibody. α-SUMO polyclonal antibodies were generated in rabbits (Covance Research Products) as previously described (). Initial experiments also made use of an α-SUMO polyclonal antibody provided by E. Johnson (Thomas Jefferson University, Philadelphia, PA). Strains used for microscopy were grown in either YPD (for synchronous culture experiments) or in fluorescent protein medium (minimal medium supplemented with adenine and containing 6.5 g/L sodium citrate). Cells were imaged at room temperature using a microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) with a plan-Apochromat 100× NA 1.4 differential interference contrast oil immersion objective (Carl Zeiss MicroImaging, Inc.) with filter sets 38 and 47 (Carl Zeiss MicroImaging, Inc.) and filter 488000 (Chroma Technology Corp.). 3D images (0.25-μm steps) were acquired with a camera (CoolSNAP HQ; Roper Scientific) and analyzed using MetaMorph software (Invitrogen). Images are presented as maximum intensity 2D projections. For fluorescence microscopy, WT and mutant proteins were tagged at the endogenous locus with GFP or a GFP variant (). Tub1-CFP–containing strains were generated by integrating plasmid pSB375 (a gift from K. Bloom, University of North Carolina, Chapel Hill, NC) digested with StuI at the URA3 locus. Spindle length measurements were performed on asynchronous cultures of Tub1-CFP–containing cells fixed in 70% ethanol and 200 mM Tris-HCl, pH 8.0, which was judged to be in anaphase by the presence of part of the spindle in the daughter cell. To visualize the haploid budding pattern, cells were incubated in PBS with 20 μg/ml Calcofluor white (Fluorescent Brightener 28; Sigma-Aldrich) at 25°C for 5 min, washed in PBS, and visualized for bud scar staining. Quantitative half-sector analysis was performed as described previously (; ). To perform the diploid bimater assay, 10 single colonies were patched onto YPD plates, replica plated to both MATa and MATα lawns, and mating products were selected. The median number of mating products from the 10 patches was compared with WT in order to calculate the increase in frequency over mating of a WT diploid strain. Fig. S1 shows that ndc10-4xK→R strains do not have cytokinesis-related defects. Fig. S2 contains representative images of CFP-Tub1–marked anaphase mitotic spindles in WT and ndc10-4xK→R strains. Table S1 lists the potential sumoylation sites targeted by site-directed mutagenesis in Ndc10, Ndc80, and Cep3. Tables S2 and S3 list the yeast strains and plasmids used in this study, respectively. Online supplemental material is available at .

Centrosomes are the primary microtubule-organizing center in most eukaryotic cells. Centrosomes are considered to be soluble cytoplasmic organelles that lack membrane-associated structures. However, centrosomes are closely associated with the nuclear envelope in most interphase cells. A variety of data suggests that a biochemical link exists between the centrosome and nuclear membrane (; ; ; ), and tethering of the centrosome to the nuclear envelope might be important for rapid formation of the mitotic spindle, nuclear positioning within the cell, and/or centrosome inheritance. SUN (Sad1-UNC-84 homology) domain–containing proteins are an emerging family of inner nuclear envelope proteins that are excellent candidates to play a role in connecting centrosomes with the nuclear envelope. Originally identified based on an ∼150-amino-acid region of homology between the C terminus of the Sad1 protein and the UNC-84 protein (; ), SUN proteins are present in the proteomes of most eukaryotes (Fig. S2, available at ; ). In addition to the SUN domain, these proteins contain a transmembrane sequence and at least one coiled-coil domain and localize to the inner nuclear envelope. SUN proteins are anchored in the inner nuclear envelope by their transmembrane segment and oriented in the membrane such that the C-terminal SUN domain is located in the space between the inner and outer nuclear membranes. Here, the SUN domain can interact with the C-terminal tail of an outer nuclear envelope protein that binds to the cytoskeleton, including the centrosome (for review see ). The spindle pole body (SPB) is the centrosome-equivalent organelle and is the sole site of microtubule nucleation in budding yeast. The SPB is a multilayered cylindrical structure that is embedded in the nuclear envelope throughout the yeast life cycle (for review see ). The soluble SPB core consists of three primary layers called the outer, inner, and central plaques. Cytoplasmic and nuclear microtubules are nucleated from the outer and inner plaques, respectively, whereas the central layer of the SPB plays an important role in tethering the organelle to the nuclear envelope. Associated with one side of the core SPB is an electron-dense region of the nuclear envelope termed the half-bridge, which is important for SPB duplication as well as for microtubule nucleation during G1 (, ). SPBs in fission yeast have a similar but not identical structure. Importantly, the SPB is also embedded in the nuclear envelope, possibly by its SUN protein, Sad1 (). Until now, the budding yeast orthologue of Sad1 had not been identified. The half-bridge is critical for SPB duplication; yet, details of its structure and function at a molecular level are only beginning to emerge. Four proteins are found at the half-bridge: Cdc31, Kar1, Mps3, and Sfi1. Kar1 and Mps3 are integral membrane proteins that localize to the cytoplasmic and nuclear sides of the half-bridge, respectively (; ; ), whereas Cdc31 and Sfi1 are soluble half-bridge components (; ). Recently, multiple Cdc31 proteins were shown to associate with Sfi1 to form a soluble cytoplasmic filament that spans the length of the half-bridge, and a model in which duplication of the Cdc31-Sfi1 filament generated a non-SPB associated end to initiate assembly of a new SPB was proposed (). However, it is unclear how Mps3 and Kar1 associate with this filament and what roles both membrane proteins play during SPB duplication and assembly. One possibility is that Mps3 and Kar1 form the physical half-bridge because mutations in and cause cells to arrest with unduplicated SPBs that lack any recognizable half-bridge structure (; ). Interestingly, Cdc31 and Sfi1 may play a role in the insertion step later in SPB duplication, although this function is poorly understood (; ). The insertion step of SPB duplication also requires the membrane proteins Mps2 and Ndc1 and their respective binding partners Bbp1 and Nbp1 (, ; ; ). These proteins may be recruited to the half-bridge to facilitate insertion of the newly formed SPB into the nuclear envelope and/or to tether the half-bridge to the core SPB. Direct binding between the half-bridge and membrane components of the SPB has not been demonstrated, but genetic and two-hybrid interactions between Bbp1 and Kar1 suggest that the half-bridge is connected to the core SPB through these two proteins (). Not only is the half-bridge important for SPB duplication, but it is also essential for nuclear migration and fusion after mating (karyogamy; for review see ). During G1, cytoplasmic microtubules are nucleated from the half-bridge instead of from the outer plaque (). After mating, half-bridge microtubules interdigitate, allowing the two nuclei to congress, and the juxtaposed half-bridges form the site where SPB and nuclear membrane fusion originates (, ). The importance of the SPB half-bridge for nuclear migration and fusion is illustrated by the fact that mutations in , , and cause defects in both steps of karyogamy (; ; ). In the present work, we show that budding yeast half-bridge protein Mps3 is homologous to the SPB component and SUN protein Sad1 (). Mps3 also contains a SUN domain, and mutational analysis of this region of Mps3 demonstrated that it is critical for Mps3 function during SPB duplication and karyogamy. The SUN domain of Mps3 binds to the C terminus of another integral membrane component of the SPB, Mps2, and the Mps2–Mps3 interaction is required for formation of an intact SPB. Our results demonstrate at a molecular level how the half-bridge is tethered to the core SPB and describe the consequences of disrupting this interaction. They also support the novel function of SUN proteins in bridging the inner and outer nuclear envelope and provide further evidence that SUN proteins function to tether centrosomes to the nuclear envelope. encodes a 682-amino-acid protein with an acidic N terminus, a transmembrane segment, two coiled-coil domains, and a poly-glutamine region (; ). Analysis of secondary structure and local compositional complexity also suggests that the C-terminal region of Mps3 (amino acids 436–682) folds into a discrete globular domain, which has been shown by deletion analysis to be important for Mps3 function (Tables S2 and S3, available at ; ). To learn more about the function of the C-terminal domain, we searched for similar sequences in protein databases using PSI-BLAST with the model inclusion cutoff set at 0.05 (). After the first round of search, only Mps3 orthologues from other budding yeasts were detected; however, after the second iteration, we identified numerous proteins from various eukaryotes that showed homology to the Mps3 C terminus. The best matches were to the C-terminal SUN domain of the Sad1 and to other SUN domain–containing proteins. Alignment of the C terminus of Mps3 with the SUN domain from Sad1 and other SUN proteins indicates a good fit between Mps3 and the sequence consensus of the SUN domain family (). The 514-amino-acid Sad1 protein shares a common overall architecture with Mps3 () and localizes to the SPB (), raising the possibility that Mps3 and Sad1 perform similar cellular roles. We overexpressed in various budding yeast temperature-sensitive (ts) mutants in and found that growth at the nonpermissive temperature was partially restored in cells (see the following section; ), indicating that Sad1 and Mps3 are functionally related. Mutation of the highly conserved asparagine to lysine at position 597 of (; N597K) suggests that the SUN domain is important for Mps3 function during mitosis and mating (). To more thoroughly analyze the role of the Mps3 SUN domain, we constructed an allelic series within the Mps3 SUN domain by mutating residues that are conserved or similar in most vertebrate SUN domains () or are conserved in at least four out of the six sequenced species (). We found that many of the conserved residues are essential for Mps3 function because their mutation results in a nonviable allele (Tables S2 and S3). However, mutation of some conserved residues, such as serine 516, proline 518, and tyrosine 563, had no effect on cell viability. In our analysis, we isolated a series of five new ts alleles (, , , , and ; and Tables S2 and S3). We also found that loss of the second half of the SUN domain (; amino acids 524–645) or mutation of the conserved tryptophan at position 477 () resulted in ts alleles if multiple copies of the mutant gene were present in the cell (Table S3). We suspected that the SUN domain was critical for Mps3 function during SPB duplication because several of our mutants spontaneously diploidize (; and Fig. S1, available at ), a phenotype shared by and some and mutants that are defective in SPB duplication. To test whether our Mps3 SUN domain mutants show defects in SPB duplication, asynchronously growing wild-type and mutant cells were analyzed after a 4-h shift to the nonpermissive temperature (37°C). Flow cytometric analysis of DNA content and budding index confirmed that all of the SUN domain mutants arrest in mitosis at 37°C. Analysis of mitotic spindle morphology by indirect immunofluorescence microscopy and EM revealed that >75% of large-budded , , , and cells contain a monopolar spindle at 37°C (; and Fig. S1). These results suggest that the SUN domain in Mps3 is required for SPB duplication. In , , and mutants, we also observed monopolar spindles in roughly half of the large-budded cells by both indirect immunofluorescence microscopy and EM (; and Fig. S1). The remaining cells contained short, bipolar spindles with a single DNA mass indicative of a metaphase arrest. This heterogeneous arrest phenotype supports the possibility that Mps3 has a second function in mitotic progression separate from its role in SPB duplication that is defective in these mutants (see Discussion). Like mutants, we observed that the growth defect of some of our SUN mutants was partially rescued by overexpression of components of the half-bridge, including (; ). The fact that these mutants also displayed defects in Cdc31 localization at 37°C suggests that their failure in SPB duplication is due to an inability to bind to Cdc31 at the half-bridge (unpublished data). Interestingly, the growth defect of all of the SUN domain mutants at 37°C was suppressed at least in part by overexpression of or (). One interpretation of this genetic interaction is that the SUN region of Mps3 might interact directly with Nbp1 or Mps2; therefore, having more of either Mps3 binding partner present in the cell could stabilize a weak interaction with the mps3 SUN mutant protein and allow the cells to proliferate at the restrictive temperature. Defects in Cdc31, Mps2, and/or Nbp1 binding could also explain why localization of most of mps3 SUN domain mutant proteins to the SPB is reduced or eliminated in cells grown at 37°C (), even though total protein levels determined by Western blotting are not dramatically altered (). Mps3 is important not only for SPB duplication but also for karyogamy. Analysis of the allele, which contains a mutation in one of the most highly conserved residues in the SUN domain, indicated that Mps3 is required for both steps of karyogamy: nuclear congression and fusion (). To test whether our mutants displayed karyogamy defects, we analyzed the position of nuclei in mating mixtures of wild-type and SUN domain mutants after 5 h at the semipermissive temperature of 30°C; at higher temperatures, karyogamy itself is affected and our mutants begin to exhibit a significant delay in mitosis, a cell cycle stage that is nonpermissive for mating. After 5 h, 95% of zygotes contained a single nucleus in the wild-type mating, whereas only 60–70% of SUN domain mutant zygotes had successfully completed both steps of karyogamy and contained a single nucleus (). The SUN domain mutants displayed a unilateral defect in karyogamy (unpublished data) and were defective in both nuclear congression and fusion, although the congression defect was more pronounced. We suspect that allele and strain-background differences are responsible for the less severe karyogamy phenotypes that we observed compared with others (). The simplest interpretation of the karyogamy defect, like the SPB duplication defect, is that SUN domain mutants fail to form an intact half-bridge, which is critical for both nuclear congression and fusion (). Because Mps3 is located at the SPB and SUN proteins are thought to interact with other integral membrane proteins, we hypothesized that the Mps3 SUN domain might bind Kar1 or Mps2, the other single pass membrane proteins at the SPB (; ; ). Full-length Kar1 and Mps2 expressed as maltose binding protein (MBP) fusions in bacteria were transferred to nitrocellulose membranes after SDS-PAGE of bacterial extracts, and the membrane was probed with the Mps3 SUN domain (amino acids 457–617) labeled with the infrared dye Alexa 680 in a gel overlay assay. We found that the SUN domain bound to full-length Mps2 (MBP-Mps2) and to the Mps2 C terminus (MBP-mps3-Ct) but did not interact with MBP, Kar1 (MBP-Kar1), or other proteins in the bacterial extracts (). Binding of the Mps3 SUN domain to Mps2 required a functional SUN domain because most mps3 SUN domain mutant proteins were unable to compete with the wild-type protein for Mps2 binding in the gel overlay assay, even when present in at least 10-fold excess ( and not depicted). To test whether Mps2 and Mps3 interact in yeast, we created strains expressing and/or from their endogenous promoters and immunoprecipitated the tagged proteins from the membrane fraction of lysates from spheroplasted cells treated with the short 12-Å membrane-permeable cross-linker dithiobis(succinimidyl)propionate. Mps2-13xMYC coimmunoprecipitated with Mps3-3xFLAG, as did the Mps3 binding protein Cdc31 (; ). The fact that Mps2-13xMYC did not precipitate in cells lacking Mps3-3xFLAG and that other SPB components such as Spc29 did not coprecipitate indicates that binding is specific and likely occurs through Mps3. Mps2 and Mps3 also coimmunoprecipitated in the absence of a cross-linker when cell lysates were prepared by grinding in liquid nitrogen (), providing further evidence of a direct interaction between Mps2 and Mps3 in yeast. However, as SPBs remain largely intact by this method of lysis (), we cannot totally exclude the possibility that the Mps2–Mps3 interaction is mediated by another SPB component. From the binding data in vitro and our ability to coimmunoprecipitate Mps2 and Mps3 from yeast extracts using two different techniques, we conclude that Mps2 binds to Mps3 in vivo and that this interaction is likely mediated by the Mps3 SUN domain. Several proteins predicted to bind to SUN domains contain a conserved C-terminal transmembrane segment followed by ∼30 amino acids known as the Klarsicht/ANC-1/Syne homology (KASH) domain (). Protease protection and protein binding experiments demonstrated that the N terminus of Mps3 is exposed to the cytoplasm/nucleoplasm, whereas the C terminus, including the SUN domain, is located in the space between the inner and outer nuclear membranes (). In contrast, the N terminus of Mps2 is exposed to the cytoplasm/nucleoplasm, whereas the C terminus of Mps2 is in the intermembrane space (). Therefore, we would predict that the Mps3 SUN domain binds to the C terminus of Mps2. However, neither the Mps2 C terminus nor any other protein sequence in contain regions of significant similarity to a hidden Markov model of the KASH domain (unpublished data); instead, Mps2 and its orthologues from other budding yeasts have a unique C-terminal region that includes seven conserved aromatic residues (). Despite its lack of a KASH domain, the Mps2 C terminus clearly binds to the Mps3 SUN domain in both the gel overlay assay and pull-down assays in vitro ( and not depicted). In addition, a version of Mps2 lacking the C-terminal amino acids (mps2ΔC; amino acids 1–328) does not bind to the SUN domain in vitro ( and ) and is unable to rescue the growth defect of and serve as the sole copy of in cells (). Thus, the C terminus of Mps2, like the Mps3 SUN domain, is critical for cell viability. Analysis of a series of point mutations (not depicted) and truncations () in the Mps2 C terminus indicates that residues throughout the region are essential for Mps2 function and for Mps3 binding. Insertion of a nonsense codon after residue 381 in Mps2 resulted in an allele of () that encodes a protein containing all but the last six amino acids. The fact that cells display a conditional growth defect (), combined with the fact that the mps2-381 protein only weakly bound to the SUN domain in vitro (), is evidence that the C terminus of Mps2 is important for the interaction with Mps3 in vivo. Additional support for the notion that the growth defect in cells is due to loss of Mps3 binding comes from our observations that overexpression of wild-type , but not SUN domain mutants, partially rescues the growth defect of mutants at 37°C (). Interestingly, overexpression is unable to suppress mutants that contain lesions in the N terminus (not depicted), whereas overexpression, which suppresses some N-terminal mutants (), only weakly rescues growth of (). These results strongly suggest that the SUN domain of Mps3 binds to the C terminus of Mps2 and supports the possibility that loss of Mps3 binding is responsible for the growth defect of and other C-terminal Mps2-deletion mutants. Asynchronously growing wild-type and mutant cells were analyzed after a 4-h shift to the nonpermissive temperature (37°C). Flow cytometric analysis of DNA content and budding index confirmed that the mutants spontaneously diploidize at 23°C and arrest in mitosis at 37°C (). Analysis of mitotic spindle morphology by indirect immunofluorescence microscopy and EM revealed that >75% of large-budded cells contain a monopolar spindle at 37°C (). Other mutants also have monopolar spindles (); however, mutants have a distinct terminal SPB mutant phenotype: they contain a single monopolar spindle on a deep nuclear invagination, and the SPB appears to have failed to initiate SPB duplication (). The morphology of the SPB is highly reminiscent of many of our SUN mutants ( and Fig. S1), and it appears that SPBs lack a normal half-bridge, although visualization of the half-bridge by EM alone on such a constricted invagination is difficult. We found no evidence by morphology or by immunolabeling of a second partially assembled SPB like that observed in mutants in spite of good overall fixation (unpublished data; ). This novel terminal SPB phenotype uncovered by the allele suggests that Mps2 has an earlier function in SPB duplication, perhaps during half-bridge formation or tethering of the half-bridge to the SPB, which was not previously indicated through analysis of other mutants. Like the Mps3 SUN domain, the Mps2 C terminus is important not only for SPB duplication but also for both steps of karyogamy. Although other alleles also have a mild karyogamy defect, >30% of cells fail to complete both nuclear congression and fusion, indicating that Mps2–Mps3 binding is important for karyogamy (). The fact that mutants have phenotypes very similar to our SUN mutants and to other mutants that lack a functional half-bridge led us to examine the localization and levels of Mps3 and the half-bridge component Cdc31 in cells. Asynchronously growing wild-type and cells that have the endogenous copy of fused to GFP were shifted to 37°C for 3 h. Western blot analysis revealed that Mps3-GFP, as well as other SPB components such as Cdc31, Spc110, and Tub4, were expressed at similar levels in and cells (). However, we consistently saw a decrease in the number of cells grown at 37°C containing Mps3-GFP, as well as a decrease in the overall intensity of Mps3-GFP epifluorescence at the SPB (). Cdc31 staining at the SPB by indirect immunofluorescence also decreased in mutants at 37°C (). Instead of a distinct SPB signal, Mps3-GFP was diffusely localized and Cdc31 appeared as punctate spots in cells at 37°C. No change in expression or localization of core SPB components such as Spc110 and Tub4 was observed in mutants at either temperature ( and ). The failure of mutants to localize Mps3-GFP and Cdc31 to the SPB is not due to defects in 9xMYC-mps2-381 expression or its localization to the SPB; 9xMYC-Mps2 and 9xMYC-mps2-381 both localized to the SPB at 23 and 37°C (). Together with our data showing that mps2-381 binding to the Mps3 SUN domain is reduced in vitro, these data strongly suggest that mutants fail to bind and localize components of the half-bridge to the SPB. We propose that the Mps3 SUN domain and the Mps2 C terminus form an intermolecular bridge spanning the nuclear membranes to tether the half-bridge to the core SPB. This interaction is critical for assembly and/or maintenance of an intact half-bridge (). Association between the core SPB and the half-bridge is critical for SPB duplication and function, yet the molecular details of this link have not been characterized at a biochemical level. In this work, we show that SPBs in Mps3 SUN domain and Mps2 C-terminal mutants lack an associated half-bridge and find that the Mps2 C terminus directly binds to Mps3 in a SUN domain–dependent manner. This strongly suggests that the Mps2–Mps3 interaction is essential to anchoring the half-bridge to the core SPB (). When combined with previous studies showing that the N terminus of Mps2 binds to Bbp1, which in turn interacts with the central plaque component Spc29 (), our results provide a missing physical connection between the half-bridge and the core SPB. Previous analysis of mutants showed that Mps2 function is required for insertion of the newly duplicated SPB into the nuclear envelope (). However, our findings that mutants lack any recognizable half-bridge structure and fail to localize Mps3 and Cdc31 to the SPB suggests that Mps2 has an additional function earlier in SPB duplication. Based on analysis of Mps2–Mps3 binding, we propose that Mps3 recruits and binds to Mps2 at the distal tip of the newly assembled bridge during SPB duplication. Once this connection between the C terminus of Mps2 and the Mps3 SUN domain is established, Mps2 can interact through its N terminus with additional proteins such as Bbp1, Nbp1, and Ndc1 to facilitate insertion of the newly duplicated SPB into the nuclear envelope. This model would explain why previously characterized alleles, which all contain N-terminal mutations, fail in the late step of SPB duplication like , , and mutants (, ; ; ). We would also predict that binding between Mps2 and Mps3 is required to tether the half-bridge to the core SPB from the point of SPB duplication throughout the rest of the cell cycle. Consistent with this hypothesis, cells fail to maintain an intact half-bridge at the restrictive temperature, resulting in defects in the earliest step of SPB duplication as well as in karyogamy. Is the Mps2–Mps3 interaction the only binding event required to tether the half-bridge to the core SPB? Two-hybrid interactions have been observed between Bbp1 and Kar1 () as well as between Ndc1 and Kar1 (unpublished data), and numerous genetic interactions have been reported among Mps2, Ndc1, Bbp1, Nbp1, and components of the half-bridge (; ; ), so we suspect that a complex set of protein–protein interactions likely anchors the half-bridge to the core SPB. Multiple overlapping interactions would help ensure the integrity of the entire SPB and allow it to withstand microtubule forces that push and pull on the SPB and nucleus during the yeast life cycle. Redundancy in protein–protein interactions holding the half-bridge to the core SPB could also explain why we observe only ∼50% reduction in Mps3 and Cdc31 localization to the SPB in mutants at 37°C. Further analysis of the molecular interactions between the membrane and half-bridge components of the SPB will allow us to better understand how the SPB is assembled and duplicated. Analysis of Mps3 SUN mutants in yeast suggest that SUN proteins have multiple evolutionarily conserved functions. First, the role of the Mps3 SUN domain in maintaining SPB integrity by tethering the soluble core proteins to the nuclear envelope components suggests the interesting possibility that this protein family is involved in connecting centrosomes to the nuclear envelope. Although the idea that centrosomes are physically linked to the nuclear envelope is controversial, old and new evidence suggests that such an association may exist (; ; ; ). Low sequence conservation between SUN domains makes it impossible to predict the phylogenetic relationships of most proteins within the SUN family with statistical significance. However, our functional data suggests that Mps3 is most related to fission yeast Sad1, SUN-1, and mammalian Spag4, which have been shown to be essential for tethering SPBs, centrosomes, and axonemes, respectively, with the nuclear envelope (; ; ). Interestingly, these Mps3-like SUN proteins not only interact with various types of microtubule-organizing centers but also appear to bind to proteins, such as Kms1, ZYG-12, and now Mps2, that show very little sequence similarity to the KASH domain of other SUN binding proteins that interact with the actin cytoskeleton (). One idea is that microtubule SUN proteins may recognize a distinct or divergent motif that is currently poorly defined because of the low number of known binding proteins. The requirement of an intact Mps3 SUN domain for nuclear migration during karyogamy is consistent with the role of UNC-84 and SUN-1 in nuclear positioning (, ). Mps3 SUN mutants also affect the second step of karyogamy. This may be due to a defect in SPB integrity; however, given that hSun1 is required for homotypic membrane fusion in vitro (), the requirement for Mps3 during nuclear fusion may reflect a more general role for SUN proteins as mediators of membrane fusion. Interestingly, the SUN domain region of Mps3 has been shown to interact with the DnaJ-like ER membrane chaperone Jem1, which is required for membrane fusion after mating (). SUN proteins also appear to have functions within the nucleus. Sad1 is required for formation of the meiotic bouquet during fission yeast meiosis (; ), and mammalian Sun1 proteins have Zn finger domains that may interact directly with chromatin (). In addition, Mps3 is required for the establishment of sister chromatid cohesion during S phase in budding yeast (), and one of our SUN domain mutants () is found in the allele that has defects in both centromere and telomere cohesion (). Additional nuclear functions could be one reason for the spindle checkpoint–dependent mitotic delay that we observed in , , and cells (unpublished data). How mutations in the SUN domain might affect nuclear functions of Mps3 is unclear at present. However, our identification of Mps3 as the budding yeast SUN protein should facilitate rapid identification and analysis of additional SUN binding partners and greatly enhance our understanding of SUN protein functions. All strains are derivatives of W303 () and are listed in Table S1 (available at ). The entire ORF of or was deleted in a diploid with or using PCR-based methods, and the strain was transformed with a centromeric -marked plasmid containing the corresponding wild-type ORF, sporulated, and dissected to generate and strains in both mating types. and alleles in pRS305 were created by oligonucleotide-directed mutagenesis and integrated into the locus in single copy in the appropriate deletion. was fused to 3xFLAG and was fused to 13xMYC using PCR. The was created by mutagenesis of (), and vectors containing both wild-type and mutant forms of were cut with BstZ17I to direct integration into the locus of SLJ1901 followed by loss of the covering plasmid on plates containing 5-fluorootic acid (5-FOA). and were amplified by PCR from pREP42- (a gift from I. Hagan, Paterson Institute for Cancer Research, Manchester, UK) and pSJ140 (), respectively, and cloned into a pRS306-based vector containing the promoter to create (pSJ379) and (pSJ146). and mutants, the SUN domain of and mutants and were amplified by PCR and cloned into pMAL-c2 (New England Biolabs, Inc.) or pQE10/pQE11 (QIAGEN) for expression in bacteria as MBP or 6xHis fusion proteins, respectively. 6xHis-Spc29 was purified from BL21(DE3) bacteria transformed with pQE11- (pSJ260) by metal affinity followed by anion exchange chromatography, and the C terminus of Mps3 was purified from BL21(DE3) pLysS bacteria transformed with pMALc2- (pSJ151) using amylose resin. Antibodies against purified 6xHis-Spc29 or MBP-Mps3-Ct were generated in rabbits (Animal Pharm) and purified according to the manufacturer's instructions on a Spc29 or Mps3 column created using the Sulfo-link kit (Pierce Chemical Co.). Mps2 antibodies were generated by injecting a KLH-coupled peptide corresponding to residues 212–233 of Mps2 into guinea pigs (Pocono) followed by affinity purification on MBP-Mps2 bound to nitrocellulose filters. The Mps3 SUN domain was also purified from BL21(DE3) pLysS bacteria transformed with pQE10- (pSJ413) by metal affinity chromatography, and 0.5 mg purified 6xHis-mps3SUN labeled with Alexa 680 as described previously (). Expression of MBP, MBP-Mps2 (pSJ426), and MBP-Kar1 (pSJ425) in BL21(DE3) pLysS cells was induced by the addition of 0.1 mM IPTG for 2 h at 23°C. Cells from 1 ml of culture were resuspended in 200 μl 2× SDS sample buffer and heated to 100°C for 5 min, 5 μl of MBP-Mps2 and MBP-Kar1 and 2 μl of MBP-only samples were separated by 8% SDS-PAGE, and proteins were electrophoretically transferred to nitrocellulose. Membranes were probed with 10 μg Alexa 680–labeled 6xHis-mps3SUN, and binding was assessed using the Odyssey imaging system (LiCor). Expression of bacterial proteins was confirmed by Western blotting with a 1:1,000 dilution of anti-MBP antibodies (New England Biolabs, Inc.). For bacterial pull-down assays, MBP, MBP-Mps2, and MBP-mps2 mutants were expressed in BL21(DE3) pLysS cells by the addition of 0.1 mM IPTG for 2 h at 23°C and cell pellets from 2 ml of culture were resuspended in 0.5 ml PBS containing 0.1% Triton X-100 and 1 mM PMSF, vortexed for 5 min at 4°C, and centrifuged at 14,000 for 10 min at 4°C. 100 μl of each extract was mixed with 1 μg of purified 6xHis-mps3SUN in a 500-μl immunoprecipitation that included 2 μl protein A–coated magnetic beads (Dynal) and 1 μl of affinity-purified anti-Mps2 antibodies. After a 1-h incubation a 4°C, immunoprecipitates were washed five times in PBS containing 0.1% Triton X-100 and an additional 250 mM NaCl. Spheroplast lysates were prepared from 100 OD midlog phase cells. Harvested cells were resuspended in 5 ml spheroplast buffer (50 mM Tris-HCl, pH 7.5, 1.2 M sorbitol, and 10 mM NaN) containing 40 mM β-mercaptoethanol and treated with 300 μg/ml zymolase 100T for 30 min at 30°C. Cells were then washed three times in spheroplast buffer, resuspended in 2.5 ml spheroplast lysis buffer (20 mM Hepes-KOH, pH 7.4, 100 mM K-acetate, 5 mM Mg-acetate, 1 mM EDTA, 1 mM PMSF, and 1 μg/ml each of pepstatin A, aprotinin, and leupeptin), and lysed by dounce homogenization. Lysates were transferred to ice and treated with 0.2 mg/ml dithiobis(succinimidyl)propionate for 10 min. Cross-linking was quenched by addition of 50 mM Tris-HCl, pH 8.0, and lysates were centrifuged at 13,000 for 15 min at 4°C. The pellet was resuspended in 2.5 ml solubilization buffer (spheroplast lysis buffer plus 1 M NaCl and 1% Triton X-100) and recentrifuged at 13,000 for 15 min at 4°C. The supernatant was added to 20 ml spheroplast lysis buffer immediately before immunoprecipitation. Ground lysates were prepared from 500 OD of midlog phase cells as described previously (). In brief, cells were frozen in liquid nitrogen and ground with a Retsch ball mill. Ground cell powder was allowed to thaw on ice and resuspended in 10 ml extraction buffer (25 mM Hepes-NaOH, pH 7.5, 300 mM NaCl, 0.1 mM EDTA, 0.5 mM EGTA, 2 mM MgCl, 0.5% Triton X-100, 1 mM DTT, 1 mM PMSF, and 1 μg/ml each of pepstatin A, aprotinin, and leupeptin). After homogenization with a Polytron 10/35 for 30 s, lysates were centrifuged at 3,000 for 10 min at 4°C, and the resulting supernatant was used for immunoprecipitations. 100 μl FLAG M2 (Sigma-Aldrich) or 9E10 conjugated (Santa Cruz Biotechnology, Inc.) beads were added to lysates to immunoprecipitate Mps3-FLAG or Mps2-13xMYC, respectively. After a 2-h incubation at 4°C, beads were washed five times in extraction buffer, and 1/10 of the bound protein and 1/1,000 of the input protein was analyzed by SDS-PAGE followed by Western blotting. The following primary antibody dilutions were used: 1:5,000 anti-MYC A14 (Santa Cruz Biotechnology, Inc.), 1:1,000 anti-FLAG (Sigma-Aldrich), 1:2,000 anti-Cdc31 (), 1:5,000 anti-Spc29, 1:2,000 anti-Mps3, 1:2,000 anti-Spc110 (a gift of T. Davis, University of Washington, Seattle, WA), 1:1,000 anti-GFP B34 (Covance), 1:2,500 anti-MBP (New England Biolabs, Inc.), and 1:10,000 anti–glucose-6-phosphate dehydrogenase (G6PDH; Sigma-Aldrich). Alkaline phosphatase–conjugated secondary antibodies were used at 1:10,000 (Promega). Approximately 5 × 10 midlog phase cells of opposite mating types grown at 23°C were mixed onto a 45-μm nitrocellulose filter, and filters were incubated on YPD plates for 5 h at 30°C. Cells were then stained with DAPI, and nuclear position was analyzed by fluorescence microscopy. Analysis of DNA content by flow cytometry, EM, and protein localization by indirect immunofluorescence and epifluorescence microscopy were performed as previously described (). Spc110 polyclonal antibodies were a gift from T. Davis and were used at a 1:2,000 dilution. Mps3 was localized using affinity-purified anti-Mps3 antibodies diluted 1:500 in PBS containing 3% BSA and 0.1% NP-40 followed by detection with Alexa 555–conjugated goat anti–rabbit secondary antibodies (Invitrogen) diluted 1:20,000 in the same buffer. Cells were examined with an Axioimager (Carl Zeiss MicroImaging, Inc.) using a 100× Plan-Fluar lens (NA = 1.45; Carl Zeiss MicroImaging, Inc.), and images were captured with a digital camera (Orca ER; Hamamatsu) and processed using Axiovision 4.0 (Carl Zeiss MicroImaging, Inc.). Homologues of Mps3 were detected through iterative searches of the nonredundant protein sequence database (National Center for Biotechnology Information) using the PSI-BLAST program (). Sequences corresponding to the SUN domain of Mps3 homologues and C termini of putative SUN binding proteins were aligned using ClustalW, and phylogenetic analysis was performed using the MEGA3 software. Fig. S1 shows that ts SUN domain mutants arrest in mitosis because of a defect in SPB duplication. Fig. S2 shows a phylogenetic analysis of SUN domains. Table S1 lists yeast strains. Table S2 shows SUN domain mutants. Table S3 gives copy number–dependent SUN domain mutants. Online supplemental material is available at .

MicroRNAs (miRNAs) are a class of small noncoding RNAs that are processed by Dicer from precursors with a characteristic hairpin secondary structure (). Hundreds of miRNAs have been identified from plants, animals, and viruses (miRBase; ). miRNAs are implicated in various cellular processes, such as cell fate determination, cell death, and tumorigenesis (for review see ). Many miRNAs are expressed in a tissue-specific manner (; ; ; ; ; ; ; ), suggesting a role of the miRNAs in the specification of the tissue during differentiation. Among the hundreds of miRNAs, only a small fraction have assigned target mRNAs or an established role. Valid target prediction is a major problem in the study of miRNAs. Although several algorithms for target prediction have been based on sequence similarity between targets and miRNAs (), the small size of the miRNAs and the tolerance for mismatches and bulges in the recognition sequence result in most of these algorithms' predicting too many targets. The mode of action of miRNAs on their targets is controversial. Classic results from miRNAs suggested that the miRNAs bind to their targets with imperfect complementarity and decrease the levels of encoded proteins without decreasing the target mRNA (; ). In contrast, target mRNA is cleaved specifically at the recognition site by siRNA (), many plant miRNAs (for reviews see ; ), and at least one animal miRNA (). In all cases where the target mRNA is cleaved, the interaction between the small RNA and the target mRNA is nearly perfect. Therefore, the degree of complementarity has been thought to be a major determinant in dictating whether a miRNA promotes mRNA degradation or inhibits protein synthesis. Although this hypothesis is supported by mutation analyses of miRNAs and their target mRNAs (; ), a recent report demonstrated that a miRNA can regulate the levels of several target mRNAs despite mismatches and bulges between the miRNA and the targets (). This was shown true for even and miRNAs (), which had been thought to block only the translational step. Differentiation down a specific lineage is characterized by the activation of tissue-specific transcription factors and modulation of gene expression. To study the role of miRNA in such a process and begin the process of identifying potential targets, we studied muscle differentiation using the C2C12 myoblast (MB) cell line as a model system (; ). Upon serum depletion, muscle-specific transcription factors such as myogenin are induced and many muscle genes are turned on. Subsequently, cells become elongated and fused to each other to form multinucleate myotubes (MTs). Another critical event during the differentiation process is a decrease in DNA synthesis and cell cycle arrest. We show here that miRNAs miR-206, -1, and -133 on their own change the gene expression profile of C2C12 toward the differentiated state and that miR-206 induces many of the markers of differentiation. The miRNAs regulate many target mRNAs as revealed by microarray screening. Antisense oligonucleotides to these miRNAs inhibit muscle differentiation and entry into cell quiescence. We predicted the putative direct targets of miR-206 by intersecting the mRNAs down-regulated in the microarray data with the computational prediction of targets based on sequence match to the miRNA. As an example of the utility of this approach, we identify four mRNAs, including that of the largest subunit (p180 subunit; Pola1) of DNA polymerase α (DNA pol α) as being directly regulated by miR-206. Two of the targets, including DNA pol α, are cleaved at multiple sites by the miRNA. The effect of miR-206 on DNA pol α, thereby DNA synthesis, is a new example of miRNA function connecting the cell quiescence event with the differentiation process. In addition, inhibitors of myogenic transcription factors are indirectly down-regulated by miR-206, thereby further promoting the differentiation process. Recently, another group reported a critical role of miR-1 and -133 in C2C12 differentiation (). Together, we can conclude that all three miRNAs induced during C2C12 differentiation are very important for myogenesis. Previous results suggested that miR-1, -133, and -206 are expressed in muscle and heart (; ; ; ). Consistent with this, RNase protection assays () showed that miR-1 and -133 are abundant in skeletal muscle and heart and that miR-206 is specifically abundant in skeletal muscle. miR-1 and -206 have an 18/21 match in sequence with each other and complete identity in the first eight nucleotides (miRBase) that constitute the seed sequence for target recognition (). C2C12 mouse MBs can be induced to differentiate into MTs by serum depletion, as indicated here by the induction of myogenin, cell cycle inhibitor p21, and myosin heavy chain (MHC) with a constant level of Cdk4 as a loading control (). miR-1, -133, and -206 were not expressed in undifferentiated C2C12 but were induced during muscle differentiation (; ), whereas two other miRNAs ( and miR-125b) were expressed constantly throughout the differentiation process. Using this in vitro differentiation of C2C12, we studied whether the miRNAs play an active role during skeletal myogenesis. To investigate the function of these miRNAs in myogenesis, we used double-stranded RNA duplexes with miRNA sequence that mimic miRNA function (; ). Transfection of these duplexes into C2C12 reduced the luciferase expression when the cognate target site is placed at the 3′UTR of luciferase gene, validating the experimental system. The cross-reactivity between miR-1 and -206 can be explained by their similar sequences. To ensure that the transfected miRNAs are not being assayed at supraphysiological levels, we measured the levels of miR-133 and -206 after transfection into MBs and compared the levels of the same miRNA after induction of differentiation into MTs. The levels of the transfected miRNAs were comparable to the levels of the naturally induced miRNAs (). We began our studies with miR-206 because of the skeletal muscle–specific expression of miR-206 () and because of its similarity to miR-1. miR-206 transfection advanced MHC expression after changing to differentiation medium (DM; ), whereas miR-133 did not (Fig. S1 A, available at ) under the same experimental condition. Upon transfection in the continued presence of serum, miR-206 markedly up-regulated the percentage of cells expressing MHC and the muscle-specific transcription factor myogenin, relative to cells at comparable density transfected with GL2 control (; and Fig. S1, B and C). In addition, 28% of MHC-positive cells are multinucleated after miR-206 treatment, whereas no multinucleated cells were detected in GL2 control ( and not depicted). Therefore, physiological levels of miR-206 induce skeletal myogenesis from C2C12 even without serum depletion. To measure the extent of differentiation and to extend the studies to the other muscle-specific miRNAs, microarray experiments were used to measure changes in global gene expression profile after muscle differentiation by serum deprivation (DM). The microarray profile after differentiation was compared with that obtained after miRNA transfection in the presence of serum (growth medium [GM]). We focused our analysis on the ∼100 genes that were induced or repressed the most after C2C12 differentiation to MTs by serum depletion (). Genes up- or down-regulated after muscle differentiation were changed similarly by transfection of the miRNAs (in GM), with miR-206 and -1 having similar effects and miR-133 affecting a slightly different subset of genes. Given the importance of the first eight nucleotides of miRNA in target interaction (), the identical seed sequences between miR-1 and -206 explain why the two miRNAs had almost identical changes in gene expression. These data suggest that the individual miRNAs change the repertoire of expressed genes in a direction mimicking that seen during muscle differentiation. Further analysis of the microarray data will be addressed in the next section. To determine whether the miRNAs were essential for differentiation, C2C12 cells were treated with 2′--methyl antisense oligonucleotides against the miRNAs (; ) and then induced to differentiate by serum depletion. Inhibition of the miRNAs reduced the number of MHC-positive cells () and decreased MHC levels in an immunoblot (), with little effect when only miR-133 was inhibited. BrdU immunostaining shows that a significant fraction of the cells remain in active DNA synthesis when the miRNAs are inhibited (42% for the anti-miR mix relative to 26% for the anti-GL2 control; P = 0.0071; ). Inhibition of the miRNAs prevented the elongation of MTs seen during differentiation (). The cells remained relatively short and thick even after expression of MHC: 86% of MHC-positive cells in the anti-GL2 control sample were elongated, as opposed to only 25% in the anti-miR mix sample. Although inhibition of miR-1 and -206 (without inhibition of miR-133) was sufficient to decrease MHC expression and derepress DNA synthesis (), it was not sufficient to inhibit cell elongation (not depicted), suggesting that miR-133 might have specific targets relevant to MT elongation. Collectively, the results suggest that the three miRNAs are required for complete differentiation to muscle, with miR-206 (and miR-1) being particularly important for induction of cell quiescence. Unlike plant miRNAs, animal miRNAs are believed to repress protein synthesis without changing mRNA levels. Yet, the microarrays () revealed a large number of mRNA changes after miRNA introduction. A significant number of up-regulated genes included muscle-specific genes such as myosin light chain (phosphorylatable, fast skeletal muscle), myosin light polypeptide 1, troponin T1 (skeletal, slow), troponin I (skeletal slow 1), myomesin 2, and titin (; Table S1, available at ). Because miRNAs are expected to be repressive, the up-regulation of genes is most likely due to indirect effects after the primary differentiation-inducing stimuli of the miRNAs. In contrast, the list of down-regulated mRNAs ( and Table S2, available at ) might include direct targets of the miRNAs if, as in plants, the miRNAs promote target RNA cleavage. To identify such direct targets, we intersected the list of genes down-regulated in the microarrays with the list of predicted genes with miRNA target sites (generated using the miRanda program []). We focused our analysis on miR-206 because of its similarity with miR-1 and its pronounced effect on muscle differentiation and cell proliferation ( and ). On this intersection list, we were pleased to find the largest subunit of DNA pol α (Pola1; available from GenBank/EMBL/DDBJ under accession no. ), the replicative polymerase expected to be very important for cell proliferation. miR-206 alone is sufficient to decrease Pola1 at the mRNA and protein levels as early as 24 h after transfection (). During differentiation of MBs to MTs, Pola1 mRNA and protein were also decreased (). The Pola1 promoter requires the activity of E2F, a transcription factor that is repressed by pRb during myogenesis (; ), so we wondered how much of the down-regulation of Pola1 mRNA during differentiation was dependent on miR-206. Antisense to miR-206 significantly delayed the down-regulation of Pola1 mRNA (). This result is in agreement with the increase of BrdU-positive cells and decrease of MHC-positive cells when miR-206 was inhibited by antisense oligonucleotide (). The antisense to miR-206 does not, however, completely block differentiation (). Therefore, residual inhibition of Cdk2 kinase, hypophosphorylation of Rb, and repression of E2F probably account for the eventual repression of Pola1 in the presence of antisense to miR-206, albeit with delayed kinetics. Together, these results suggest that miR-206 is important for the early down-regulation of Pola1 during differentiation. Pola1 has two potential target sites of miR-206 (; Fig. S5, available at ; and see ) in the 4991–5345 segment of the 3′UTR. Fusion of a luciferase reporter to this segment rendered luciferase repressible by miR-206 (). The level of repression by cotransfected miRNAs is comparable to the two- to threefold repression by miRNAs in all published experiments where a reporter gene is fused to naturally occurring target 3′UTRs (; ; ). As a negative control, no repression was seen upon insertion of another part of the 3′UTR of Pola1 without any predicted target sites for miR-206 (DNA pol α 3985–4657). Point mutations in the target sites in pol α 4991–5345 revealed that the one encompassing nucleotides 5007–5029 (M1; ; Fig. S5; and see ) is necessary for the repression by miR-206, whereas the second site at 5090–5114 (M2) is not required. Therefore, miR-206 directly down-regulates Pola1 mRNA, a down-regulation that probably contributes to the prompt suppression of cell proliferation seen during differentiation. Pola1 is similarly down-regulated by miR-1 but not as much by miR-133 ( and Fig. S2). However, miR-206 is likely to be more important than miR-1 for this function during C2C12 differentiation because it is present at a much higher level than miR-1 (). We measured whether down-regulation of DNA synthesis is an early event after miR-206 introduction. In good agreement with the Pola1 decrease, DNA synthesis was significantly inhibited by 24 h after transfection of miR-206, eventually leading to a decrease in cell proliferation that became evident at 72 h (). Because cell cycle arrest is a critical step during muscle differentiation, we wondered whether the DNA synthesis inhibition induced by miR-206 leads to the cell cycle arrest. Upon serum depletion of MBs to form MTs, the Cdk inhibitors p21 and p27 are induced and Cdk2 kinase activity is decreased accompanied by a reduction in the amount of the faster moving form of the Cdk2 that is phosphorylated on its activating site, Threonine160 (; ). miR-206 transfection induces p21 and p27 as early as 24 h after transfection. The FACS profile for DNA content () and the activity of Cdk2 kinase (), however, demonstrate that G1 accumulation is not evident at 24 h after transfection, even though DNA synthesis was already reduced significantly (). At 96 h after transfection, the miR-206 transfected cells eventually accumulate in G1 with repressed Cdk2 activity (), consistent with a G1 arrest (not depicted). Therefore, miR-206–mediated inhibition of DNA synthesis precedes the cell cycle arrest. Next, we tested whether cells synchronized in the cell cycle responded to miR-206 with an arrest in G1. C2C12 cells were synchronized at the G1/S boundary by thymidine aphidicolin block (, middle). During the synchronization, we introduced miR-206 or GL2. Upon removal of aphidicolin, GL2-treated cells were released from G1 through the subsequent stages of the first cell cycle (2–8 h) and proceeded through the second cell cycle (8 h and later; ). miR-206–treated cells were released into S phase with delayed kinetics (2–8 h), and the delay was even more marked in the second cell cycle ( [bottom] and E [>8 h]). The time of second G1 (, 10–12 h) corresponds to 34–36 h after transfection. Because the kinetics of target repression after miR-206 transfection is likely to be slow, it is entirely expected that the cell cycle effect is more marked as the cell proceeds through successive cell cycles. Besides Pola1, we tried to find additional targets of miR-206 by two approaches. First, we reasoned that inhibitors of differentiation could be putative targets of the miRNAs. MyoR (musculin) and Id 1–3 antagonize the action of the bHLH (basic helix-loop-helix) myogenic transcription factors like MyoD, whereas Hedgehog-interacting protein (Hhip) inhibits a factor called Hedgehog that promotes differentiation (; ; ). All these mRNAs were down-regulated after transfection of miR-206 (; and Table S2). However, Id1-3 and MyoR are unlikely to be regulated directly by miR-206, as indicated by the failure of miR-206 to repress luciferase reporter fused to these genes in transient transfection assays () and the absence of any predicted target site. Hhip has a predicted target site (Table S3, available at ), but is also not a direct target of miR-206 in the luciferase fusion assay (). Indirect repression of these inhibitors of differentiation upon introduction of miR-206 into C2C12 cells indicates how the chain of events initiated by direct down-regulation of key targets by a miRNA can have a profound effect on the differentiation program. The second approach to finding direct targets was to focus on genes that were decreased in the microarrays after miRNA introduction and contained putative target sites in the 3′ UTR. These include a Rac1 interacting protein named butyrate-induced transcript 1 (B-ind1), brain-derived neurotrophic factor, the gap-junction protein connexin43 (Cx43), nuclear receptor coactivator 5, AMP activated protein kinase, Hhip, junction adhesion molecule 4, and an adiponectin-related protein expressed in microglia called monocyte-to-macrophage differentiation-associated protein (Mmd). All of these mRNAs were decreased by normal differentiation and by transfection of the miRNA (). Although all of them have predicted target sites (Table S3), fusion to luciferase reporter reveals that only B-ind1, Cx43, and Mmd are direct targets of miR-206 like DNA pol α (). Of the four direct targets, Pola1 has already been shown to be important for cell quiescence. B-ind1 might influence differentiation through its interaction with the G protein Rac-1. Down-regulation of B-ind1 mRNA was inversely related to the up-regulation of miR-206 during differentiation and was abrogated when miR-206 was inhibited by anti–miR-206 (). We do not have any evidence, however, that B-ind1 down-regulation is critical for differentiation. The known functions of Cx43 and Mmd do not suggest that their repression is critical for differentiation. Having confirmed a critical role of miR-206 in the down-regulation of at least two targets, Pola1 and B-ind1 mRNA ( and ), we next turned our attention to the mechanism of the down-regulation. miRNAs, unlike siRNAs, are not perfectly matched to their targets and have been reported to repress their targets at the protein synthesis level (; ; ). Plant miRNAs (for reviews see ; ) and at least one animal miRNA () with perfect match to the target mRNA induce cleavage of the target. Short RNAs have also been shown to alter the chromatin state and thus repress the transcription of some genes (for review see ). As was the case for Pola1, the addition of B-ind1 mRNA sequence downstream from a luciferase reporter confers suppression by miR-206, suggesting that B-ind1 is a direct target of miR-206 (). Point mutations indicate that at least two predicted target sites contribute to the down-regulation (, M1 and M2; and Fig. S3, available at ). Nuclear run-on experiments were performed to determine whether there is any transcriptional repression of these direct target genes induced during differentiation or after introduction of miRNAs miR-1 or -206. Transcriptional activity of Pola1 was too low to be measured by nuclear run-on (unpublished data), but that of B-ind1 was measured and found to not decrease after C2C12 differentiation or after the introduction of miR-1 and -206 (). The Northern blots that accompanied this experiment show that the steady-state mRNA level of B-ind1 is repressed by at least fivefold (), suggesting that the miRNAs repress B-ind1 at the posttranscriptional level. To test if the posttranscriptional reduction is due to the cleavage of mRNA by the miRNAs, we performed modified rapid amplification of cDNA ends (RACE)–PCR () to map the 5′ ends of potential cleavage fragments of B-ind1 and Pola1. Differentiated MTs (not depicted) and C2C12 MBs transfected with miR-206 and -1 () generated more cleavage products from the B-ind1 transcript compared with the MBs or cells transfected with either GL2 control or miR-133. Similar results were obtained from Pola1 (unpublished data), supporting the hypothesis that the mRNAs are cleaved in the presence of the miRNAs. The sizes of RACE-PCR products derived from the B-ind1 () or Pola1 (not depicted) after transfection of miR-206 are distinct from those obtained after transfection of miR-1 ( and Fig. S3). As miR-1 and -206 recognize similar target sites and similarly decreased B-ind1 (), the different cleavage sites between the two miRNAs suggest that the miRNAs do not select the cleavage sites. To support this idea, sequencing of the RACE-PCR products revealed that the 5′ ends of cleavage fragments mapped at multiple sites in the B-ind1 mRNA ( and Fig. S3) and were not confined to putative target sites of miR-206. B-ind1 mRNA was cleaved similarly during differentiation (Fig. S4, available at ). DNA pol α mRNA was cleaved by transfection of miR-206, and these cleavage sites also did not match with putative target sites ( and Fig. S5). In contrast, siRNAs cleave the target mRNA at the center of RNA duplex by RISC (RNA-induced silencing complex; ; ). A control experiment with miR-206 siRNA against a perfectly matching site in luciferase 3′UTR showed that cleavage in this case was exactly at the siRNA target site (unpublished data). Thus, although the down-regulation of B-ind1 and Pola1 mRNA depends on miR-206 target sites with multiple mismatches and bulges, and although the target mRNA is cleaved, the 5′ ends of the cleavage fragments are distributed and do not match the target sites. We show that miRNAs not only promote differentiation of MBs in vitro but are required for differentiation. Individual miRNAs like miR-206 produce large changes in gene expression at the mRNA level that mimic changes seen upon differentiation, and the down-regulation of a direct target of miR-206, Pola1, might contribute to the cell cycle quiescence upon differentiation. In addition, several inhibitors of myogenic transcription factors are repressed after miR-206 transfection, leading to further amplification of the prodifferentiation function. At least two of the direct targets are cleaved at the posttranscriptional level, but the 5′ ends of the cleavage fragments suggest the involvement of exonucleases, a finding that has implications for the proposed involvement of processing bodies (P-bodies) in miRNA function. Finally, we describe an efficient strategy for identifying direct targets of miRNAs by intersecting mRNAs down-regulated by miR-206 in microarray screens with computationally predicted targets and then checking the intersection list by transient transfection and luciferase fusion assays. Transfection of miR-1 in a nonmuscle cell has been reported to alter the gene expression profile toward that of muscle (), but there was no indication of the magnitude of the change or the importance of the miRNAs in the physiology of muscle differentiation. Our results suggest that muscle-specific miRNAs promote skeletal myogenesis through direct down-regulation of several target genes and are required for normal differentiation. The microarray result indicates that the gene expression changes effected by miR-206 or -1 are closer to that of normal differentiation compared with that effected by miR-133. Because miR-206 is specifically expressed in skeletal muscle, whereas miR-1 is present in both skeletal and cardiac muscle, it would be interesting to compare miR-206 and -1 in greater detail to identify overlapping or unique roles in myogenesis. A block to DNA synthesis through the direct down-regulation of DNA pol α precedes the cell cycle withdrawal and is the first demonstration of miRNAs directly affecting DNA replication. This result is supported by the recent observation that overexpression of miR-1 decreased the pool of proliferating ventricular cardiomyocytes in transgenic mice (). It is not clear yet whether down-regulation of the three other targets identified, B-ind1, Cx43, and Mmd, contributes in any way to the muscle differentiation program. To fully understand the role of miR-206 in skeletal myogenesis, we expect to uncover more targets of the miRNA in the future. Combinatorial down-regulation of multiple targets may be necessary to fully mimic the effect of miR-206 transfection into MBs. miR-135b, -338 (), and -181 () are also muscle specific, suggesting that there may be additional miRNA involved in the regulation of myogenesis. miR-181 (), -1, and -133 () have already been shown to play a role in muscle differentiation by repressing their cognate targets, and our results add miR-206 to the list. We expect that extensive cooperation between several miRNAs and several transcription factors is necessary to effect the complete differentiation program. The mode of action of miRNAs is still being debated. The multiple mismatches with target sites are believed to prevent cleavage of the target by a siRNA-like mechanism. Our results indicate, however, that at least for four endogenous targets, the repression by a miRNA is accompanied by a decrease in mRNA levels. For one of these direct targets (B-ind1), we rule out any transcriptional component, and for two of them (B-ind1 and Pola1), we show that the repression is accompanied by generation of mRNA cleavage fragments with multiple 5′ ends that do not map to the essential target sites matching the miRNA sequence. The exact 5′ ends vary and do not overlap when B-ind1 is down-regulated by the transfection of miR-206 or -1 or serum deprivation. This variability of the 5′ ends of the cleavage fragments suggests that they may be created by exonuclease activity after an initial endonucleolytic cleavage. RNA P-bodies have recently been proposed as sites where miRNAs sequester their target mRNAs from the protein-synthesis machinery (; ). P-bodies contain exonucleases, so that the degradation of target mRNAs observed in this paper could be the end stage of the sequestration of targets in P-bodies. By serving as a common platform for both processes, P-bodies might explain how miRNAs can in some cases repress protein levels but not mRNAs and in other cases promote the degradation of mRNAs. An important point is that changes in mRNA profile during differentiation, thought to be mainly the province of transcription factors, can also be dictated by miRNA-mediated posttranscriptional cleavage of mRNAs. The approach described here of intersecting mRNAs discovered to be down-regulated in microarray screens after miRNA transfection with mRNAs that have computationally predicted matches to miRNAs provides a high yield of potential targets. Fusion of candidate 3′UTRs to luciferase can then be used to quickly narrow down the list of targets to those that are directly down-regulated by miRNAs. Combining this strategy with in vitro differentiation systems is likely to yield many miRNA targets critical for differentiation down specific lineages. We are aware, of course, that this strategy might overlook targets that are exclusively down-regulated at the protein level, such as HoxA11 by miR-181 () and histone deacetylase by miR-1 (). If, however, down-regulation of protein synthesis by miRNAs involves sequestration of the target mRNAs to P-bodies, we suspect that many direct targets of miRNAs will eventually be down-regulated at the mRNA level and thus be revealed in the microarray screens. Because different computational algorithms predict different sets of targets for a given miRNA, a future extension of this work will be to intersect the microarray data with the output from other target prediction programs. The ability of miRNAs to affect many mRNAs is similar to the ability of transcription factors to regulate many promoters simultaneously. Thus, just like transcription factors, we predict that miRNAs will induce complex changes in rate-limiting steps of cell metabolism and thus have a profound effect on the differentiation program. An additional point of interest is the interaction between these miRNAs and transcription factors known to be involved in muscle differentiation. For example, MyoD has already been suggested to be involved in the induction of miR-1 in cardiac myogenesis (). We suggest here that one of the muscle-specific miRNAs, miR-206, indirectly down-regulates Id1-3 and MyoR, inhibitors of myogenic transcription factors like MyoD. Such a positive feedback loop between the myogenic transcription factors and the muscle-specific miRNAs will be expected to push the equilibrium toward differentiation and could be very significant for holding muscle cells in a permanently differentiated state. C2C12 (American Type Culture Collection; ) mouse skeletal MBs were maintained at subconfluent densities in DME supplemented with 20% FCS (GM). Myogenic differentiation () into MT was induced by changing subconfluent cells to DME containing 2% heat-inactivated horse serum (DM). Antibodies to p21 (C-19), p27 (C-19), Cdk2 (M2), Cdk4 (C-22), Pola1 (G-16), and myogenin (mouse mAb F5D) were purchased from Santa Cruz Biotechnology, Inc. Mouse mAb against MHC and anti–β-actin antibody were obtained from Sigma-Aldrich. Double-stranded RNA oligonucleotides containing on one strand the sequences of miR-1 (5′-uggaauguaaagaaguaugua-3′), -133 (5′-uugguccccuucaaccagcugu-3′), -206 (5′-uggaauguaaggaagugugugg-3′), and -125b (5′-ucccugagacccuaacuuguga-3′), as well as GL2 () were synthesized by Invitrogen. Complementary sequence of each oligonucleotide was designed to produce a two-nucleotide overhang at both the 3′ ends of the duplex. 2′--methyl antisense oligonucleotides against miR-1, -133, and -206 and GL2 were synthesized by Dharmacon RNA Technologies. Transfection into C2C12 was performed with Lipofectamine 2000 reagent (Invitrogen), combined with 267 nM of each siRNA duplex or with 133 nM of 2′--methyl antisense oligonucleotide. Immunostaining was performed as described previously (). Unless otherwise specified, all manipulations were at room temperature. Cells on sterile glass coverslip were fixed with 2% formaldehyde in PBS for 15 min and were permeabilized with 0.2% Triton X-100 and 1% normal goat serum (NGS) in ice-cold PBS for 5 min. After blocking with 1% NGS in PBS two times for 15 min, incubation with primary antibody (1:400 in 1% NGS) for 1 h was followed by the FITC-conjugated anti–mouse IgG (dilution 1:500; DakoCytomation) for 1 h. Control experiment with H-3 (anti-hexahistidine) as a primary antibody ensured no cross-reactivity of the secondary antibodies. In case of sequential anti-BrdU probing, the above steps are repeated, but treatment with 1.5 N HCl for 30 min was included before the incubation with Alexa Fluor 594–conjugated anti-BrdU (Invitrogen) antibody (1:100 in 1% NGS) for 1 h. After washes, nuclei were counterstained with DAPI (H-1200; Vector Laboratories) for 1 min before mounting. Images were visualized using a microscope (Microphot-SA; Nikon) with 60× magnitude, captured using a camera (UFX-DX; Nikon), and processed using SPOT (version 3.5.4 for MacOS; Diagnostic Instruments) software. 50 μg of protein was immunoprecipitated with anti-Cdk2 antibody overnight and then pulled down on protein G–Sepharose beads for 2 h. The beads were washed three times with lysis buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 0.1% NP-40, 5 mM EDTA, 50 mM NaF, 1 mM sodium vanadate, and protease inhibitors) and twice with kinase buffer (50 mM Hepes-NaOH, pH 7.4, and 25 mM MgCl). The beads were then incubated with 25 μl kinase reaction mixture (50 mM Hepes-NaOH, pH 7.4, 25 mM MgCl, 0.5 mM DTT, 50 μM ATP, 5 μCi γ-[P]ATP, and 2 μg histone H1). The reactions were incubated at 30°C for 30 min and stopped by the addition of 12.5 μl of 3× SDS sample buffer. Cell growth was measured with CellTiter 96 nonradioactive cell proliferation assay kit (Promega). DNA synthesis was measured as described previously () with minor modifications. Cells were treated with 1 μCi of methyl-[H]thymidine in 0.25 ml of medium in a 24-well dish. Incubation at 37°C for 1 h was followed by washing with PBS two times. Radioactively labeled cells were treated with ice-cold stop solution containing 10% (vol/vol) TCA and 0.2 M sodium pyrophosphate for 20 min, washed in 95% ethanol two times, and solubilized in 1% (vol/vol) SDS and 10 mM NaOH. Solubilized samples were transferred onto paper (3MM; Whatman) and dried under an infrared lamp. The radioactivity was measured with a liquid scintillation counter. RNase protection assay was performed with miRNA detection kit (Ambion) with minor modifications. 20 μl of reaction contained the indicated amount of RNA, 0.6 μg of yeast tRNA, and the probe RNA, which was in vitro transcribed, labeled with α-[P]UTP using miRNA probe construction kit (Ambion), and purified from polyacrylamide gel. Digestion products were resolved by electrophoresis in an 18% polyacrylamide gel with 7 M urea. C2C12 cells (MB) were induced to MT or were transfected six times at 24-h intervals with GL2 or miR-1, -133, or -206, respectively. Total RNAs from each sample were isolated on day 7 by using Trizol reagent. Subsequent steps for the hybridization to Affymetrix GeneChip Mouse Genome 430 2.0 Array (containing ∼45,000 transcripts) were done according to standard Affymetrix protocols. The array data were analyzed with the GeneChip Operating Software. The varied genes were identified by comparing the two samples in fold change and/or “present” and “absent” calls. The results of the microarray screen can be found on Gene Expression Open Source System () and are freely available to the public. For nuclear run-on assay, 0.5 μg of each DNA fragment was slot blotted onto a positively charged Nylon membrane (Nytran; Schleicher & Schuell). Nucleus isolation and run-on transcription reaction were performed as described previously () with modifications. After the transcription reaction, Trizol reagent was added, and RNA probe was prepared according to the manufacturer's instructions. Membranes were probed with P-labeled run-on RNA for 16 h at 68°C in hybridization buffer containing 0.25 M NaHPO, pH 7.2, 1 mM EDTA, and 7% (wt/vol) SDS. Membranes were washed twice in 2× SSC at room temperature for 15 min each, followed by a wash in 0.2× SSC and 1% (wt/vol) SDS at 65°C for 20 min. To map the cleavage sites of B-ind1 and Pola1 mRNA, RLM-PCR was performed with GeneRacer kit (Invitrogen) with modifications (). Total RNAs from C2C12 cells transfected with GL2 or with miR-206 were ligated with RNA adaptor without any pretreatment (). B-ind1– or Pola1-specific primers for the cDNA synthesis, the first-round PCR, and the second-round PCR are indicated in Figs. S3, S4, and S5. The PCR products were separated on 1.1% agarose gel. Each lane on the agarose gel was quantified with ImageQuant 5.2 software and normalized to the B-ind1 mRNA level in the input mRNA. Titration of input mRNA ensured that the amount of RLM-RACE PCR product in a given sample is proportional to the input mRNA. The PCR products were subcloned into PCR4-TOPO vector and sequenced. For ease of subsequent subcloning, pRL-CMV(MCS) was modified from the original vector pRL-CMV (Promega) by inserting a synthetic linker with various restriction sites into the XbaI restriction site downstream of the ORF of Renilla () luciferase gene. At 20 h after transfection of the miRNAs and the luciferase plasmids into C2C12 cells, luciferase assays were performed with Dual-luciferase reporter assay system (Promega) per the manufacturer's instructions. Luminescent signal was quantified by luminometer (Monolight 3020; BD Biosciences). Each value from Renilla luciferase construct (Rr) was first normalized to the firefly () luciferase assay value (Pp) from the cotransfected pGL3-control vector (Promega). Each Rr/Pp value was again normalized to the Rr/Pp value from miR-125b as a control. Each value is a mean of three transfections. The luciferase constructs containing various putative target genes include B-ind1 (1248–2334 of accession no. , available from GenBank/EMBL/DDBJ), brain-derived neurotrophic factor (700–1756 of accession no. ), Cx43 (82–2700 of accession no. ), Hhip (800–2982 of accession no. ), junction adhesion molecule 4 (91–1491 of accession no. ), and Mmd (301–2510 of accession no. ). The indicated regions were PCR amplified and inserted into pRL-CMV. Fig. S1 shows an examination of muscle markers and cell density after transfection of the miRNAs. Fig. S2 shows that miR-1, but not -133, directly down-regulates DNA pol α through the M1 site. Figs. S3, S4, and S5 show cleavage sites on B-ind1 or Pola1 mRNA. Tables S1 and S2 show the 30 most up- and down-regulated genes, respectively, in C2C12 transfected with miR-206 duplex. Table S3 shows that the potential target sequences for miR-206, predicted by miRanda, exist in several putative target genes. Online supplemental material is available at .

Dynamic control of the actin cytoskeleton is essential for cell polarization, migration, and division (; ). Proper actin remodeling involves the activation of different eukaryotic nucleation factors that can generate specific types of new actin filaments. The actin-related protein 2/3 (Arp2/3) complex nucleates new actin filaments while remaining anchored to the sides of existing filaments, creating branched actin networks (). Formin proteins nucleate unbranched filaments, creating actin cables, contractile rings, and stress fibers (; ; ; ). Although the Arp2/3 complex is known to play important roles in processes such as cell polarization, motility, and vesicle trafficking, much less is know about the biological functions of formins, especially in higher eukaryotes. Similarly, although molecular regulation of the Arp2/3 complex by upstream signals is understood in some detail (), much less is known about the regulation and function of the much larger formin family. The formins are defined by a conserved C-terminal formin homology (FH) 2 domain that mediates effects on actin (; ; ; ; ). The FH2 domain functions as a dimer and has varying effects in different formins, including actin filament nucleation, filament severing, and barbed-end binding with elongation (anticapping effect) or without elongation (capping effect; ; ; ; ; ; ; ). The function of the conserved N-terminal region found in most formins is still unclear. In vitro and in vivo studies with mouse diaphanous 1 (mDia1) have demonstrated a role for its N terminus in mediating the autoinhibition of FH2 activity (see the following paragraph; ; ; ; , ). Other studies with the fungal formins Bni1p, Fus1, and SepA have implicated their N termini in regulating localization in vivo (; ; ). For both Bni1p and Fus1, perturbation of cellular localization disrupts biological activity (; ). It is unclear whether the effects of the N terminus on FH2 activity and cellular localization are linked or whether the N terminus simply serves different roles in different formins. Elucidating the general regulation of formin localization and activity in higher eukaryotes is important for understanding the biological activities of these molecules. In the diaphanous-related formin (DRF) subfamily of formins, a short (∼20 residue) conserved region called the diaphanous autoregulatory domain (DAD) follows the FH2 domain in sequence (; ). In mDia1, the DAD binds to the N terminus to inhibit the actin assembly activity of the FH2 domain through an unknown mechanism (; ). The DAD-binding element in the N terminus is called the diaphanous inhibitory domain (DID; ). Rho binds to the mDia1 N terminus through a region that spans a portion of the DID and an adjacent sequence element termed the G region (for GTPase binding; ; ). Rho binding destabilizes interactions between the N and C termini, leading to partial activation of the mDia1 FH2 domain (; ). The mDia1 N terminus forms a dimer through its dimerization domain (DD) and coiled coil (CC) region (; ). Based on sequence similarity, the G-DID-DD-CC architecture of the mDia1 N terminus is likely to hold for other DRF family members (). The formin FHOD1 is also likely to be autoinhibited based on cytoskeletal effects of mutant proteins studied in vivo (). However, the generality of FH2 domain autoinhibition through a DAD–DID interaction and activation by Rho GTPases has yet to be established directly for DRFs other than mDia1. FRLα (formin-related gene in leukocytes α) is a macrophage-enriched DRF whose FH2 domain can nucleate new actin filaments and sever existing filaments in vitro (; ). FRLα has been reported to bind to Rac and modulate cell adhesion, migration, and survival, but its molecular function in macrophages has not been explored in detail (). Because of its unique expression profile, we hypothesized that understanding FRLα may reveal new aspects of formin biology in higher eukaryotes as well as elucidate general principles governing the biological and biochemical regulation of formin proteins. We find that the FRLα N terminus binds to the C terminus in a DAD-dependent manner and inhibits the actin assembly activity of the FH2 domain. For both FRLα and mDia1, autoinhibition also regulates a previously unrecognized plasma membrane localization activity of the N-terminal domains. This activity has both GTPase-dependent and -independent components. For FRLα, active Cdc42 relieves the autoinhibition of FH2 activity in vitro and membrane localization activity in macrophages. Knockdown experiments reveal that FRLα is required for efficient Fc-γ receptor–mediated phagocytosis, which is consistent with its role as a macrophage-enriched Cdc42 effector. Live cell imaging shows that FRLα is transiently recruited to the phagocytic cup in a Cdc42-dependent manner. These studies reveal a general mechanism of DRF regulation in which autoinhibition controls actin assembly activity and cellular localization. In addition, we identify a new biological function for DRFs in the immune system of higher eukaryotes. Most previous in vivo studies of DRF N-terminus function have used fragments based on GTPase-binding domain (GBD)–FH3 sequence elements (; ). However, recent structural studies of the mDia1 N terminus have demonstrated that these sequence elements do not demarcate structural elements of the protein. Therefore, the results from previous studies using constructs representing divided structural domains are difficult to interpret (; ). In this study, we have used predicted structural domains of FRLα based on sequence alignments with mDia1 to examine the function of its N- and C-terminal domains (). Our N-terminal construct (residues 1–450) contains the regions of FRLα that align with the G, DID, and DD elements of mDia1 (FRLα appears to lack a CC region), whereas our C-terminal construct (residues 612–1,094) contains the FH2 and DAD domains (). Pull-down assays show that the immobilized GST-tagged N terminus can interact directly with the untagged C terminus (). Binding is also observed when immobilized maltose-binding protein (MBP)–tagged C terminus is used to pull down GST-tagged N terminus (). The DAD motif is required for binding because a C-terminal construct mutated at a conserved leucine residue within the DAD motif (L1062D) cannot interact with the N-terminal fragment (). Mutation of the same residue in mDia1 (unpublished data) or mDia2 () also prevents interactions between N- and C-terminal fragments. As previously described, the FH2 domain of FRLα can stimulate actin filament assembly, although it is ∼50-fold less potent than the FH2 domain of mDia1 (; ; ). The N terminus inhibits the activity of the C terminus in a dose-dependent manner with an IC of ∼100 nM () but has no effect on the L1062D DAD mutant or FH2-independent actin assembly (). Thus, the biochemical activity of the FRLα FH2 domain is regulated by high affinity DAD-mediated autoinhibitory interactions with the N terminus. We transfected RAW 264.7 cells, a mouse macrophage cell line, with GFP-tagged FRLα. Full-length FRLα is cytoplasmic and is excluded from the nucleus, as shown by confocal images of cells coexpressing FRLα-GFP and monomeric RFP (mRFP), a uniformly distributed fluorescent control (). In contrast, the N-terminal fragment of FRLα is located primarily at the plasma membrane, suggesting that in the full-length protein, the cellular localization of the N terminus may be controlled by binding to the C terminus (). To test this hypothesis, we determined the cellular localization of FRLα-GFP proteins with mutations that would impair the N + C interaction. Introduction of the L1062D DAD mutation results in plasma membrane localization of full-length FRLα (). In mDia1, an L260E mutation in the N-terminal DID region blocks DAD binding without affecting Rho interactions (). The analogous mutation in full-length FRLα (V281E) causes the protein to localize at the plasma membrane (). Quantitation of the ratio of GFP fluorescence intensity at the plasma membrane to the intensity in the cytosol shows that all membrane-localized FRLα constructs exhibit 2–4.5-fold enrichment at the plasma membrane, whereas cytosolic FRLα constructs have a ratio near 1 (). The degree of enrichment is largely independent of the overall expression level for all constructs shown here (). To test whether interactions between the N terminus and DAD are sufficient to mediate the control of FRLα cellular localization, we constructed a truncated FRLα protein called mini-FRLα, which lacks the FH2 domain. Mini-FRLα contains the G, DID, and DD regions tethered by a (Gly-Gly-Ser) linker to the DAD-containing C-terminal 101 residues (). When purified from overexpressing bacteria, this protein is dimeric as assessed by multiwavelength static light scattering (measured mol wt = 148.2 ± 0.2 kD compared with 71.9 kD for the monomer; unpublished data), which is the expected molecular organization of full-length FRLα. When expressed in macrophages, mini-FRLα–GFP is located in the cytoplasm of transfected cells (). In the crystal structure of the DAD–DID complex from mDia1, the DAD forms an amphipathic helix with its hydrophobic face contacting the DID. Mutation of conserved residues on the hydrophobic face significantly decreases DAD–DID affinity, and the mutation of hydrophilic residues on the solvent exposed face has little effect on this interaction (; ; ). With mini-FRLα, the mutation of any one of the conserved hydrophobic residues I1058, I1059, or L1062 on the hydrophobic face of the analogous predicted helix results in plasma membrane localization of the protein (). In contrast, the mutation of residues G1053 or E1056 on the hydrophilic face has no effect on localization (). Furthermore, the FRLα N terminus is cytoplasmic in cells coexpressing the DAD-containing C-terminal 100 residues of the protein but is membrane localized in cells coexpressing an L1062D mutant C terminus (Fig. S1, available at ). These results suggest that binding of the hydrophobic face of the amphipathic DAD helix is necessary and sufficient to block membrane localization mediated by the FRLα N terminus. Our combined in vitro and in vivo data show that the interactions between the N and C termini are mutually autoinhibitory, with the N terminus blocking the actin assembly activity of the FH2 domain and the C terminus blocking the plasma membrane localization activity of the N terminus. We next sought to determine whether Rho GTPases could relieve the autoinhibition of FRLα. Previously reported pull-down assays indicated that an N-terminal fragment of FRLα can specifically associate with Rac1 in a nucleotide-independent manner (). However, we could not detect an interaction between recombinant Cdc42 or Rac1 loaded with a GTP analogue, β,γ-imidoguanosine 5′-triphosphate (GMPPNP), and GST-tagged FRLα N terminus using similar assays (unpublished data). Unexpectedly, high concentrations of Cdc42-GMPPNP can relieve N-terminal inhibition of the FRLα FH2 domain (). The Cdc42 effect is dose dependent and saturates at ∼400 μM GTPase, a concentration that slightly decreases the activity of the isolated C terminus (). Under these conditions, although Cdc42-GMPPNP exhibits low potency, it essentially fully relieves inhibition by the N terminus. Cdc42-GDP is a much weaker activator compared with Cdc42-GMPPNP (). Rac1-GMPPNP is unable to relieve the autoinhibition of FRLα actin assembly activity () even at concentrations (∼400 μM) at which Cdc42-GMPPNP is maximally effective (). A V161D mutation in the mDia1 DID decreases Rho binding without affecting DAD binding (). The analogous T126D mutation in the FRLα N terminus does not affect binding to the C terminus () or inhibition of FH2-mediated actin assembly () but blocks the ability of Cdc42-GMPPNP to relieve inhibition (). When cotransfected with constitutively active Cdc42, both FRLα-GFP and mini-FRLα–GFP localize at the plasma membrane (). The localization of FRLα is nucleotide and GTPase specific, as the protein remains primarily cytosolic in cells coexpressing dominant-negative Cdc42 or constitutively active Rac1 or RhoA ( and I). In all cases, expression of the different GTPase mutants was confirmed by immunostaining (Fig. S2, available at ). Like our GFP fusions of FRLα, endogenous FRLα is primarily cytosolic in the absence of Cdc42. In cells transfected with constitutively active Cdc42, endogenous FRLα becomes enriched at the plasma membrane (Fig. S2). Because Cdc42 is membrane anchored through prenylation of its C-terminal CAAX motif, we tested whether Cdc42 was directly recruiting FRLα to the plasma membrane of RAW cells. The T126D mutant FRLα N terminus, which is unresponsive to Cdc42 in actin assembly assays, is enriched at the plasma membrane but to a lesser degree than the analogous wild-type fragment (). As with our other FRLα constructs, the degree of enrichment does not vary systematically with varying expression levels (). Thus, Cdc42 binding likely contributes to but is not necessary for membrane enrichment of the FRLα N terminus. However, signaling through Cdc42 does appear to be required for this membrane localization activity because coexpression of the FRLα N terminus with the GBD of the Wiskott-Aldrich syndrome protein (WASP; a reagent that at high levels should sequester the active GTPase) causes the FRLα fragment to be cytoplasmic ( and S1). This relocation to the cytoplasm is not observed when the FRLα N terminus is coexpressed with a mutant WASP-GBD that cannot bind Cdc42 ( and S1). These results suggest that membrane localization activity of the FRLα N terminus derives, in part, from direct binding to Cdc42 and also from an interaction with some other membrane-associated factor that lies downstream of Cdc42. Although our data suggest a direct interaction between Cdc42 and the N terminus of FRLα, we have been unsuccessful in demonstrating a direct interaction in pull-down or fluorescence-based binding assays using bacterially expressed recombinant proteins (unpublished data). However, when cotransfected into mammalian 293T cells, constitutively active Cdc42 and the wild-type FRLα N terminus can be coimmunoprecipitated (, lanes a–c). The N terminus does not coimmunoprecipitate with constitutively active Rac1 or RhoA (, lanes g–i and j–l, respectively), suggesting specificity for Cdc42 over these other two Rho family members. One caveat here is that despite significant effort, the level of Rac expression was always lower than that of Cdc42 in these experiments. However, along with the biochemistry and aforementioned localization data, these results do support a much greater role for Cdc42 than Rac in regulating FRLα. The T126D mutant N terminus does not coimmunoprecipitate with Cdc42, suggesting that in vitro and in cells, the Cdc42 responsiveness of FRLα is mediated by the same set of interactions (, lanes d–f). Furthermore, full-length FRLα and mini- FRLα do not coimmunoprecipitate with Cdc42 even though they localize at the plasma membrane when coexpressed with constitutively active Cdc42 (, lanes m–o and p–r, respectively). These results suggest that Cdc42 has higher affinity for the FRLα N terminus as compared with an autoinhibited FRLα molecule, which is consistent with the idea that Cdc42 must compete with the autoinhibitory interactions present in full-length or mini-FRLα. As in FRLα, a GFP fusion of the N terminus of mDia1(residues 1–570) is localized at the plasma membrane (). Both full-length mDia1-GFP and mini-mDia1–GFP, analogous to our mini-FRLα construct, are cytoplasmic and are excluded from the nucleus of RAW cells (). Thus, autoinhibition controls the localization of mDia1 in a manner analogous to FRLα. Also, like FRLα, the membrane localization activity in the mDia1 N terminus has GTPase-dependent and -independent components. The V161D mutant N-terminal fragment of mDia1 (analogous to the aforementioned FRLα T126D mutant) is strongly defective in Rho binding (). This protein localizes to the plasma membrane in macrophages but to a lesser degree than the wild-type N terminus (). Interestingly, the degree of membrane enrichment of this protein decreases as expression level increases (), suggesting that the GTPase-independent membrane interaction is saturable. A nearly identical behavior is also observed for a fragment of the mDia1 N terminus lacking the G region entirely (; mDia1_ΔG_N-term), which also is severely impaired in Rho binding (; ). Analogous fragments of FRLα are not expressed in bacteria (in contrast to mDia1_ΔG_N-term, which expresses well; unpublished data), are expressed only weakly and inconsistently in macrophages, and were not analyzed in detail. Together, our data suggest that for both FRLα and mDia1, plasma membrane localization of the N terminus is mediated by GTPase-dependent and -independent interactions and that these are controlled by autoinhibitory contacts within the formins. We next sought to determine whether FRLα is involved in Fc-γ receptor–mediated phagocytosis. This process is Cdc42 dependent and requires extensive actin rearrangement at the cell surface of macrophages (). Using two different sets of siRNAs directed against FRLα, we achieved significant knockdown of endogenous FRLα in RAW macrophages (). These reduced Fc-γ receptor–mediated phagocytosis of IgG-opsonized RBCs by 45 ± 4 and 29 ± 2%, respectively, compared with cells expressing control siRNAs (). Overall cell appearance, adhesion, and filamentous actin content judged by phalloidin staining are all normal in the siRNA-treated cells (Fig. S3 A, available at ). Moreover, complement receptor-mediated phagocytosis, a process that proceeds through a Rho–mDia1 pathway (), is also unaffected by siRNA treatment (Fig. S3 B). These results demonstrate that the knockdown of FRLα does not lead to generalized cytoskeletal defects in RAW cells and establish an important role for FRLα specifically in Fc-γ receptor–mediated phagocytosis. In cells coexpressing FRLα-GFP and mRFP, we used ratiometric imaging to determine specific sites of FRLα accumulation. During phagocytosis, FRLα-GFP accumulates in the developing phagocytic cup and appears to concentrate at the tips of the extending pseudopods ( and Videos 1–3, available at ). FRLα-GFP accumulation is maximal immediately before pseudopod fusion, which we set as a reference time point to compare multiple phagocytic events, and rapidly dissipates as the RBC is internalized (). This spatial and temporal pattern is reproducibly observed in different transfected cells; additional phagocytic events are shown in Fig. S4. A control GFP construct fused to the C-terminal 101 residues of FRLα does not accumulate at the phagocytic cup (). Consistent with previous findings that Rho is not involved in Fc-γ receptor–mediated phagocytosis (), mDia1-GFP does not localize to the cup during this process (). The T126D mutant of FRLα shows a consistent modest increase in localization at the cup relative to mDia1 and the FRL C terminus, but its enrichment does not change appreciably over the course of internalization (). This enrichment is likely caused by the ability of the T126D mutant N terminus to interact with membranes (), which accumulate at the cup as a result of its topology. Importantly, the wild-type protein shows much different behavior, with a greater degree of accumulation and substantial changes over the course of the process, suggesting recruitment to the cup by an active mechanism. Active Cdc42 has also been shown to accumulate transiently at the extending pseudopod tips of the phagocytic cup (). We monitored the localization of active Cdc42 during phagocytosis using the WASP-GBD fused to GFP, a reagent with high (19 nM) affinity for active Cdc42 (; ; ; ). We note that at high levels of expression, this reagent blocked phagocytosis (unpublished data), but, at lower levels, it had no effect on phagocytic efficiency, and we could use it to visualize active Cdc42 during this process. As a control, we used a mutant GBD-GFP construct that is unable to bind Cdc42. We find that Cdc42, like FRLα, is recruited transiently to the phagocytic cup and that its accumulation peaks immediately before pseudopod fusion (; and Videos 4 and 5, available at ). The similarities in timing and localization of active Cdc42 and FRLα support the idea that FRLα function during phagocytosis is linked to Cdc42 signaling. We have generated N-terminal fragments of FRLα based on recent crystal structures of the mDia1 N terminus (; ) to investigate the mechanisms of autoinhibition and activation in FRLα. Based on fragmented DID constructs, earlier work suggested that FRLα may bind Rac (). However, using constructs containing intact structural domains that are biochemically well behaved, we now show that FRLα is a Cdc42-specific effector. Several lines of evidence support this assertion. Cdc42 is able to relieve the autoinhibition of FH2-mediated actin assembly by the wild-type N terminus but not by the T126D mutant N terminus. Mutation of the analogous residue in mDia1 prevents direct Rho binding and Rho-mediated activation (). Furthermore, the wild-type FRLα N terminus coimmunoprecipitates with active Cdc42 from cell lysates, but the point mutant does not. In addition, the timing and localization of Cdc42 and wild-type FRLα are largely coincident during phagocytosis, and the recruitment of FRLα to the phagocytic cup is blocked by the T126D mutation. Similar biochemical, coimmunoprecipitation, and localization assays fail to show a link between Rac1 activity and FRLα function. However, the interaction between Cdc42 and FRLα appears to be of low affinity, and the interaction between them may be modulated by cellular factors. The functional significance of the N terminus of DRFs has been thought to be limited to mediating the autoinhibition of the C terminus (; ; ). We now show for the DRFs FRLα and mDia1 that interactions between the N and C termini mediate the mutual autoinhibition of DRF localization and biochemical activity. The membrane localization activity of the DRF N termini derives, in part, from interactions with Rho GTPases. However, our results suggest that the N termini can also bind an additional membrane-associated factor. This factor binds the N termini competitively with the DAD and, thus, should contribute to DRF activation. Depending on the structural details of the interaction, the factor may act cooperatively with GTPases to drive DRF membrane localization and activation. Cdc42 plays an essential role in Fc-γ receptor–mediated phagocytosis (). Formins have not previously been implicated in this process, although mDia1 has recently been shown to be involved in complement-mediated phagocytosis, a Rho-mediated process (). Our identification of FRLα as a Cdc42 effector led us to test whether this macrophage-enriched DRF is involved in Fc-γ receptor–mediated phagocytosis. Two previous studies on Fc receptor–mediated phagocytosis reported that dominant-negative Cdc42 inhibits phagocytosis by ∼40–70% (; ), which is similar to what we observe for our strongest FRLα knockdown (45%). The consistent degree of inhibition by dominant-negative Cdc42 and FRLα RNAi suggests that FRLα is an important effector of Cdc42-mediated signaling during Fc-γ receptor–mediated phagocytosis. It is useful to distinguish three phases of Fc-γ receptor–mediated phagocytosis: pseudopod extension, pseudopod fusion, and particle internalization (; ; ). Time-lapse imaging shows that fluorescently labeled actin accumulates at the base and extending pseudopods of the phagocytic cup (; ). At closure, actin forms a sphere around the phagosome (). After closure, actin dissipates from the phagosome starting at the base, resulting in a short-lived actin cap structure at the top of the phagosome (). The actin architectures at each phase of phagocytosis and the signaling pathways that control their formation have yet to be determined. Recent studies have started to define the spatio-temporal patterns and functional roles of different actin regulatory proteins during Fc-γ receptor–mediated phagocytosis, focusing mostly on Cdc42 and Rac (; ; ; ; ). The model that has emerged from these studies suggests that active Cdc42 localizes primarily at the tips of the extending pseudopods, mediating their extension and dissipating rapidly after their fusion. In contrast, active Rac1 is localized at the base and throughout the pseudopods during extension and then concentrates at the base to control fusion and internalization. Thus, Cdc42 and Rac likely play different roles during Fc-γ receptor–mediated phagocytosis. In contrast to the Rho GTPases, the actions of downstream signaling molecules during phagocytosis have been studied less extensively. WASP family proteins are important for Arp2/3 complex activation during cell migration (), and static images reveal that the WASP and Arp2/3 complex are recruited to the phagocytic cup (; ). However, the detailed spatio-temporal patterns of these molecules during phagocytosis have not been reported. Fc-γ receptor–mediated phagocytosis represents one of the first isolated cytoskeletal systems in higher eukaryotes in which both Arp2/3 () and formins (this study) have essential functions. Why might two different actin nucleation machines be required for Fc-γ phagocytosis? Although both the Arp2/3 complex and FRLα can nucleate actin filaments, there are important differences in their resulting actin networks. Arp2/3 networks are branched and sensitive to capping proteins, and, therefore, the behavior of the Arp2/3 network depends on the balance of capping and nucleation. FRLα-generated filaments are unbranched and resistant to capping proteins. Based on its linkage to Cdc42, FRLα function may be important primarily for pseudopod extension, whereas the Arp2/3 complex may be required for the other phases of phagocytosis. In addition to nucleating new filaments, the FRLα FH2 domain can also sever actin filaments (). This activity could lead to an increase in ATP-capped filament ends, which appear to be preferential sites of Arp2/3-mediated branching (). Indeed, FRLα may not serve a nucleation function during phagocytosis at all but rather may facilitate pseudopod extension by functioning as an anticapping or severing protein to modify the Arp2/3 actin network. Currently, the specific biochemical and biological roles of FRLα and Arp2/3 complex during Fc-γ receptor–mediated phagocytosis remain elusive. Understanding the detailed molecular mechanisms of phagocytosis will shed light on this essential function of the immune system and will help determine the relative contributions of different actin regulatory elements to cytoskeletal dynamics in higher eukaryotes. Full-length FRLα (residues 1–1,094), the FRLα N terminus (residues 1–450), the FRLα C terminus (residues 612–1,094), and the FRLα ΔFH2–C terminus (residues 994–1,094) were cloned into pET vectors containing N-terminal tobacco etch virus protease-cleavable GST or MBP affinity tags. FRLα cDNA was provided by T. Watanabe (Kyushu University, Fukuoka, Japan). Full-length mDia1 (resides 1–1,255), the mDia1 N terminus (residues 1–570), and the mDia1 ΔG–N terminus (residues 131–570) were cloned by PCR. All FRLα and mDia1 constructs were cloned into a modified pCMV-Script vector containing C-terminal EGFP or mRFP. Template cDNA for mRFP was provided by R.Y. Tsien (University of California, San Diego, San Diego, CA). Mini-FRLα (residues 1–536 and 994–1,094 tethered by a GGSGGS linker) and mini-mDia1 (residues 1–457 and 1,168–1,255 tethered by a GGSGGS linker) were generated using overlap extension PCR. Myc-tagged GTPase constructs were provided by M.H. Cobb (University of Texas Southwestern, Dallas, TX). The WASP-GBD comprises residues 230–288 of the human protein. The WASP-GBD mutant was generated by scrambling the amino acid sequence of the internal Cdc42/Rac interactive binding motif (IGAPSGFKHVSHVGWD mutated to KWDVPIHHGGFGASVS), which is necessary for high affinity interactions with Cdc42 (). All proteins were expressed in the strain BL21(DE3). FRLα N-terminal constructs were expressed as GST fusion proteins, whereas C-terminal constructs were expressed as MBP fusions. N-terminal constructs were purified using glutathione–Sepharose beads (GE Healthcare), anion exchange chromatography, and gel filtration. C-terminal constructs were purified by cation exchange chromatography and gel filtration. Before some assays, the affinity tags were cleaved using tobacco etch virus protease, and the tag-free proteins were purified either with anion or cation exchange chromatography. Cdc42 (residues 1–179) was prepared and loaded with GMPPNP as previously described (; ). Actin was purified from rabbit skeletal muscle and labeled with pyrene as described previously (). 6 0 0 p m o l G S T - F R L α N t e r m i n u s w a s l o a d e d o n t o g l u t a t h i o n e – S e p h a r o s e b e a d s ( G E H e a l t h c a r e ) a n d i n c u b a t e d w i t h 6 0 0 p m o l F R L α C t e r m i n u s i n 2 0 0 μ l o f b i n d i n g b u f f e r ( 2 0 m M T r i s , p H 8 . 0 , 1 0 0 m M N a C l , a n d 1 m M D T T ) . B e a d s w e r e w a s h e d t h r e e t i m e s w i t h 2 0 0 μ l o f t h e b i n d i n g b u f f e r c o n t a i n i n g 0 . 1 % T r i t o n X - 1 0 0 , w a s h e d t w i c e w i t h 2 0 0 μ l o f t h e b i n d i n g b u f f e r w i t h o u t d e t e r g e n t , a n d a n a l y z e d b y S D S - P A G E a n d C o o m a s s i e b l u e s t a i n i n g . T h e s a m e p r o t o c o l w a s a l s o p e r f o r m e d w i t h 6 0 0 p m o l M B P – C t e r m i n u s o n a m y l o s e r e s i n ( N e w E n g l a n d B i o l a b s , I n c . ) i n c u b a t e d w i t h 6 0 0 p m o l o f G S T – N t e r m i n u s . sub #text RAW 264.7 cells were purchased from the American Type Culture Collection and maintained in DME with 10% FBS and 1 mM sodium pyruvate (Invitrogen). 10 cells were transfected with 5 μg DNA using LipofectAMINE 2000 (Invitrogen). For siRNA transfection, 2 × 10 cells were transfected with siRNAs using 1.5 μl siQuest reagent (Mirus Bio Corporation) and 250 nM siRNAs. Under these conditions, nearly 100% of RAW cells were transfected by siRNAs as judged by experiments using LabelIT Fluorescent RNA delivery controls (Mirus Bio Corporation). To determine the efficiency of FRLα knockdown, cell lysates were prepared 48 h after transfection. Western blots were performed using anti-FRLα antibodies provided by H.N. Higgs (Dartmouth College, Hanover, NH). HEK293T cells were maintained in 10% calf serum (American Type Culture Collection) and 1 mM penicillin/streptomycin (Invitrogen). For coimmunoprecipitation, 10 cells plated in 10-cm dishes were transfected with 10 μg DNA using the calcium phosphate precipitation method. HEK293T cells were cotransfected with expression vectors for each formin-GFP and myc-GTPase construct. 36 h after transfection, cell lysates were prepared in lysis buffer (20 mM Tris, pH 7.4, 100 mM NaCl, 1 mM DTT, 1 mM MgCl, 0.2% NP-40, 1 mM PMSF, 1 mM benzamidine, 1 μg/ml leupeptin, 1 μg/ml antipain, 0.5 μg/ml pepstatin, 20 mM NaF, and 0.5 mM NaVO). The cleared lysate was incubated with anti-GFP antibody (Invitrogen) for 1 h at 4°C. Immune complexes were precipitated with UltraLink immobilized protein A/G beads (Pierce Chemical Co.) and analyzed by Western blotting. 24 h after transfection, cells were replated onto 12-mm glass coverslips (Fisher Scientific), allowed to adhere for 6–8 h, and were fixed. Imaging was performed on a laser scanning microscope (LSM510 META; Carl Zeiss MicroImaging, Inc.) using a 63× oil immersion objective (Carl Zeiss MicroImaging, Inc.). Images were analyzed using Slidebook software (Intelligent Imaging Innovations). Membrane and cytosolic intensities were determined by calculating the mean fluorescence in masks placed either at the membrane or in the cytoplasm. The expression level was expressed as the sum of the fluorescence intensity across a given confocal plane divided by the cross-sectional area. 24 h after transfection, cells were replated onto 35-mm glass-bottom dishes and allowed to adhere for 6–8 h. For imaging, cells were incubated at 37°C on a heated stage connected to a humidifier module and covered with an optically clear foil cover (Carl Zeiss MicroImaging, Inc.). To start phagocytosis, 1 ml of 50-fold diluted opsonized RBCs prepared as described previously () was added dropwise to the macrophages. For each imaging run, 61 time points were collected. For each time point, a four-plane z stack of GFP and mRFP fluorescent images spaced 1.5 μm apart was acquired along with one differential interference contrast (DIC) image (50-ms exposure). A total of nine images per time point was collected at maximum speed, which, on our system, resulted in one time point per 13.7 s. Imaging was performed using a 63× oil immersion objective (Carl Zeiss MicroImaging, Inc.) on an inverted microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) equipped with a 75-W Xenon lamp and CCD camera (Sensicam; PCO Computer Optics). Fluorescence images were acquired using FITC and Cy3 filter sets (Chroma Technology Corp.), 200-ms exposure times, 2 × 2 binning, and a 10% neural density filter. For each phagocytic event, the optimal plane (where the phagoyctic cup was in best focus) was determined at each time point. These planes were stitched together to create a 2D time-lapse series. For each event, the time point at which pseudopod fusion occurred, as seen in the fluorescent images, was set as the zero time reference. Only events in which the macrophage and RBC could be observed 8 time points before fusion and 23 time points after were further analyzed. Thus, for each event we have analyzed 32 time points, which is sufficient for complete analysis of the dynamics that we observe. To analyze fusion protein accumulation during phagocytosis, we averaged the GFP/mRFP ratio at the site of phagocytosis (R) and normalized it to the GFP/mRFP ratio in the cytoplasm (R) in a manner similar to previously described methods (). R was calculated by creating a mask around the region of contact between the macrophage and the RBC. An R mask, which was placed in the cytoplasm, was also created. For each pixel in each mask at each time point, we calculated the ratio of GFP intensity to mRFP intensity. We then averaged the ratios for each mask to generate the GFP/mRFP ratio at the site of phagocytosis or in the cytoplasm. 24 h after transfection, RAW cells were replated onto 12-mm glass coverslips. 48 h after transfection, cells were washed and returned to the 37°C incubator in 0.2 ml of buffer with divalents (20 mM Hepes, pH 7.4, 125 mM NaCl, 5 mM KCl, 5 mM glucose, 10 mM NaHCO, 1 mM KHPO, 1 mM MgCl, and 1 mM CaCl). After 1 h, 20 μl of 10-fold diluted opsonized RBCs were added. For complement phagocytosis, cells were treated with 100 nM PMA 30 min before the addition of RBCs. After 1 h, cells were washed, incubated in distilled water to lyse extracellular RBCs, and fixed. Cells were permeabilized with PSG buffer (PBS, 0.01% saponin, 0.25% gelatin, and 0.02% NaN). Intracellular RBCs were stained with 1:200 diluted FITC-conjugated anti–rabbit antibodies (Jackson ImmunoResearch Laboratories). Imaging was performed on an inverted microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) using a 63× oil immersion objective (Carl Zeiss MicroImaging, Inc.). The number of internalized RBCs was counted using the FITC fluorescence signal, and the number of macrophages was counted using cellular autofluorescence. For complement phagocytosis, internalized RBCs were counted using DIC. Each experiment was performed in duplicate, and a minimum of 300 macrophages was counted for every condition. Table S1 provides the mean GFP/GFP ratios for all constructs and conditions tested. Fig. S1 shows that membrane localization of the FRLα N terminus is inhibited by DAD binding in trans. Fig. S2 shows the localization of endogenous FRLα and expression of myc-tagged Rho GTPase mutants in RAW cells. Fig. S3 shows the characterization of cells treated with FRLα siRNA, and Fig. S4 shows additional images of FRLα recruitment to the phagocytic cup. Videos 1 and 2 are DIC and fluorescence time lapses, respectively, of FRLα-GFP–expressing macrophages undergoing Fc-γ receptor–mediated phagocytosis. Video 3 shows a magnified phagocytic cup from Video 2. Videos 4 and 5 are DIC and fluorescence time lapses, respectively, of GBD-GFP–expressing macrophages undergoing Fc-γ receptor–mediated phagocytosis. Online supplemental material is available at .

About 30% of all proteins synthesized in an cell execute their cellular function in extracytoplasmic compartments like the inner membrane, the periplasm, or the outer membrane. The distribution of newly synthesized proteins to their final location is therefore a crucial issue, and bacteria have evolved multiple targeting pathways for selecting and properly directing these extracytoplasmic proteins (). The majority of these proteins engage the SecY translocon for exiting the cytoplasm and are targeted to this protein-conducting channel by two distinct pathways. Secretory proteins, e.g., proteins that are translocated across the inner membrane to reside in the periplasm or in the outer membrane are transported posttranslationally by the bacterial-specific SecA–SecB pathway (). In this pathway, the chaperone SecB binds to most secretory proteins before they are transferred to the ATPase SecA, which is distributed between the cytoplasm and the membrane (). Membrane binding of SecA is mediated by its affinity for both phospholipids and the SecY translocon (). Importantly, only the translocon bound SecA is thought to interact with the signal sequence of secretory protein in a highly specific manner (). Finally, the secretory protein is threaded through the SecY channel by repeated ATP-dependent conformational changes of SecA (; ). Different to secretory proteins, bacterial inner membrane proteins are recognized cotranslationally by the universally conserved signal recognition particle (SRP; ). SRP binds to the signal anchor sequence of a nascent membrane protein when it emerges from the ribosome and subsequently targets the SRP–ribosome-associated nascent chain (RNC) complex to the membrane via a GTP-dependent interaction with FtsY, the bacterial SRP receptor (SR). At the membrane, the RNC is transferred from the SRP–SR complex to the SecY translocon, and the membrane protein is cotranslationally inserted into the lipid bilayer. This step requires the sequential dissociation of SRP from both the signal anchor sequence and SR and is regulated by the GTPase activities of both Ffh, the protein component of the bacterial SRP, and FtsY (; ). In eukaryotic cells, the SRP-interacting SRα subunit of the SR is tethered to the membrane via its specific interaction with the membrane-integral SRβ subunit (). As SRβ is suggested to interact with the Sec61 channel, the eukaryotic SRP might target its substrates directly into close vicinity of the Sec61 translocon (). One particular facet of the bacterial SRP pathway is that the bacterial SR consists of only the FtsY protein, which is homologous to the eukaryotic SRα (; ). Although the bacterial SR lacks a membrane integral subunit, ∼30% of the cellular FtsY is stably associated with the membrane. Importantly, only the membrane bound FtsY appears to be able to induce the dissociation of SRP from the signal anchor sequence (). The association of FtsY with the membrane is thought to involve both protein–lipid contacts and protein–protein contacts (). Recent data indicate that FtsY is at least transiently associated with the SecY translocon (), and it has been proposed that the SecY translocon provides one binding site for FtsY at the membrane. In this study, we have further analyzed the membrane association of FtsY. Our data indicate that FtsY binds to two discrete binding sites at the membrane and that both interactions are stabilized by guanosine 5′-[β,γ-imido] triphosphate (GMP-PNP), a nonhydrolysable GTP analogue. Binding to the first site requires only the NG-domain of FtsY and results in a proteinase K–resistant conformation of FtsY. The second binding site is provided by the SecY translocon, to which FtsY binds in a carbonate-resistant manner. This interaction results in the formation of a 400-kD FtsY translocon complex and is stabilized by the N-terminal A-domain of FtsY, which probably serves as a transient lipid anchor. Despite the lack of a transmembrane domain, FtsY has been shown to be partially resistant toward alkaline carbonate extraction (; ), a method that is routinely used to differentiate between membrane-inserted and soluble proteins (). We used this method to determine whether membrane binding of FtsY was influenced by nucleotides. When in vitro–synthesized FtsY was incubated with inner membrane vesicles (INVs) and subsequently extracted with 0.2 M NaCO, we observed a small but reproducible increase in carbonate resistance (, compare lanes 2 and 4). Strikingly, when FtsY was incubated with INVs in the presence of the nonhydrolysable GTP analogue GMP-PNP, a significantly larger portion of FtsY was bound to the membrane in a carbonate-resistant manner (, compare lanes 4 and 8). This was not due to a GMP-PNP–induced aggregation of FtsY, because in the absence of INVs, the addition of GMP-PNP did not change the amount of FtsY present in the pellet fraction after carbonate extraction (, lane 6). The increase in carbonate resistance in the presence of INVs was strictly dependent on GMP-PNP; the addition of GTP, GDP, ATP, or adenyl-5′-yl imidodiphosphate (AMP-PNP) did not significantly influence the ability of FtsY to associate with the membrane in a carbonate-resistant manner (). The data described in the previous paragraph could suggest that blocking the GTPase activity of FtsY by GMP-PNP stabilizes its association with the membrane. We therefore analyzed the carbonate resistance of FtsY mutants that were affected in their GTPase activities. The FtsY mutant FtsY(G455V) has been shown to bind GTP but is unable to hydrolyze it efficiently (). In contrast to wild-type FtsY, in vitro–synthesized FtsY(G455V) exhibited a strong carbonate resistance even in the absence of GMP-PNP (). When assayed in the presence of GMP-PNP, the carbonate-resistant interaction with the membrane increased even further (). This could reflect the fact that the GTPase activity of this particular mutant is only partially blocked () but could also suggest that additional GTPases, like SRP, are involved in the GMP-PNP–dependent carbonate-resistant interaction of FtsY with the membrane. This was addressed by analyzing the carbonate resistance of two additional FtsY mutants: FtsY(G455W) and FtsY(A334W). Both mutants exhibit no significant GTP hydrolysis (<5% of wild type; ); they differ, however, in their ability to form an FtsY–SRP complex (). FtsY(G455W) is like the aforementioned FtsY(G455V) mutant, defective in SRP–FtsY complex formation, whereas complex formation is not impaired in the FtsY(A334W) mutant (). Nevertheless, the G455W and A334W mutants exhibited the same carbonate-resistant phenotype, which was almost completely independent of GMP-PNP (). These data suggest that the GTPase activity of FtsY is directly correlated with its ability to interact with the membrane in a carbonate-resistant manner. We have recently shown that FtsY transiently interacts with the SecY translocon (). Whether this interaction was responsible for the GMP-PNP–induced carbonate-resistant membrane binding of FtsY was analyzed by using INVs derived from translocon mutants. The SecY40 mutant has been shown to specifically block the integration of SRP-dependent membrane proteins (), and it has been proposed that it is particularly the SecY–FtsY interaction that is impaired in this mutant (). In agreement with this hypothesis, FtsY did not gain carbonate resistance in the presence of INVs upon the addition of GMP-PNP (). In contrast, if INVs derived from the mutant were tested, the GMP-PNP–induced association of FtsY with the membrane was comparable to wild-type INVs. The mutation blocks a functional SecA–SecY interaction () without disturbing the SRP/FtsY-dependent steps of membrane protein integration (; ; ). Importantly, no GMP-PNP–induced carbonate resistance of FtsY was observed in SecE-depleted INVs (, CM124), in which SecY is rapidly degraded (). These data are consistent with the idea that GMP-PNP stabilizes a specific interaction between the SecY translocon and FtsY, which, as a consequence, acquires carbonate resistance. To obtain evidence for a possible complex formation between FtsY and the SecY translocon upon addition of GMP-PNP, we used blue native PAGE (BN-PAGE), a technique that has been widely used for studying membrane bound protein complexes (). We first determined whether FtsY would assemble into differently sized complexes in the presence or absence of GMP-PNP. His-tagged FtsY was in vitro synthesized, purified by metal affinity chromatography, and incubated with INVs. Subsequently, the membranes were solubilized with dodecyl-β--maltoside and separated by BN-PAGE. In the absence of GMP-PNP, we observed a strong 200-kD radiolabeled FtsY complex (, lane 1). However, if FtsY was incubated with INVs in the presence of GMP-PNP, the 200-kD band became significantly weaker and instead a new complex of ∼400 kD appeared (, lane 2). The appearance of the 400-kD FtsY complex was GMP-PNP specific and not observed in the presence of GTP, GDP, ATP, or AMP-PNP (unpublished data). To analyze whether the 400-kD FtsY complex represented the carbonate-resistant state of FtsY, membranes were incubated with FtsY in the presence or absence of GMP-PNP and carbonate extracted. Only the carbonate-resistant material was then solubilized and separated on BN-PAGE. Importantly, the 400-kD FtsY complex that was formed in the presence of GMP-PNP was resistant to carbonate extraction (, lane 4), whereas the 200-kD FtsY complex was carbonate sensitive, i.e., only barely detectable (, lane 3). Thus, the carbonate-resistant interaction of FtsY with the membrane is reflected by the formation of the 400-kD complex. The formation of the 400-kD complex was further analyzed in a time-course experiment. Membrane vesicles were incubated with radiolabeled FtsY and GMP-PNP, and at different time points after GMP-PNP addition, samples were solubilized and separated on BN-PAGE. shows that the 200-kD complex was converted within minutes into the 400-kD complex. Because FtsY does not acquire carbonate resistance in SecE-depleted membranes (), one would expect that the 400-kD FtsY complex is not formed in the presence of these membranes. Indeed, the conversion of the 200-kD complex into the 400-kD complex was only barely detectable in the presence of SecE-depleted membranes (), suggesting that the formation of the 400-kD complex required the presence of the SecY translocon. The very small amount of the 400-kD complex observed in these membranes is probably due to residual amounts of the Sec translocon, which, as an essential protein, cannot be completely depleted. Independently of whether GMP-PNP had been added, two weaker radiolabeled bands below the 140-kD protein marker band were detectable (, lanes 1 and 2); these were not carbonate resistant (, lanes 3 and 4) and were also detected in the SecE-depleted INVs (). Like the dominating 200- and 400-kD complexes, they were recognized by α-FtsY antibodies (unpublished data). Despite its predicted size of 54 kD, FtsY migrates in SDS-PAGE as a doublet band of 100 and 75 kD. The 75-kD band represents a truncated FtsY derivative that lacks the first N-terminal 14 amino acids (unpublished data; ). We currently do not know whether the two bands recognized below the 140-kD marker band correspond to these 100 and 75 kD species or whether they reflect additional proteolytic fragments, like the 56-kD fragment observed by . Importantly, these data are all consistent with a GMP-PNP stabilized FtsY translocon complex of 400 kD. Different oligomers of the SecYEG complex have been reported to exist in membranes, with a preponderance of a SecYEG complex of ∼230 kD (). We confirmed the presence of this SecYEG complex by performing a BN-PAGE of solubilized wild-type INV proteins. After Western transfer, a single complex of ∼230 kD was immunodetected with antibodies directed against SecY, SecE, or SecG, the core subunits of the bacterial Sec translocon (). Thus, the 400-kD FtsY complex observed in the presence of GMP-PNP would be compatible with the association of the 200-kD FtsY complex with an ∼230-kD SecY translocon complex. To verify this, His-tagged FtsY was synthesized in a large-scale in vitro reaction and incubated with INVs in the absence or presence of GMP-PNP. INVs were isolated and solubilized, and the solubilized material was purified by metal affinity chromatography. Independently of whether GMP-PNP had been added, similar amounts of radiolabeled FtsY were eluted from the column (, top, lanes 9 and 12). A small aliquot of the eluted material was separated on BN-PAGE to verify that both the 200- and 400-kD complexes were detectable, which was indeed the case (). The remaining material was separated on SDS-PAGE for immune detection. SecY, SecE, and SecG were detectable in the eluted 400-kD FtsY complex (, lane 12). All three translocon components were, however, undetectable in the eluted 200-kD FtsY complex (, lane 9). Neither the 200- nor the 400-kD complex appeared to contain significant amounts of YidC or Ffh, the protein component of the bacterial SRP (, lanes 9 and 12), although some YidC was detectable in the wash fraction (, lane 8). This is probably due to a partial overloading of the column. To validate the copurification method, we also in vitro synthesized FtsY lacking the His tag (, lanes 3 and 6) and repeated the purification procedure, but neither FtsY nor the translocon components were eluted from the column. In conclusion, our data strongly suggest that the 400-kD complex represents an FtsY–SecYEG translocon complex. Several reports have suggested that the association of FtsY with the membrane involves two distinct binding sites (, ; ). Our data suggest that the SecY translocon provides the proteinaceous binding site proposed by . Because FtsY has been shown to bind to liposomes and to insert into lipid monolayers (), lipids probably provide a second binding site for FtsY. For analyzing SecY-independent binding of FtsY to the membrane, we adopted the approach of limited proteolysis, which has been used to detect lipid- or nucleotide-induced conformational changes in FtsY (; ). In the absence of INVs, FtsY was almost completely degraded by proteinase K, independently of whether GMP-PNP had been added (). In the presence of INVs, FtsY was also almost completely degraded, unless GMP-PNP had been added. In fact, under these conditions, a strong protease-protected fragment of 33 kD was detected, which over time was further degraded into a 25-kD protease-protected fragment (). Both fragments were immunoprecipitated by α-FtsY antibodies, confirming that they were derived from full-size FtsY. Two additional fragments recognizable only after immunoprecipitation (, asterisk) are probably the result of an additional cleavage event during the immunoprecipitation procedure. Antibodies directed against the C-terminal (His) tag of FtsY immunoprecipitated only full-size FtsY but not the protease-protected fragments, suggesting that in addition to the cleavage at the N-terminal part (; see ), proteinase K also cleaves at the C-terminal part of FtsY. As for the carbonate resistance, the appearance of the proteinase K–resistant fragments was strictly dependent on the addition of GMP-PNP; the addition of GTP, GDP, ATP, or AMP-PNP did not result in proteinase K protection of FtsY (unpublished data). Finally, we tested whether the proteinase K protection of FtsY reflected a SecY-independent interaction by using SecE-depleted INVs. In contrast to the carbonate-resistance data, protease protection of FtsY in the presence of GMP-PNP was observed for both the SecE-depleted and the SecY40 INVs, albeit at a slightly reduced level in comparison to wild-type INVs (). In summary, our data strongly suggest that the carbonate- and proteinase K–resistant states of FtsY reflect two different types of FtsY membrane interactions: to achieve carbonate resistance, FtsY has to be in close contact with the SecY translocon, whereas the proteinase K–resistant state of FtsY is independent of the SecY translocon. Membrane binding of the FtsY is thought to require at least its N-terminal A-domain (; ; ) but probably also involves the N-domain (). Surprisingly, it has recently been shown that adding a single phenylalanine residue to the N terminus of a nonfunctional NG derivative of FtsY is sufficient to support the growth of (). To analyze the role of the A-domain in more detail, we analyzed membrane binding of the FtsY(NG+1) derivative both by carbonate-resistance and proteinase K–protection assays. Different from full-size FtsY, FtsY(NG+1) did not become carbonate resistant in the presence of GMP-PNP (). Although FtsY(NG+1) did not gain carbonate resistance in the presence of GMP-PNP, the 33-kD proteinase K–protected fragment was clearly visible (). The presence of identical protease-protected fragments of full-size FtsY and FtsY(NG+1) suggests that proteinase K predominantly cleaves off the A-domain of FtsY. Importantly, our data suggest that in the absence of the A-domain, FtsY is still able to interact with the membrane and to acquire a proteinase K–resistant conformation in the presence of GMP-PNP, although protease protection is slightly less pronounced than for full-size FtsY. On the other hand, the lack of the A-domain seems to completely diminish the carbonate-resistant interaction with the SecY translocon. To verify this, we made use of the observation that the A-domain of membrane bound FtsY is accessible to proteinase K. When full-size FtsY was incubated with GMP-PNP and INVs, a large portion was carbonate resistant (, lane 2). When membrane-associated full-size FtsY was first treated with proteinase K and carbonate extracted, the 33- and 25-kD FtsY fragments were exclusively found in the supernatant. This suggests that cleaving off the A-domain prevents a carbonate-resistant interaction of FtsY with the SecY translocon. We also analyzed the membrane association of FtsY(NG+1) by BN-PAGE analyses. However, because the electrophoretic mobility of FtsY lacking the A-domain in SDS-PAGE () and BN-PAGE (unpublished data) is completely different from full-size FtsY, it is difficult to draw internally consistent conclusions from this particular experiment. In light of the recent observation that FtsY(NG+1) is at least partially functional in vivo (), we used a complementation assay () to test the functionality of FtsY(NG+1) in vitro. The integration of the SRP-dependent membrane protein mannitol permease (MtlA) was drastically reduced in the presence of SecY40 INVs () but was almost completely restored by the addition of 1 μg wild-type FtsY. In contrast, the same amount of FtsY(NG+1) did not significantly improve the integration of MtlA. It was only by increasing the concentrations of FtsY(NG+1) that a measurable integration of MtlA in INVs was observed (). This suggests that although the A-domain is not essential for the interaction between FtsY and the SecY translocon, it improves the efficiency of this interaction. In light of the fact that carbonate resistance is only observed in the presence of the A-domain, the A-domain probably serves as a transient lipid anchor that stabilizes the FtsY–SecY interaction. Although the bacterial SR lacks a membrane-integral SRβ homologue, the SRα homologue FtsY is still found partly associated with the cytoplasmic membrane (). Importantly, only the membrane-associated FtsY appears to mediate the transfer of the RNC from the SRP to the SecY translocon (), suggesting that FtsY requires the context of the membrane to be functional in the targeting reaction. This raises the important question of how FtsY is tethered to the cytoplasmic membrane and how this is coordinated with the targeting reaction. Our data provide clear evidence that the SecY translocon provides one binding site for FtsY, which supports our recent hypothesis () that SecY is most likely the proteinaceous factor proposed to be involved in membrane binding of FtsY (; ). Binding of FtsY to the SecY translocon is reflected by its carbonate resistance and the formation of a 400-kD FtsY–SecY translocon complex. It involves most likely the surface-exposed cytoplasmic loop C5 of SecY, which is suggested to provide an important interface for cytosolic factors during protein transport (). A single amino acid exchange in this loop, like in the mutant, blocks the carbonate-resistant interaction between SecY and FtsY. This nicely explains why, in this mutant, the SRP- but not the SecA-dependent protein transport is impaired (; ). The A-domain of FtsY appears to be involved in stabilizing the FtsY–SecY interaction. The important function of the A-domain is also reflected by the observation that FtsY derivatives lacking the complete A-domain are nonfunctional in (). Surprisingly, however, it has been shown recently that adding a single phenylalanine residue to the N terminus of an otherwise nonfunctional NG-domain is sufficient to support growth of in vivo (), although its ability to bind to the membrane is reduced. In agreement with this observation, our analyses show that this (NG+1) derivative is partly functional in an in vitro–complementation assay using mutant INVs. However, to suppress the phenotype, significantly higher concentrations of FtsY(NG+1) are required in comparison to wild-type FtsY. In addition, unlike wild-type FtsY, FtsY(NG+1) does not associate in a carbonate-resistant manner with the SecY translocon. These data could indicate that the primary contact between FtsY and SecY is mediated via the NG-domain of FtsY, which is then further stabilized by the A-domain. This would also be in agreement with the observation that some bacterial FtsY lack the A-domain (). Because carbonate resistance is considered to primarily reflect protein–lipid contact, the A-domain could serve as a transient lipid anchor in . It has been shown that even in the absence of the SecY translocon, FtsY can still bind to the membrane and is able to induce with low efficiency SRP release from a nascent chain (). However, in the absence of the SecY translocon, SRP release is not coupled to membrane insertion (). To prevent such a futile targeting cycle, it has been suggested that FtsY needs to be primed for coordinating the efficient SRP–RNC targeting with the subsequent insertion process (; ). Consistently, our data could suggest that it is the interaction with SecY that primes FtsY for efficient binding of the SRP–RNC complex. In this scenario, the targeting reaction would be directly coupled to the accessibility of the translocon. In eukaryotes, a role of SRβ in sensing the availability of the translocon has been suggested (), which is in agreement with data showing that SRβ plays an important role in the transfer of the RNC from SRP to the translocon (). A recent model suggests that only in the presence of an empty translocon does SRα assemble with SRβ, thus forming a functional receptor for SRP–RNCs in close vicinity to an accessible translocon (). The recent cryo-EM structure of an SR–SRP–RNC complex does not reveal any direct contact between SRα and the ribosome () but confirms cross-linking data suggesting a direct contact between SRβ and ribosomal proteins (). Even though biochemical data indicate a role also of SRα in ribosome binding (), it is evident that SRβ is an important factor for membrane tethering of the SRP–RNC complex. In bacteria, the absence of SRβ might necessitate the formation of an FtsY translocon complex to provide binding sites for both SRP (via FtsY) and the ribosome (via SecY), thereby allowing the efficient and highly selective delivery of an SRP–RNC complex to the protein-conducting channel. The ability of FtsY to bind to the membrane in the absence of the Sec translocon suggests the presence of a second, SecY-independent binding site, which could be provided by phospholipids (, ; ) or by an as-yet-unidentified protein component. We show here that in the absence of the SecY translocon, FtsY forms a 200-kD membrane bound complex, which in contrast to the 400-kD FtsY translocon complex, is sensitive to carbonate extraction. So far, we have been unable to detect any additional protein besides FtsY in this complex; in particular YidC, which has been shown in chloroplasts to interact with FtsY (), does not seem to be present. This could indicate that the 200-kD FtsY complex reflects just lipid bound FtsY and does not involve any additional protein. This would be in agreement with gel filtration data showing that purified native FtsY elutes in a single peak of 200 kD (). However, because of the aberrant mobility of FtsY in SDS-PAGE and in gel filtration studies, it is currently not possible to unambiguously determine whether the 200-kD FtsY species corresponds to an FtsY monomer with aberrant mobility, an FtsY oligomer, or a complex between FtsY and an as-yet-unknown component. In any case, the presence of a second, SecY-independent binding site for FtsY is also suggested by our observation that the NG-domain of FtsY acquires protease protection upon membrane contact, which does not require the presence of the SecY translocon. In the absence of the SecY translocon, only the 200-kD complex is observed, suggesting that the protease-protected fragments most likely derive from this complex. Protease protection of FtsY is also largely independent of the A-domain, suggesting that this domain is not essential for the SecY-independent membrane binding of FtsY. One striking result of our experiments is that binding of FtsY to both the SecY translocon and the SecY-independent binding site is greatly stabilized by blocking the GTP hydrolysis of FtsY. This was achieved by using either GMP-PNP or by analyzing GTPase mutants of FtsY. In solution, GMP-PNP has been shown to stabilize the complex between FtsY and SRP (; ); however, we found no indication for the presence of SRP in the 200- or 400-kD FtsY complexes. In addition, the FtsY mutants used in this study show distinct phenotypes as to their ability to bind to SRP. The FtsY(G455W) and FtsY(G455V) mutants are not only defective in GTP hydrolysis but also impair the FtsY–SRP complex formation (; ). Thus, although their interaction with SRP is impaired, they are still carbonate resistant. The same carbonate-resistant phenotype is also observed for the FtsY(A334W) mutant, in which the GTPase activity is blocked but not its ability to interact with SRP (). Thus, it appears that the formation of the 400-kD FtsY–SecY translocon complex does not depend on SRP but that FtsY and SecY probably associate before the delivery of the SRP–RNC. Membranes have been shown to stimulate the GTPase activity of FtsY in the absence of any SRP–RNC complex (). Thus, the strongly increased membrane binding of FtsY, induced by blocking its GTPase activity, could in principle indicate that GTP hydrolysis is involved in dissociating not only the SRP–FtsY interaction but also the FtsY membrane–translocon interaction. A precedent for such an interaction is found in the ATPase SecA, which, like FtsY, shows a low basal hydrolysis activity that is stimulated by membranes (). Likewise, the membrane association of SecA is enhanced in mutants with reduced ATPase activity () or in the presence of AMP-PNP (). Thus, it appears that FtsY, the receptor for bacterial membrane proteins, and SecA, the receptor for bacterial secretory proteins, have adapted a similar nucleotide-dependent mechanism for membrane association. Because SecA and FtsY are suggested to bind to the same cytoplasmic loops of SecY (; ), they probably cannot simultaneously bind to the Sec translocon. A possible dissociation of FtsY from the Sec translocon upon GTP hydrolysis would ensure that FtsY only transiently occupies the Sec translocon and would thereby also allow SRP/FtsY-independent substrates to access it. Binding to lipids would be advantageous because it would locally increase the FtsY concentration at the membrane and subsequently increase the chances to rebind to the limited number of SecYEG translocons. In summary, our data suggest that FtsY binds to the membrane via two distinct contact sites: the Sec translocon and probably the lipids. Both interactions are stabilized by blocking the GTPase activity of FtsY. In many respects, the features of FtsY described in this study and in previous reports are reminiscent of the features of SecA, which is considered to be a peripheral subunit of the SecY translocon during the transport of secretory proteins (). Thus, it is possible that the dual ability of the SecY translocon to translocate secretory proteins as well as to integrate membrane proteins is determined by its interaction with two peripheral subunits, either SecA or FtsY. The strains and plasmids used for in vitro synthesis were described previously (; ). The gene containing the G455V mutation was amplified by PCR from the plasmid pJH15 (provided by H. Bernstein, National Institutes of Health, Bethesda, MD; ) using the NcoI–FtsY (5′-GATAACCATGGCGAAAGAAAAAAAACG-3′) and the FtsY–HindIII (5′-GATAAAGCTTATCCTCTCGGGC-3′) primers. FtsY(G455V) PCR product was then digested by NcoI and HindIII and cloned in place of (wild type) gene in the pTD37 (), resulting in the pSAFtsY(G455V) plasmid. The FtsY(NG+1) construct was provided by E. Bibi (Weizmann Institute of Science, Rehovot, Israel) but reconstructed by PCR using the plasmid pTD37 and the two respective primers, NdeI–NG+1–FtsY (5′-GGAATTCCATATGTTCGCGCGCCTG-3′) and FtsY–HindIII. FtsY(NG+1) PCR product was digested by NdeI and HindIII and introduced in the pET22b (Novagen), generating the pSAFtsY(NG+1) plasmid. The FtsY mutants G455W and A334W were constructed by whole plasmid PCR amplification of pTD37 using the mutagenic primers FtsY-455Wf (5′-ACGGCGAAAGGCTGGGTAATTTTCTCGGTGGCT-3′) and FtsY-455r (5′-AGCCACCGAGAAAATTACCCAGCCTTTCGCCGT-3′) and FtsY-334Wf (5′-GGTGATACTTTCCGTTGGGCTGCGGTTGAACAG-3′) and FtsY-334Wr (5′-CTGTTCAACCGCAGCCCAACGGAAAGTATCACC-3′), respectively. In vitro protein synthesis and the composition of the reconstituted transcription/translation system of was described previously (). For purification of S-FtsY, the protein was first in vitro synthesized and then applied to a Talon metal affinity resin (BD Biosciences). After a 30-min incubation at 4°C, the resin was filled into a 2-ml gravity-flow column (QIAGEN) and washed with 20 bed volume of buffer A (30 mM TeaOAc, pH 7.5, 50 mM KOAc, 9 mM Mg[OAc], and 10% glycerol). S-FtsY was eluted with buffer A supplemented with 200 mM imidazole/HCl, pH 7. Fractions were analyzed on SDS gel, and FtsY-containing fractions were directly used for the subsequent experiments. For carbonate extraction, samples were incubated with 0.18 M NaCO, pH 11.3, for 15 min at 4°C and subsequently centrifuged for 30 min at 70,000 rpm in rotor (TLA 100.2; Beckman Coulter). The supernatants were neutralized with glacial acetic acid, precipitated with 1 vol 10% TCA, and resuspended in SDS loading buffer. Pellets were directly dissolved in SDS loading buffer and, like supernatants, separated on SDS-PAGE. For the protease treatment, samples were incubated with 0.5 mg/ml proteinase K for 20 min at 25°C and subsequently TCA precipitated. Immunoprecipitations with α-FtsY and α-His antibodies were performed as reported in . For proteinase K–treated samples, 1 mM PMSF was added before immunoprecipitations. 50 μg INV was incubated for 20 min at 37°C with the purified S-FtsY in the presence or absence of GMP-PNP (final concentration, 2 mM). After incubation, INVs were solubilized for 15 min at 4°C in lysis buffer (0.2% [wt/vol] dodecylmaltoside [Roche], 5 mM 6-aminohexanoic acid, 50 mM imidazol, pH 7, 50 mM NaCl, 10% glycerol, and 2 mM PMSF). After solubilization, samples were centrifuged at 4°C for 15 min at 30 000 . Soluble material was analyzed by BN-PAGE as described in . For large-scale in vitro synthesis of FtsY, a 100-fold in vitro reaction mixture was incubated for 45 min at 37°C and subsequently incubated for 20 min with 3 mg INV in the presence or absence of GMP-PNP. INVs were isolated and solubilized in 500 μl lysis buffer for 30 min at 4°C. After centrifugation, the supernatant was applied to a 3-ml Talon column. Washing and elution was performed as described above. All samples were analyzed on 15% SDS-PAGE gels or 4–13% or 5–13% BN-PAGE gels. Radiolabeled proteins were visualized by phosphorimaging using a phosphorimager (Molecular Dynamics) and quantified using the Imagequant software (Molecular Dynamics).

Eukaryotic plasma membranes are thought to contain dynamic microdomains called lipid rafts, which are proposed to be cholesterol- and sphingolipid-rich ordered domains that “float” in an environment of more fluid regions (; ). However, some authors have argued that rafts are not preexisting structures in cell membranes, but are induced by clustering of raft components (). It has also been suggested that rafts may simply be artifactual (). Rafts were initially defined by their insolubility in the detergent Triton X-100 (). Many proteins, including important signaling molecules, preferentially partition into these fractions (). However, because cold detergents can scramble lipids (; ), these membranes cannot be equated with native microdomains (). In cells, glycosyl-phosphatidylinositol–anchored and myristoylated/palmitoylated proteins have been reported to be present in cholesterol-dependent domains that are tens of nanometers in diameter (; ; ). In addition, caveolae represent a subtype of lipid raft that form flask-shaped membrane invaginations containing the structural protein caveolin1 (Cav1; ). Cav1 directly interacts with cholesterol, palmitic acid, and stearic acid and has been implicated in signal transduction () and endocytosis (), particularly of cholera toxin subunit B (CtxB), which binds to the ganglioside GM1 (). Cell adhesion to the ECM is mediated mainly by integrins, which, when in culture, cluster with numerous cytoskeletal and signaling proteins at focal adhesions and focal complexes (). Integrins control many signaling events that are critical for cell survival, growth, and gene expression (; ). Both integrin clustering and changes in conformation caused by ligand binding contribute to these signaling events. Additionally, integrin binding to ECM proteins is controlled by intracellular signaling pathways. There are several reports linking integrins with lipid rafts and/or caveolin. Functional activation of integrins, i.e., conversion to the high affinity state, appears to be linked to lipid raft localization (; ; ; ). Caveolin was reported to physically associate with integrins (; ; ). Recently, it was shown that integrin-mediated adhesion regulates the trafficking of lipid raft components such that when cells are detached multiple raft markers are rapidly internalized (). This process requires dynamin2 and caveolin1 phosphorylated at Tyr 14 (). The phosphorylated caveolin1 (pYCav1) localizes to focal adhesions in adherent cells, but relocalizes to caveolae when cells are detached. We analyzed membrane order at focal adhesions and determined its dependence on integrins, Cav1 expression, and phosphorylation. We used the membrane dye Laurdan in conjunction with two-photon laser scanning microscopy, which has been used extensively to define ordered domains in artificial membranes () and in live and fixed cells (). In macrophages, neutrophils, and activated T lymphocytes, condensed membranes cover a significant proportion of the cell surface and are frequently associated with actin-rich membrane protrusions (; ) or immunological synapses (). Our results show that membrane order is highly dependent on integrin binding to the ECM, that focal adhesions are sites of high membrane order, and that these events are partially dependent on caveolin. To assess the physical state of cell membranes, we analyzed Laurdan fluorescence using two-photon microscopy. The Laurdan probe does not preferentially partition into either lipid phase, but aligns parallel to the phospholipids () and undergoes a shift in its peak emission wavelength from ∼500 nm in fluid membranes to ∼440 nm in ordered membranes. The shift to longer emission wavelength is caused by partial penetration of water molecules into more fluid membranes; a polar environment favors an internal charge transfer state of the probe with an energetically lower excited state and, hence, a longer emission wavelength (). A normalized ratio of the two emission regions, given by the general polarization (GP), provides a relative measure of membrane order. GP values are, in principle, between −1 and +1, with fluid domains ranging from ∼0.05 to 0.25 and ordered domains ranging from 0.25 to 0.55 (). Although Laurdan has reported different lipid phases in liposomes (), phase separation in cell plasma membranes has not been observed (). In the complex environment of cell membranes, GP values reflect the overall membrane structure. When cells are fixed and immunolabeled, Laurdan does not bind to or become trapped in complexes or membrane domains, as indicated by complete extraction by 0.1% Triton X-100 (Fig. S1 A, available at ). However, we cannot completely exclude that proteins, or perhaps physical parameters of membranes other than lipid packing, could affect Laurdan's spectral properties. Serum-starved pig aortic endothelial cells (PAEC) on fibronectin (FN)-coated glass coverslips were imaged at the basal membrane. GP images () show punctuated, irregularly distributed, high GP domains pseudocolored yellow to red. Shown at a higher magnification in , these ordered domains are typically a few pixels in diameter, where a single pixel (∼215 × 215 nm) is close to the spatial resolution of the microscope (183 nm). The high GP domains seen in adherent PAEC are therefore 0.2–1.0 μm in diameter. It is important to note that these areas do not necessarily represent single membrane domains, but rather areas in which the fraction of ordered domains is higher. To identify distinct membrane structures, we immunolabeled these cells with antibodies against phosphorylated FAK (pFAK) and phosphorylated caveolin-1 (pYCav1) as focal adhesion markers (), total Cav1 as a caveolar marker, or Cy3-conjugated CTxB, which binds to GM1 and is a well-established marker for lipid rafts (). To correlate GP with focal adhesions, we used the immunofluorescent images to mask the GP images; the masked GP images only show the immunostained pixels, using the same pseudocoloring to indicate GP values. shows the masked GP images of Cav1, pYCav1, and pFAK, respectively. Particularly when magnified (), it becomes apparent that pYCav1- and pFAK-stained pixels are substantially enriched in high GP areas colored yellow and red, whereas Cav1-stained pixels select GP values at the border between red and green. Even within a focal adhesion, GP values are not homogenous, indicating the absence of phase boundaries, as well as the complexity of these membrane sites. To quantify GP values within the immunoselected areas, we determined the average GP values of the masked images and calculated the mean of images. This pixel-per-pixel comparison gives a “fluidity” index for pixels that stain positively for selected markers. clearly shows that pFAK-positive regions are highly ordered, with a mean GP value (0.502 ± 0.067) that is even higher than CTxB-stained areas (0.430 ± 0.084). In contrast, GP values averaged over the entire cell are 0.23 (). As a control, we determined the GP value of transferrin receptor (TfR)–stained regions (Fig. S1; 0.165 ± 0.066), which was consistent with the exclusion of TfR from cholesterol-rich domains (). Triggering TfR uptake resulted in a similar mean GP value, suggesting that neither surface-bound TfR nor coated pits or endosomes contain a high fraction of ordered domains. Collectively, these results show that focal adhesions are highly ordered. It has been shown previously that Cav1 phosphorylated on Y14 localizes to focal adhesions (; ; ). This phosphorylated fraction comprises <1% of the total Cav1 (); thus, the total Cav1 staining of focal adhesions is very weak. Cav1 instead localizes to distinct regions of the cell (), where, presumably, caveolae are abundant. Overlaying GP with images of Cav1 or pYCav1 showed that these regions are also highly ordered (, and ). Interestingly, GP was higher in pixels positive for pYCav1 than for Cav1 (P < 0.05), which is consistent with focal adhesions being very highly ordered. We next determined whether membrane structure at pYCav1-stained regions depends on cholesterol (Fig. S2, available at ). After cholesterol depletion with 10 mM methyl-β-cyclodextrin (mβCD) for 1 h, the mean GP value at these regions decreased to 0.293 ± 0.077 ( = 19; P < 0.001). The mean GP value was restored to 0.508 ± 0.049 ( = 17) after cholesterol-depleted cells were incubated with 15 μg/ml cholesterol complexed to 0.37 mM mβCD for an additional 1 h. These data indicate that membrane order at focal adhesions depends on cholesterol. The localization of CtxB to Cav1-enriched domains increased 2–10 min after the detachment of cells from the substratum and was followed by internalization of the CtxB (). Therefore, we analyzed the effects of cell adhesion on membrane structure. After detachment, all membrane domains become more fluid (); CtxB-positive domains in particular showed a drastic, time-dependent decrease in mean GP. However, a difference in mean GP after 1–2 min of detachment was only significant for the focal adhesion marker pYCav1, suggesting that the structure of pYCav1-containing membrane domains is dependent on integrin engagement. It is also noteworthy that Cav1 and pYCav1 are located in membrane domains of similar GP value in detached cells, whereas pYCav1 was found in more ordered domains in adherent cells. This observation is consistent with increased colocalization of pYCav1 with Cav1 upon detachment (). Collectively, the data suggest that membrane order at focal adhesions is higher than that at caveolae, and that focal adhesion, but not caveolar membrane structure, is dependent on integrin engagement. Keeping endothelial cells in suspension triggers the internalization of GM1-containing membrane domains (unpublished data), just as in fibroblasts (). Consistent with this result, the mean GP of internalized GM1-containing raft membranes showed a statistically significant, time-dependent decrease after detachment (). This finding is in agreement with the increased solubility of GM1 in cold detergent after cell detachment (). Collectively, the data suggest that not only are ordered domains internalized when cells are detached but also that membrane organization is drastically perturbed as a result of this process. compares the global membrane structure of adherent PAEC with PAEC suspended for 1–2 min. Normalized GP histograms can be accurately described by fitting to two Gaussian populations (error function [ERF] < 0.01). In adherent cells, the fluid population covered 82.3% of the surface and the mean GP was 0.166; ordered domains covered 17.7% and mean GP was 0.508. Immediately after detachment, the GP of the ordered population decreased to 0.444, while its abundance decreased to 10.8%. It is interesting to note that the fluid population increased its coverage to 89.2%, but became more ordered (mean GP 0.255). This result suggests that components that confer order, such as cholesterol, may have moved from the rafts into the bulk membrane. To investigate whether the high degree of order within focal adhesions requires Cav1, we examined mouse embryonic fibroblasts (MEFs) from wild-type (WT) and Cav1 animals. Distribution of pFAK in MEFs on FN was similar in both cell types (), demonstrating that focal adhesions are not compromised in Cav1 cells, as previously reported (). However, analysis of GP revealed that focal adhesions were significantly more ordered in WT MEFs compared with Cav1 MEFs ( and ), although lower than those in PAEC. We next compared CtxB-enriched domains in both cell types. The intensity of CTxB staining was similar in both cell types (unpublished data) and a comparison of GP values in CTxB-positive pixels revealed no significant difference between Cav1 and WT MEFs (). Previous work suggested that lipid rafts may influence integrin function, with integrin activation, clustering, and adhesion being raft-dependent (). Therefore, we considered that integrin affinities may be lower in Cav1 compared with WT cells, which in turn could influence membrane order at focal adhesions. Affinity state for integrin α5β1, which is the main FN receptor, was measured by the binding of a soluble integrin-binding fragment of FN that has been used previously to measure α5β1 affinity (). These measurements showed no difference between WT and Cav1 cells (Fig. S3, available at ). Integrin αvβ3 activation state, assayed by binding of the Fab fragment WOW1 (), also showed no difference (unpublished data). Thus, effects of caveolin on membrane order are not caused by changes in integrin affinity state. Next, we compared the global GP distribution in adherent WT and Cav1 MEFs (). In WT MEFs, two populations were evident, one with mean GP = 0.178 and 71.4% coverage, and a second with GP = 0.565 and 28.6% coverage. In Cav1 MEFs, the fluid population had a mean GP of 0.161 and 91.9% coverage, while the ordered population had mean GP of 0.471 and 8.1% coverage. Hence, ordered domains in adherent Cav1 MEFs are both less abundant and less ordered. This difference is most likely caused by the combination of less-ordered focal adhesions and loss of caveolae. It should also be noted that Cav1 MEFs are enriched in cholesterol esters but depleted of unesterified cholesterol compared with WT MEFs (), whereas the levels of major phospholipid classes are unaltered (unpublished data). These results are consistent with the known mechanism for autoregulation of cholesterol levels through SREBP cleavage (). Changes in cholesterol levels and distribution could also contribute to the observed difference in GP distribution. Nevertheless, membranes outside of focal adhesions and caveolae were not significantly different, suggesting that these effects are specific. These results led us to ask whether changes in focal adhesion order in Cav1 cells are caused by global changes in membrane structure or by specific loss of pYCav1. Therefore, Cav1 MEFs were transfected either with FLAG-tagged WT Cav1 () or FLAG-tagged Y14F Cav1 mutant (). WT and Y14F Cav1 expression levels were similar (not depicted), as were their staining patterns (). Flag-positive pixels had similar order (GP = 0.354 ± 0.057 for Y14F Cav1 and GP = 0.358 ± 0.060 for WT Cav1; see FLAG in ), which matched caveolin-positive pixels in nontransfected WT MEF (GP = 0.382 ± 0.070; ) or mock-transfected WT MEFs (GP = 0.376 ± 0.042; Table S1, available at ). These results are consistent with the finding that Y14F Cav1 forms caveolae (). Global GP histograms showed that both WT and Y14F Cav1 () increased membrane order relative to nontransfected or mock-transfected Cav1 MEFs ( and Table S1, respectively). However, WT Cav1 increased order to a greater extent (mean GP of ordered domains 0.535 vs. 0.487 in WT Cav1 vs. Y14F Cav1, respectively; ) and coverage of ordered domains was higher (21.6% vs. 13.5% for WT vs. Y14F Cav1, respectively). WT Cav1 was particularly more effective at increasing GP values within focal adhesions (marked by pFAK; GP = 0.421 ± 0.050), relative to Y14F Cav1 (GP = 0.359 ± 0.063; ). The more ordered state within pFAK-positive pixels in WT Cav1–expressing cells is also visible in compared with , which shows GP values in pFAK-stained areas in Y14F Cav1–expressing cells. Thus, phosphorylation of Cav1 on Tyr14 is important for the highly ordered state of focal adhesion membranes. To address whether the difference in GP distribution between WT Cav1– and Y14F Cav1–expressing cells can be attributed entirely to higher membrane order within focal adhesions, we analyzed pFAK-negative pixels in both cell types (unpublished data). We found no difference in membrane order outside of focal adhesions. GP in pFAK-negative areas were 0.297 ± 0.073 ( = 12) and 0.309 ± 0.096 ( = 12) for Cav1 cells transfected with WT Cav1 or Y14F Cav1, respectively. GP values outside focal adhesions in WT Cav1– or Y14F Cav1–expressing cells were significantly higher than those found in Cav1 cells (0.237 ± 0.052; = 12), but not WT cells (0.328 ± 0.076; = 12). It is likely that the presence of caveolae account for the difference in membrane order between Cav1 and WT cells, whereas the difference in membrane order between WT Cav1– and Y14F Cav1–expressing cells appears to be caused by the change in focal adhesions. Detachment of cells from the substratum induces internalization of raft components in WT, but not Cav1, cells (). However, detachment could also perturb domains through other mechanisms. Therefore, we examined WT and Cav1 MEFs at various times after detachment (). shows mean GP values of CTxB-positive pixels. As in PAEC, the mean GP value of GM1-positive domains slightly decreased immediately after detachment in both cell types. However, in WT cells, mean GP values continued to decrease as in PAEC. In contrast, Cav1 MEFs showed no further decrease in GP within CtxB-positive regions. When we defined plasma membrane as the outer 0.5–1.2 μm (∼3–6 pixels) of Laurdan-stained cells and internalized membranes as those inside this zone (indicated in ), the outer regions became increasingly fluid in WT, but not Cav1, MEFs (). The internalized membranes also became more fluid in WT cells. These data therefore support the hypothesis that the time-dependent decrease in fluidity of both the plasma membrane and internalized membranes after detachment are dependent on caveolin. Whether lipid rafts are preexisting structures in cell membranes that are determined by the self-organization of cholesterol and membrane lipids or are induced by clustering of membrane components has been a controversial question. We demonstrate that membrane order in cells, as detected by the reporter molecule Laurdan, is highly sensitive to integrin clustering and the presence and phosphorylation of caveolin. These data therefore indicate that a significant portion of the raft structure is protein dependent. That ordered membranes at focal adhesions partially depend on the expression of Cav1 and its phosphorylation on Tyr14 suggests that localization of pYCav1 to focal adhesions recruits membrane components that induce order. Cholesterol may be one such component, which is consistent with the effects of cholesterol depletion observed here and the known binding of cholesterol by caveolin (). Some of the order within focal adhesions is independent of Cav1 and most likely depends on integrin clustering. This idea is supported by the results that focal adhesions in Cav1 cells are still more ordered than surrounding regions, and that cell detachment decreases total order at early times before endocytosis of GM1. It also fits well with the association between activated integrins and lipid rafts (). These results suggest that the clustering of integrins and of pYCav1 both contribute to the assembly of membrane components into domains that are more ordered than in the unclustered state. Consistent with the finding that cell detachment from the ECM triggers internalization of membrane domains in a caveolin-dependent manner (), the fluidity of the plasma membrane showed a time-dependent decrease after detachment in WT, but not Cav1, cells. Interestingly, CTxB-stained domains became more fluid after endocytosis, suggesting that they mix with other membrane components during trafficking and lose their ordered state. The mechanisms that govern these events will have to await further characterization of the trafficking pathways for these domains; however, the new data clearly confirm previous results based on solubility and localization of lipid raft markers. The physical properties of membranes can have major effects on cell functions (). Ordered membrane domains may affect focal adhesion signaling through multiple mechanisms. These domains are believed to localize signaling components, including Src family kinases, H-Ras, heterotrimeric G proteins, and activated Rho family GTPases (; ) and may concentrate phosphoinositides (), which recruit or regulate focal adhesion proteins including α-actinin, vinculin, and talin (). Ordered membrane domains may also generate an environment that localizes kinases and excludes phosphatases (), and that can affect kinase activities directly by altering the configuration of membrane-associated proteins (). Although the mechanism has not been elucidated, Cav1 has been reported to modulate integrin function in several systems (; ; ). The extent to which changes in local membrane composition and physical state within focal adhesions contribute to these effects will be an interesting area of future work. However, to distinguish between the direct effects of caveolin via protein–protein interactions and those caused by changes in membrane order, more tools need to be developed to specifically manipulate membrane order at focal adhesions. Although it is functionally important, how membrane domains are targeted to or formed at specific sites within the plasma membrane is poorly understood. The lipid raft hypothesis suggests that small, highly mobile domains are formed by the self-assembly of cholesterol and sphingolipids (; ). For glycosyl-phosphatidylinositol–anchored () or palmitoylated proteins (), these domains are 5–10 nm in diameter; hence, clustering, possibly by actin-dependent mechanisms (), is required for detection by light microscopy. The “picket fence” model proposes that nanoscale rafts are trapped in areas with a high density of transmembrane proteins and intra- and/or extracellular anchors to the membrane (). We have recently shown that T cell activation sites are areas of condensed membranes (), although the T cell receptor complexes are assembled by protein–protein interactions (). Hence, large multimolecular protein complexes consisting of an extracellular anchor, transmembrane proteins, and a link to the actin cytoskeleton can exert an “ordering” effect on the lipid bilayer (). A similar scenario could be envisaged for focal adhesions, where the substratum and the actin cytoskeleton are linked across the lipid bilayer by integrins. This idea is consistent with our data, which demonstrate that loss of anchorage affects membrane structure quite rapidly. Within multimolecular complexes, a particular subset of proteins, such as dually palmitoylated LAT (linker for T cell activation) or caveolins, may facilitate interactions between protein complexes and raft lipids, creating a characteristic membrane structure. To what extent the recruitment of small, submicroscopic rafts contribute to the ordered membrane structure at specific sites remains to be seen. It is likely that both lipids and proteins cooperate to establish and maintain ordered membrane domains at focal adhesions. Pig aortic endothelial cells (PAEC; Cell Application, Inc.) were cultured in M199 containing 20% (vol/vol) FBS and 0.1 mg/ml heparin at 37°C in 5% CO. MEFs were prepared from 13.5-d-postcoitus embryos obtained by homozygous crossings of cav-1 KO or WT mice (). MEFs cells were immortalized by continuous passage until growth rates in culture resumed the rapid rates seen in early passage MEFs. MEFs were cultured in DME supplemented with 10% (vol/vol) FBS, 2 mM -glutamine, 100 units/L penicillin, and 100 μg/L streptomycin at 37°C in 5% CO. Antibodies against Cav1, pYCav1, and FAK were all obtained from BD Biosciences. Anti–phosphorylated FAK pY397 (pFAK) was purchased from Biosource, and anti-Flag M2 antibody was purchased from Sigma-Aldrich. Secondary donkey anti–rabbit or donkey anti–mouse IgG conjugated to either Cy3 or Cy5 were obtained from Jackson ImmunoResearch Laboratories. CTxB conjugated to Alexa Fluor 555 was obtained from Invitrogen. The antitubulin hybridoma E7 developed by Michael Klymkowsky was obtained from the Developmental Studies Hybridoma Bank at the University of Iowa. Confluent cells were serum starved (0.2% FBS) overnight, and labeled with 5 μM Laurdan (6-dodecanoyl-2-dimethylaminonaphtalene; Invitrogen) in media with 0.2% FBS for 30–60 min at 37°C (), followed by replating on fibronectin-coated (10 μg/ml) coverslips for 1–4 h in 10–20% FBS. To deplete cholesterol, cells were incubated with 10 mM mβCD for 1 h in starvation media. Where indicated, cholesterol-depleted cells were additionally incubated with 15 μg/ml cholesterol complexed to 0.37 mM mβCD for 1 h at 37°C. Adherent cells were washed twice in warm PBS and fixed in 4% paraformaldehyde (Sigma-Aldrich) at RT for 20 min. For suspension studies, cells were detached with 0.05% EDTA-trypsin, which was stopped by addition of 0.25 mg/ml soybean trypsin inhibitor (Sigma-Aldrich). Cells were sedimented, resuspended in medium containing 10 mg/ml BSA (Sigma-Aldrich; ), and incubated for the indicated times; they were then fixed in 4% paraformaldehyde (for 20 min at RT), cytospun onto poly--lysine–coated (Sigma-Aldrich) coverslips (for 5 min at 950 RPM), and fixed again in 4% paraformaldehyde (for 20 min at RT). To stain for GM1, live cells on ice were incubated with 0.1 mg/ml CTxB–Alexa Fluor 555 for 10–15 min, washed, and fixed. MEFs were transfected with plasmids encoding either FLAG-tagged WT Cav1 or Y14F Cav1 using MEF2 solution combined with the T20 program of the Amaxa system. Transfections with empty vectors were used as controls. Successfully transfected cells were identified by immunostaining with anti-FLAG antibodies. Images were obtained with a microscope (DM IRE2; Leica) equipped with photon-multiplier tubes and acquisition software (Leica). Laurdan fluorescence was excited at 800 nm with a multiphoton laser system (Verdi/Mira 900; Coherent). Laurdan intensity images were recorded simultaneously and emissions were in the range of 400–460 and 470–530 nm (). Microscopy calibrations were performed as described previously (). For confocal microscopy a helium-neon laser was used to excite Cy3 (Ex: 543 nm; Em: 550–620 nm) and Cy5 (Ex: 633 nm; Em 650–720 nm) with appropriate cut-off filters and pinhole widths. For fixed cells, a 100× oil objective, NA 1.4, was used; for live cell, a 63× water objective, NA 1.3, was used, and images were recorded at RT. The GP, which is defined aswas calculated for each pixel in the two Laurdan intensity images using software from WiT (). The custom-made WiT algorithm converts the intensity images into floating point format, calculates the GP value for each pixel, and converts the image back to an 8-bit unsigned format. To set background values to zero, the denominator (I + I) is converted to a binary image with background values set to zero, nonbackground values set to one, and the binary image multiplied with the GP image. Final GP images were pseudocolored in Photoshop (Adobe). GP distributions were obtained from the histograms of the GP images, normalized (sum = 100), and fitted to two Gaussian distributions using the nonlinear fitting algorithm using Excel software (Microsoft). The quality of the fit was determined by the ERF, as follows:where y(i) and y(i) are the experimental and calculated values, respectively. A fit is regarded as excellent when the ERF < 0.01 (). To determine GP values at focal adhesions, background-corrected confocal images were used to mask the GP images; the confocal images defined the regions of interest and the mean GP value of the regions of interest was determined for each image. GP values were corrected using the G-factor obtained for Laurdan in DMSO for each experiment (). The means and SD of two populations were compared with unpaired tests, assuming unequal variances. For multiple comparisons, one-way analysis of variance with Tukey's posttesting was performed, assuming Gaussian distributions (Prism; GraphPad Software, Inc.). Fig. S1 shows Laurdan microscopy. Fig. S2 shows cholesterol depletion. Fig. S3 shows integrin activation. Table S1 shows GP values of mock-transfected MEFs. Online supplemental material is available at .

An ESC has the capacity to transform into myriad cell types, but it is unclear what gene activities confer that ability. One possible factor that intrigues researchers such as Jeanne Loring (Burnham Institute for Medical Research, La Jolla, CA) is the distribution of methyl groups on the cell's DNA. Methyl groups don't just flip genes off; their effect on gene expression is complicated, Loring says. But methylation patterns do seem to shape a cell's developmental decisions. After fertilization, for example, eggs shed their methyl groups as they prepare to divide and differentiate. Although other researchers have measured methylation on individual ESC genes, nobody has tracked the tags across the genome. Loring and colleagues used microarrays to check for methyl attachments at more than 1,500 sites scattered across 371 genes. The researchers compared the marks in 14 ESC lines to those in adult stem cells and cancer cells. Previous work on gene expression suggests that more genes are active in ESCs than in more specialized cells, and Loring expected a similarly “wide-open” arrangement when it came to methylation. However, 35% of the sites bore methyl groups in ESCs, 3% more than in adult stem cells (). Although ESCs weren't as free and easy as the researchers expected, the cells did show a distinct methylation signature that separated them from adult stem cells, cancer cells, and differentiated cells. That pattern might underlie the cell's versatility, Loring says, and might help researchers track down the genes that bestow this ability. Loring and her colleagues are developing larger arrays to assess more genes. She would like to test a cell produced through SCNT to see whether its methylation signature matches that of ESCs. Once is definitely not enough for Alison Murdoch (Newcastle Fertility Centre at Life and University of Newcastle in the UK). Last year, her group used SCNT to create a human blastocyst, a feat matched by only one other researcher, discredited South Korean scientist Woo Suk Hwang. The blastocyst perished before yielding any stem cells, and Murdoch and colleagues see it as only a start. They hope to refine the SCNT technique until minting a cell line that matches a patient's tissues is “reliable and routine,” she says. The procedure is illegal in some countries that permit other forms of ESC work (), and no other lab in the UK has government sanction for this type of research, Murdoch says. But she'll soon have company. Scientists at Harvard, the University of California, San Francisco, and at least four other institutions plan to pursue SCNT. Murdoch's position at the head of the pack is ironic because, when her team published its results in June of last year (), it looked like an also-ran. The month before, Hwang and colleagues had announced fantastic success with SCNT, claiming not only to have created human blastocysts, but to have turned them into 11 ESC lines. Murdoch and company, meanwhile, reported comparatively modest results using 36 unfertilized eggs and donor nuclei from ESCs. Nuclei from ESCs are presumably easier to genetically reprogram to an uncommitted state. To unite eggs and nuclei, the researchers goaded them with a combination of chemical and electrical stimulation. Only three of the fused cells launched into division, and only one went further. Compared with Hwang's abundance of blastocysts and ESCs, the results “seemed like a tiny, tiny advance,” Murdoch says. By the time Hwang's scam came to light in late 2005—investigators determined that his group had produced a few human blastocysts but no ESC lines—Murdoch's team had diverted to other questions. “We thought the technology [for SCNT] was there,” she says. “It set us back a year at least.” Now that the lost year is over, the group is trying to pin down the problems that impede SCNT. Murdoch won't reveal what obstacles they've identified, but one limitation has loomed from the start: a shortage of eggs. SCNT works best on eggs freshly removed from a woman's body. But Murdoch and colleagues found that, even if they convinced in vitro fertilization patients to donate spare eggs, they could only garner a grand total from all the patients of about 10 eggs per month (). The Harvard team plans to solicit egg donations from young, healthy women. However, the UK body that regulates ESC research has rejected on ethical grounds Murdoch's application to tap the same source. She plans to apply again. With good eggs scarce, researchers like her who have worked with in vitro fertilization have an advantage, says Murdoch. They are adept at coaxing star performances from a single, recalcitrant egg. Eggs aren't the only cells vexing researchers. Human ESCs follow a different timetable than do the rodent stem cells that scientists are more accustomed to working with, says Duncan. Like the animals themselves, rodent stem cells are frenetic, dividing and differentiating swiftly. By comparison, human ESCs are leisurely—and fragile. Moreover, human ESCs shrug off the growth factors that galvanize their mouse counterparts. Because researchers know so little about rearing human ESCs, the obstacles holding up progress in the field are more technical than intellectual, he says. For seven years, Duncan and colleagues have been trying to prod ESCs to mature into oligodendrocytes. In the central nervous system, these cells make myelin, the fatty material that insulates nerves. An infusion of oligodendrocytes might help ease the symptoms of multiple sclerosis—in which myelin deteriorates under immune system assault—and several inherited diseases in which the material breaks down or doesn't form. Duncan and colleagues started by growing mouse ESCs into oligodendrocytes and transplanting them into the spinal cords of rats that lacked myelin. The implants settled in and laid down new insulation (). Next, Duncan teamed up with University of Wisconsin colleague Su-Chun Zhang and other researchers to try a similar experiment with human cells (). They nudged human ESCs into differentiating into nervous system precursor cells. In the lab dish, these cells would mature into several types of cells, including neurons, astrocytes, and a few oligodendrocytes. But when the researchers inserted the precursors into immunodeficient mice, no oligodendrocytes developed, and no new myelin formed. Hans Keirstead of the University of California, Irvine, and colleagues () have achieved these feats, Duncan notes. To improve their cells' performance, Duncan and colleagues are focusing on a molecular decision maker, the transcription factor Olig2. They plan to follow its expression during differentiation to understand how it helps a cell choose a particular path. The options for patients with type I diabetes are sour: continued insulin doses or an infusion of pancreatic islets that contain hormone-making β cells. Although islet transplants can help control blood sugar levels, donors are scarce and the benefits often wane after a few months or years. Henrik Semb's lab (Lund University, Sweden) is one of several around the world hunting for a recipe that will drive ESCs to morph into β cells, thereby providing an alternative source for transplants. Nobody has gotten an ESC to go all the way, but with a helping paw from mice, Semb and colleagues have cultivated human cells that manufacture insulin. Even under normal culture conditions, human ESCs began to differentiate and produce a transcription factor, Pdx1, that characterizes β cells, Semb's group found. Pdx1's presence suggests that the cells had started along the right road. However, they wouldn't go further and crank out insulin. To provide a boost, Semb and colleagues packed the cells into a section of mouse pancreas and tucked them under a rodent's kidney. The change of surroundings did the trick. Not only did the cells harbor three transcription factors—including Pdx1—that mark β cells, but they made the raw form of insulin (). “We are fairly confident that the cells are really close to a β cell,” Semb says. This technique couldn't produce transplant-ready β cells because of the danger of contamination with mouse proteins, which could trigger an attack by the recipient's immune system. To overcome that problem, the researchers are working to decipher the signals between the endoderm, which gives rise to β cells, and the surrounding mesenchyme in the developing pancreas. Mouse models capture some aspects of Alzheimer's disease (AD), amassing the β-amyloid plaques that riddle the brains of patients with the fatal illness. But the animals share one big drawback, says Larry Goldstein (University of California, San Diego, CA): “They don't really get AD.” Why the rodents don't lose their memories isn't certain—their physiology might differ in a key way from ours, or they might not live long enough to develop symptoms, he says. Searching for a more human alternative, Goldstein and colleagues turned to models that don't even have brains—ESCs. The cells are an improvement over existing neural cultures because they live longer and are easier to grow, and therefore are easier to genetically modify, says Goldstein. For the last 18 months, Goldstein's group has been coaxing the versatile cells into specializing into neurons. The researchers are still learning how to keep human stem cells happy—Goldstein describes them as “high-maintenance”—and nudge them down the right path. “We've had to develop a lot of the methodology ourselves,” he says. Goldstein plans to deploy the neurons to test an alternative hypothesis for AD's cause. The leading explanation blames β-amyloid buildup for starting the disease. But Goldstein wants to determine whether AD begins when the cell's transportation system goes haywire. In the axon of a neuron, a protein called kinesin-I trucks vesicles, organelles, and other molecular cargo away from the cell body, running on microtubules that function like railroad tracks. Last year, Goldstein's team reported that they found train wrecks of vesicles and organelles in brain neurons from an AD mouse model and from patients in the early stages of the illness (). More pile-ups formed in mice that fashioned 50% less kinesin than normal. Furthermore, the brains of kinesin-poor animals accrued more β-amyloid plaques, suggesting that the transportation tie-up somehow spurs the protein's accumulation. To evaluate that explanation, the researchers are genetically altering ESCs to cut kinesin levels and to carry mutations associated with inherited varieties of AD. Lab-reared neurons won't supplant mouse models, but they'll be essential for probing what instigates AD and for testing possible treatments, Goldstein says. “I don't see how you can generate a molecular description of the pathogenic events if you don't work on human cells,” he says.

Since their original discovery, the 2-μm plasmids of yeast have provided a fascinating paradigm of selfish DNA. Typically present at a copy number of 60 per cell, they impose a perceptible cost in fitness to the host yeast cell. Yeast cells that have been cured of 2-μm plasmids (or Cir strains) have a growth rate advantage of 1.5–3% over their plasmid-containing counterparts (Cir strains; ). In spite of this fitness disadvantage to their hosts, 2-μm plasmids are ubiquitous in strains, and related plasmids are found in other highly divergent species, implying that 2-μm–like plasmids have successfully hitchhiked in budding yeast cells for more than tens of millions of years. They accomplish this remarkable evolutionary stability by using two different strategies to ensure their own transmission: partitioning and amplification (; ). After DNA replication, partitioning ensures a high fidelity of segregation of plasmid molecules. This requires two encoded proteins, Rep1 and Rep2, which bind specifically to a cis-acting (for stability) locus and, together, mediate this plasmid segregation. In the absence of such an active segregation mechanism (i.e., if plasmid segregation was simply diffusion dependent), the rate of loss of 2-μm plasmids from daughter cells coupled with the fitness advantage of Cir cells would lead to the rapid clonal extinction of 2-μm plasmids. In the case of a decrease in copy number below a steady state value, plasmid amplification reestablishes this steady state by increasing rounds of replication. This amplification uses an elaborate series of recombination events mediated by the encoded Flp recombinase and two cis-acting () sites using a single DNA replication initiation event (; ). This redundancy of functions is perplexing for a selfish plasmid that is only 6 kb in length. One can easily imagine why amplification is necessary to spread to new host cells; in essence, this is a common property of all selfish elements, including transposons. Less obvious is why a selfish plasmid would go through the trouble of encoding an active partitioning mechanism if it had a robust amplification mechanism, given that the probability of a reduction of copy number from 60 to 0 in one cell cycle is extremely low. A partial answer emerged from cytological visualization of 2-μm plasmids in yeast cells, which revealed that these 60 individual plasmid molecules are present in just one to two clusters and that these clusters are inherited en masse (; ). Missegregation of these clusters can quickly lead to clonal extinction of the 2-μm plasmids. Indeed, the deletion of Rep1 or Rep2 can lead to a 30-fold higher loss of 2-μm plasmids relative to wild type after just seven to eight cell divisions (). Thus, it is mostly by virtue of its partitioning function that the 2-μm plasmid avoids extinction. How does the 2-μm plasmid achieve this high fidelity of segregation? Clues began to emerge from previous studies that showed a striking genetic and cytological concordance of plasmid segregation to that of yeast chromosomes. First, 2-μm plasmids were found concentrated near the poles of the yeast mitotic spindle (), which is a cytological localization akin to yeast centromeres. Second, in an yeast strain ( is the yeast Aurora kinase gene), both 2-μm plasmids and yeast chromosomes were found to missegregate in tandem fashion (). Finally, like yeast chromosomes, the 2-μm plasmid was found to recruit the yeast cohesin complex using Rep1 and Rep2, presumably to pair newly replicated plasmids (). This recruitment of cohesin was dependent on a specialized chromatin structure at the locus and on the chromatin remodeling activity of the RSC2 complex (; ). All of this circumstantial evidence had pointed to the 2-μm plasmid usurping the chromosome segregation machinery from the host yeast cell to ensure its own partitioning. Now, add a critical piece to this puzzle by demonstrating that the 2-μm plasmid utilizes the yeast centromeric histone Cse4 () to ensure correct segregation. Centromeric histones like Cse4 are exquisitely specific markers of centromeric chromatin in virtually all eukaryotes. The authors first show that Cse4 and 2-μm plasmids colocalize in chromosome spreads. Using chromatin immunoprecipitation followed by PCR, they show that Cse4 localizes specifically to the locus. Using high-salt extractions, they show that the locus of the 2-μm plasmid indeed assembles a Cse4-containing nucleosome, which protects the locus from restriction enzyme–mediated cleavage. They further demonstrate that -localized Cse4 is protected from proteolytic degradation in the same manner as (yeast centromere)-localized Cse4 (). Finally, they demonstrate that Cse4 is genetically required for the correct partitioning of the 2-μm plasmid. In the absence of wild-type Cse4, Rep2 does not localize to (although Rep1 does), RSC2 complex–mediated chromatin remodeling does not take place, and cohesin assembly at is blocked. Indeed, argue that correct partitioning requires correctly remodeled chromatin containing a Cse4 nucleosome, which can nucleate Rep1–STB–Rep2 interactions. These results together imply that the locus effectively mimics function. However, there are important differences between 2-μm plasmid and yeast chromosome segregation. The 3′ element of sequences specifically recruits proteins of the CBF3 complex, including Ndc10 (; ), which in turn helps recruits a Cse4 nucleosome, most likely at the element based on genetic data (; ). Thus, Cse4's localization to sequences is disrupted in an mutant at nonpermissive temperatures, but its localization to loci is unaffected. Conversely, although Cse4 localization to loci requires Rep1 and Rep2 proteins (), the deletion of Rep1 or Rep2 has no effect on its localization. This implies that 2-μm plasmids have invented a new means to recruit Cse4 to their loci, which is probably key to their longevity. speculate that a complex containing Rep1 and Rep2 may help deposit (and perhaps maintain) Cse4 at . This further implies that although some components of both the 2-μm plasmid and yeast kinetochores are bound to be in common, others (e.g., Ndc10) are not. Thus, the 2-μm plasmid kinetochore may prove to be valuable in future studies of -based kinetochores. Centromere function is under extremely strong constraints in budding yeasts, and the general architecture of sequences has been largely preserved over tens of millions of years (). In contrast, loci are highly variable in sequence and especially in length even within (; ). Previous studies have strongly suggested that this is a result of antagonism between 2-μm plasmids and their host cells (; ), as the latter try to evolve genetic solutions to their 2-μm infestations. In addition, 2-μm plasmid alleles compete with each other for survival within yeast cells (). This means that Cse4 is forced to negotiate with both very slowly evolving () as well as extremely rapidly evolving () centromeres, which is an interesting challenge for an essential histone.

Experiments performed by over a century ago revealed the essential requirement for accurate centrosome inheritance and its role in regulating genome integrity in the developing embryo. In many metazoans, the establishment of the bipolar spindle during the first zygotic cell division is dependent on the paternal contribution of a microtubule organizing center. After fertilization, this organelle will recruit pericentriolar material present within the oocyte cytoplasm to assemble the two functional centrosomes that will define the first mitotic spindle. In addition to this essential role of the centrosome in organizing the spindle, in , this structure is also required to specify the anterior/posterior axis after sperm entry in a microtubule-dependent and -independent manner (; ; ). Therefore, the appropriate regulation of centrosome number is pivotal because aberrations in these controls result in asymmetrical chromosome segregation and/or severe polarity defects. Although centrosomes are associated with most nuclei in , including those in the germ line, they are absent in oocytes, whereas they are clearly detectable and required for fertility in the sperm (). The loss of the centrosome from the oocyte is common to many species, but the mechanism responsible for this elimination is currently unknown. During our characterization of a Cdk inhibitor (CKI; ) we noticed that compromise of function caused embryos to arrest at the one-cell stage with a multipolar spindle. We show that this defect is due to a role of in centrosome elimination, and our data provide pioneering evidence on how centrosomes are appropriately eliminated from the developing oocyte. Recently, large-scale screens using RNAi-based strategies have provided a framework for understanding many maternally controlled embryonic processes (). However, not all genes respond equally to RNAi. Our initial use of RNAi analysis to understand the role of a CKI called was not informative because of the variable penetrance and frequency of the RNAi-related phenotypes. Furthermore, no loss-of-function alleles are currently available. We therefore turned to an alternative reverse genetic approach called cosuppression, which is an RNAi-related posttranscriptional gene-silencing mechanism that is conserved among many phyla (). In wild-type animals, mRNA is normally present in the hermaphrodite germ line but is excluded from the distal mitotic zone (). To test whether could be compromised through the cosuppression pathway, we expressed the 3′ portion of the gene (), which could not encode a functional protein and shared a very low degree of sequence conservation with , a second CKI (Fig. S1, available at ). The cosuppression transgenic array included a GFP marker facilitating our detection of animals that possessed the transgene. We obtained several transgenic lines in different genetic backgrounds, all of which indicated that reduction of consistently resulted in reproducible embryonic lethality wherein ∼60% of the GFP transgene-bearing embryos (GFP+) failed to complete embryogenesis (). The abundance of mRNA was reduced substantially throughout the gonad in these GFP+ animals (), whereas the observed embryonic lethality could be reversed by genetically disrupting this silencing mechanism using mutants in the downstream components of the cosuppression pathway ( and ), indicating that the observed lethality was specifically due to the reduction of through cosuppression (). We therefore refer to these GFP+ animals as cosuppressed (). Although ∼40% of the embryos survive embryogenesis and continue larval development without visible abnormalities, we found that these animals are irradiation sensitive (). This indicates that despite their wild-type appearance, the DNA damage response in animals is nonetheless compromised. Therefore, reduction of function results in cell cycle–related abnormalities that reflect the various thresholds of activity required to appropriately execute these cellular processes. Among the embryonically arrested embryos, we noticed that 7% of the embryos ( = 558) arrested at the one-cell stage with multiple micronuclei (9.1%; = 66), consistent with abnormal chromosome segregation and/or cytokinesis (). Examination of the affected zygotes by differential interference contrast indicated that early events (contractions of the anterior membrane or ruffling and pseudocleavage) before the pronuclear meeting were not significantly different from wild type (unpublished data). Shortly after nuclear envelope breakdown, however, the two pronuclei reformed and several de novo micronuclei became apparent. Cleavage furrows appeared occasionally but would regress, and ∼50% ( = 18) of the micronuclei-containing embryos did not form a cleavage furrow. The remaining 50% were defective in cleavage plane orientation, although both classes did undergo multiple rounds of karyokinesis ().To better understand the basis of the “one-cell” arrest phenotype, we imaged embryos that harbored GFP-histone and GFP–β-tubulin transgenes. In some embryos, we observed a second maternal pronucleus (4.5%; = 66), a meiotic defect that arises because of abnormal polar body exclusion (). We also noted that chromosomes failed to align correctly after nuclear envelope breakdown, whereas the spindle microtubules appeared to be organized around multiple foci, typical of extra microtubule organizing centers or centrosome-like structures ( and Video 1). To confirm that this unique multipolar spindle phenotype was due to the reduction of and not due to cosuppression-related phenomena or nonspecific effects on , we used an RNAi-sensitive strain () to reduce either or levels to reproduce the –associated multipolar spindle phenotype. We did detect one-cell embryos with supernumerary centrosomes after in ( and see ), although the penetrance of the defect was considerably lower than that observed in animals. On the other hand, despite causing a high frequency of embryonic arrest in the background, never caused a one-cell arrest or a multipolar spindle phenotype (). Therefore, we conclude that the supernumerary centrosomes and the resulting multipolar spindle defect observed in embryos were not due to effects on function or due to cosuppression per se but, rather, to a loss or reduction of function. To address whether affected the centrosome cycle during spermatogenesis or, alternatively, during oogenesis, we examined centrosome numbers in early pronuclear stage embryos using an antibody against SPD-2, a coiled-coil protein that associates with the centrosome (). We noticed that unlike wild-type embryos, strong SPD-2 expression was visible at distinct foci in both the paternal and maternal pronuclei (pronuclear meeting stage; ). To ascertain whether the presence of the extra centrosomes was indeed due to their contribution from the maternal pronucleus, as opposed to defects associated with failed cytokinesis (), we imaged embryos from meiosis to pronuclear meeting using GFP–γ-tubulin, revealing that GFP–γ-tubulin was associated with the maternal pronucleus in prepronuclear migration stage embryos obtained from animals (6.7%; = 60; ), whereas we never observed GFP–γ-tubulin associated with the maternal pronucleus in wild-type embryos ( = 80; ). Collectively, these results indicate that the supernumerary centrosomes were already associated with the maternal pronucleus at the time of fertilization in embryos, possibly because they were not appropriately eliminated in the maternal germ line as a result of a reduction in function. However, because we could not show definitive live images of an embryonic cell division beginning in the prepronuclear stage to the first mitotic division, we cannot formally rule out the possibility that the supernumerary centrosomes may arise from a cytokinesis failure after the first mitotic division. Therefore, to test whether centrosome elimination is defective in oocytes, we stained the gonads of affected (GFP+) and unaffected (GFP−) animals with an anti–SAS-4 antibody to determine whether centrioles were abnormally present in the oocytes of animals. SAS-4 is associated with all centrioles in and is required for their duplication (). In wild-type animals, SAS-4 is associated with all germ cell nuclei, although SAS-4 staining foci were noticeably absent from oocytes (). The absence of the SAS-4/centriole staining in oocytes is consistent with previous observations that the centrosomes are eliminated from the germ cell nuclei at or around the stage of oocyte commitment (). Anti–SAS-4 staining of the oocytes from the hermaphrodite animals revealed that SAS-4 staining structures were present next to the oocyte nuclei at a frequency consistent with the penetrance of the extra centrosome defect caused by the transgene (8.9%; = 79), whereas no obvious SAS-4 foci were ever observed in oocytes in wild-type animals ( and not depicted). Although this is the strongest evidence that is required for appropriate centriole elimination during oogenesis, we wanted to further confirm that the anti–SAS-4 staining recognized bona fide centrioles and not simply SAS-4 aggregates in the oocyte. We therefore stained the oocytes of wild-type and animals using anti–SAS-4 and anti–SAS-6, both of which recognize the centriole ( ; ). Both antibodies recognized the centrioles of embryos, where they colocalize with γ-tubulin (Fig. S3, available at ). After double staining, we compared the number of overlapping signals between wild-type and germ lines (). Consistent with our previous observation (), we noted that significantly more SAS-6 staining oocytes showed overlapping positive signals with anti–SAS-4 in the animals (14/55 SAS-6–positive oocytes) compared with wild-type (1/29 SAS-6–positive oocytes; this single overlapping SAS-4 signal may be due to juxtaposition of the signals during the deconvolution process; ). Therefore, our staining with two independent centriole-specific antibodies suggests that the observed foci are indeed centrioles, which are not appropriately eliminated in the oocytes. In , oogenesis occurs in an assembly line–like fashion (; ). We observed that the SAS-4 staining structures persisted into the late stages of oogenesis in hermaphrodites (). These data are consistent with playing a critical role in the timely elimination of the maternal centrioles during oogenesis, and when its activity is reduced below a critical threshold, the centrioles persist and eventually will give rise to the supernumerary centrosomes. Although our results strongly argue that is involved in the elimination of maternal centrioles, ultrastructural studies would provide more definitive evidence of centriolar perdurance. Intriguingly, although the maternally contributed centrosomes are the likely cause of the abnormal division observed in the one-cell–arrested embryos, we have been unable to show that these supernumerary centrosomes can nucleate microtubules and/or duplicate beyond the first division. We also noticed that the polarity of the affected embryos seems consistently normal based on GFP–PAR-2 (100%; = 17; ) or P-granule staining (; ). Our observation that anterior/posterior polarity does not seem to be affected in zygotes suggests that although the maternally contributed centrosomes appear competent to organize a mitotic spindle, they are seemingly not equivalent to the paternal centrosome in providing the polarity cue in the zygote. The basis of this difference between the centrosome pairs is currently unknown, as no difference in centrosomal morphology or molecular composition has been identified between the centrosomes of paternal and maternal origin. Our observations, although obtained with fixed embryos, suggest that a functional difference may distinguish the maternal and the paternal centrosome in establishing the anterior/posterior polarity at fertilization. However, we have been unsuccessful in imaging the maternally contributed centrosomes into and beyond the first division while simultaneously monitoring the establishment of the PAR-2 domain. Therefore, we cannot formally rule out the possibility that the polarity is established early by the sperm and that the extra centrosomes we observe in the multinucleate embryos are paternal in origin that have duplicated and appear later due to cytokinesis defects (). Because meiotic defects were also observed in embryos, we determined whether the abnormal presence of centrosomal components on the meiotic spindle might disrupt the normal mechanism of the acentriolar meiotic division. We found that the morphology of the meiotic spindle in early zygotes is disorganized (Fig. S2 C, available at ), whereas SPD-2 was detectable as a diffuse haze surrounding the spindle (Fig. S2, A and B). We also found that ZYG-1, a protein that is also required for centrosomal duplication (), was similarly present on the meiotic spindle in zygotes (unpublished data), suggesting that the atypical presence of these ectopic centrosomal materials may be responsible for the meiotic spindle abnormalities and the consequent meiotic defects observed in embryos. The loss of could result in misregulated levels of Cdk activity within the oocyte, causing a centrosomal anlage to persist and eventually form the tetrapolar spindle that results in a one-cell arrest. To test this scenario, we compromised G1/S Cdk activity by performing , which is the only E-type cyclin in (). Loss of cyclin E has no effect on the first cell division in (). However, after in animals, the characteristic one-cell arrest phenotype was suppressed substantially, which was also reflected in the nearly twofold reduction in the frequency of the multipolar spindle defect (). A similar degree of suppression was also observed after , where is the predicted Cdk2 homologue (; ). Control animals injected with double-stranded (dsRNA) corresponding to cyclin D showed no such effect (unpublished data). That this effect of cyclin E occurs independently of Cdk activity () seems unlikely based on the current accepted mechanism of CKI function and our observation that suppressed the frequency of the persistence of the maternal centrosomes to levels comparable to . Our data are thus consistent with the loss of resulting in misregulated cyclin E/Cdk2 activity in the germ line that consequently allows centrioles to perdure into the developing oocyte. That both ZYG-1 and SPD-2 persist during oogenesis and are present on the meiotic spindle in embryos suggests that their levels may be regulated by cyclin E/Cdk activity, in a manner similar to Mps1 (). The loss of therefore reveals a previously undescribed function of cyclin E–Cdk complexes in centrosome stabilization in the germ line. Through the timely regulation of this activity, the maternal centrosomes are eliminated as the germ cell acquires its oocyte fate. This novel function of Cdks and CKIs in centrosome inheritance would probably not have been uncovered through conventional gene targeting in mouse models. Unlike most animals, the sperm does not contribute the centrioles in the mouse; instead, they arise de novo in the fertilized zygote (). Why, then, do most metazoans selectively eliminate the centrosomes within the maternal germline? The answer may come from species that can develop parthenogenetically, where the oocyte is thought to harbor a centriolar anlage (). This would be selected against in species that undergo a biparental mode of development based on sperm-specific centriolar contribution. The elimination of the maternal centrosomes, either through CKI-mediated or related mechanisms, would block the ability of the oocyte to develop parthenogenetically and strongly favor the union of sperm and egg to trigger the onset of cell division in the zygote. Because the mode of centrosome inheritance in shares considerable parallels with that of many animals, identification of the Cdk targets in this model may provide invaluable insight pertinent to the mode of centrosome inheritance shared by most metazoans, including humans. The following strains were used: N2 Bristol was used as the wild type throughout. MR258 (N2; [∷; ∷GFP]), MR306 (N2; [∷GFP; ∷GFP]), MR294 (; [∷; ∷GFP]), MR303 (; [∷; ∷GFP]), NL917 ( []), WM29 ( []), MR446 (; [(+); ∷GFP∷H2B]; [(+); ∷GFP∷ TBB-2]; [∷; ∷GFP]), XA3501 (; [(+); ∷GFP∷H2B]; [(+); ∷GFP∷TBB-2]), TH27 (; [(+); ∷GFP∷ TBG-1]), MR628 ( [(+); ∷PAR-2∷GFP]; [∷; ∷GFP]), MR824 (; [(+); ∷GFP∷TBG-1]; [∷; ∷GFP]), NL2099 (), and KK866 ( [(+); ∷PAR-2∷GFP]). All strains were cultured using standard techniques and maintained at 20°C unless stated otherwise (). For cosuppression, 3 kb of genomic sequence upstream of the translational start site was PCR amplified from N2 genomic DNA followed by SphI–Pst1 digestion and insertion into pPD49.26 to yield pMR220. The fragment (amino acids 116–259; lacking a translational start site; Fig. S1) was prepared by PCR and then inserted into pMR220 at the BamHI–XmaI sites to create pMR221. The promoter fragment was inserted into pPD95.77 at SphI–PstI sites to yield pMR266. For RNAi of , a template for dsRNA synthesis was generated by subcloning the cDNA into the PstI–KpnI sites of pBluescript II to generate pMR215. dsRNA was prepared as described previously (). dsRNA was prepared as described previously (). dsRNA template was amplified from a clone of the bacterial feeding RNAi library (I-1D09) using PCR and inserted into the SacI–SacII sites of pBluescript II to generate pMR330. pMR220 and pMR221 were coinjected (50 μg/ml) with 100 μg/ml GFP as a coinjection marker into N2 hermaphrodites as described previously (). F1 progeny expressing ∷GFP were singled, and their progeny (F2) were scored for transmission of the extrachromosomal array. Embryonic lethality was scored from each transgenic line. dsRNA was obtained by in vitro transcription reactions, annealing, and injection as described previously (). Injected animals were transferred to new plates every 24 h, and the F1 progeny was examined for visible abnormalities that affected development or cell division. The following primary antibodies were used: anti–α-tubulin (Sigma-Aldrich), polyclonal anti–rabbit SPD-2 (a gift from K. O'Connell, National Institutes of Health, Bethesda, MD), rabbit polyclonal anti–SAS-4 (a gift from P. Gonczy, Swiss Institute for Experimental Cancer Research, Epalinges, Switzerland), Cy3-conjugated anti–SAS-6 and Cy5-conjugated anti–SAS-4 (a gift from K. Oegema, University of California, San Diego, La Jolla, CA), and rabbit polyclonal anti–P-granule (a gift from S. Strome, Indiana University, Bloomington, IN). Secondary antibodies were anti–rabbit or anti–mouse Texas red or FITC-conjugated secondary antibodies or anti–rabbit Alexa Fluor 594 secondary antibody (all obtained from Invitrogen). DAPI (Sigma-Aldrich) was used to counterstain slides to reveal DNA. Embryos or hermaphrodite gonads were fixed and stained as described elsewhere (). Indirect immunofluorescence microscopy was performed using a 60× oil-immersion objective lens in a compound microscope (DMR; Leica) equipped with a digital camera (C4742-95; Hamamatsu), imaging an ∼0.5-μm-thick optical section. Image analysis, computational deconvolution, and pseudocoloring were performed using Openlab 4.0.2 software (Improvision). All images using Cy3-conjugated anti–SAS-4 and Cy5–conjugated anti–SAS-6 were acquired (using a 60× oil-immersion objective lens) and deconvolved using an image restoration system (DeltaVision; Applied Precision). Data were collected as a series of 35 optical sections in increments of 0.25 μm under standard parameters using the SoftWoRx 3.0 program (Applied Precision). Images were processed using Photoshop 8.0 (Adobe). All microscopic works were performed at 20°C. Digoxigenin-labeled antisense and sense probes were generated using T7 and T3 kits with digoxigenin-11-UTP (Roche). In situ hybridization was performed on the gonads dissected from wild-type or (GFP+) adult hermaphrodites as described previously (). Fig. S1 shows protein sequence alignment of CKI-2 with -1. Fig. S2 depicts centrosomal material persisting on the meiotic spindle in one-cell embryos. Fig. S3 shows an embryonic cell labeled with GFP–γ-tubulin, anti–SAS-6, and anti–SAS-4. Video 1 shows a one-cell embryo labeled with GFP histones and GFP–β-tubulin. Video 2 shows a wild-type one-cell embryo (pronuclear migration stage) labeled with GFP–γ-tubulin. Video 3 shows a one-cell embryo (pronuclear migration stage) labeled with GFP–γ-tubulin. Video 4 shows a one-cell embryo (prepronuclear migration stage) labeled with GFP–γ-tubulin. Online supplemental material is available at .

The systematic ultrastructural analysis of intact cellular components in their native, hydrated state has only recently become possible because of major improvements of noninvasive EM methods for specimen preservation and imaging—cryoelectron tomography (cryo-ET) and cryoelectron microscopy of vitreous sections (CEMOVIS). In CEMOVIS, vitrified biological material is sectioned and imaged at a low temperature (for review see ). Cryo-ET can directly image thin regions of cells that adhere or grow on EM grids (), and it is becoming the method of choice for providing 3D information about intact intracellular structures at molecular resolution (). Recently, these emerging techniques have provided valuable insights into cellular architecture, as well as into complex macromolecular assemblies (for reviews see ; ). Microtubules have been studied by ET in sections from freeze-substituted and plastic-embedded material (), and recently, tomograms of frozen-hydrated sea urchin sperm flagella revealed details of their microtubules and associated proteins (; ). Microtubule-associated proteins (MAPs) maintain the stability of microtubules and regulate their dynamics, whereas microtubule-based motors mediate the transport of cargo along microtubule tracks. Although all MAPs and motors studied so far bind to the outside of the microtubule wall, small molecules such as taxol can associate with the luminal side of the microtubule (). It has also been proposed that a short repeat motif of the MAP tau can be localized on the inner surface of the microtubule lattice (). Intriguingly, the presence of electron-dense material was observed within the microtubule lumen in plastic-embedded and heavy metal–stained preparations of insect epithelia and spermatids (; ) and blood platelets (; ). Such luminal material appears to be especially prominent in neuronal cells (; ; ). However, none of these studies has revealed details about the form and distribution of this material along microtubules or the nature of its association with the microtubule wall. We used cryo-ET and CEMOVIS to examine the microtubules of neuronal and other cells in a state of optimal structural preservation. We demonstrate that these microtubules contain within their lumens discrete particles with connections to the microtubule wall, and we analyze the occurrence, size, and distribution of these particles. To study microtubules in their native state, we acquired cryo-ETs of thin processes of cultured cells, as well as images of vitrified sections, both depicting the morphology of the cytoplasm as it is believed to occur in living cells (; ). For cryo-ET we cultivated hippocampal neurons and other mammalian cells directly on the carbon support of gold EM grids. These cells have processes or peripheral areas that are thinner than 0.5 μm, and thus, are amenable to being studied in detail by ET without the need for sectioning. The morphology of neurons cultured on these EM grids was similar to that of neurons grown on classical glass coverslips, as assessed by light microscopy (). The use of “finder” EM grids facilitated correlating light microscopy images of living neurons with EM images and tomograms that were acquired after preservation by rapid freezing in liquid ethane to localize the imaged region of the cell (). Among a variety of membrane organelles and complexes, the cytoskeletal elements are visible on electron micrographs () and could be analyzed in detail in tomograms (, and Video 1, available at ). Close examination of microtubules from >20 tilt series revealed electron-dense material within the lumens that comprised discrete, closely spaced globular particles (). These particles were reliably detected only when analyzing the tomograms and not in projection images (compare to and Video 1). The typical distances between particles were 8, 12, 16, and, occasionally, 20 nm (). This distribution, with an apparent multiple of 4 nm () corresponding to the size of a tubulin monomer, suggests that the luminal particles could systematically interact with tubulin along the microtubule wall. It also raises the possibility that the particles might associate with both α and β tubulin or bind to a site located between the α and β subunits, possibly along stretches of the same protofilament. Occasionally, stretches devoid of luminal particles were also observed over distances of several tens of nanometers. Some tomograms revealed microtubules with widely flared ends (; compare with the nondepolymerizing end of the microtubule in ), suggesting that they were captured during shrinking, which can take place as a result of the naturally occurring cycles of neurite extension and retraction (). In such shrinking microtubules, the luminal particles showed ∼1.5-fold higher packing densities compared with microtubules in growing neurites ( and ). The luminal particles at the depolymerizing ends reached the minimum measured interparticle distance of ∼8 nm. This packing may explain the “beaded fiber” appearance of some classical transmission EM preparations (). However, the particulate character of the structures becomes obvious in cryotomograms (). In side views, some of the luminal particles show distinct connections with the microtubule wall ( and ). The visual inspection was strongly reinforced by particle-averaging procedures (; ). We analyzed 518 luminal particles using 3D averaging (). All analyzed particles had a very similar overall size of 7 nm; however, the alignment and averaging of such small particles in noisy tomograms failed to uncover fine structural details. Nevertheless, it revealed three subtypes of particles, based on their association with the microtubule wall (). The particles belonging to the first group ( = 193) appeared rodlike and were elongated normal to the microtubule axis. The bulk of the averaged density was located centrally in the microtubule lumen, whereas an elongated protrusion pointed toward and contacted the microtubule wall. The particles assigned to the second group ( = 169) appeared crescent shaped, with two arms protruding radially from a central mass at an angle of ∼140° relative to each other, and with their ends attached to the inner aspect of the microtubule. Particles with no evident connection to the microtubule wall ( = 156) were classified in a third group. The lack of a visible contact does not necessarily indicate that the particles bear no attachment points. One limitation of ET is the restricted range of viewing angles (), resulting in incomplete sampling of data. Connecting stalks could incline parallel to the tomographic Z axis so as to fall into the “missing wedge” of data (), and thus, appear less distinctly in the tomograms. In support of such an explanation, there were ∼30% of “unconnected” particles within microtubules, and the missing wedge within the tomographic data comprises a third of the complete volume. Because microtubules in cells grown on flat supports for cryo-ET are principally oriented perpendicularly to the imaging direction, we could obtain relatively complete representations of longitudinal views. On the other hand, top views of microtubules can be particularly well demonstrated by sectioned material. Therefore, we imaged vitreous sections of neuronal cells in organotypic cultures derived from hippocampus (). In top views of microtubules (), the luminal material appeared as discrete densities, some of which displayed connections to the tubulin wall. In 50-nm-thick sections, we expect an average of 3–4 luminal particles () superimposed in a projected image along the microtubule axis. Occasionally, we also observed transversely sectioned microtubules that were devoid of distinct luminal material (), presumably corresponding to the empty stretches we detected by cryo-ET. In longitudinally sectioned microtubules, the internal material was also clearly observed as discrete particles (), which is consistent with the cryo-ET observations. The luminal densities were also detected with similar frequencies within the microtubules in cell bodies on vitreous sections (), indicating that they are a general microtubule constituent in neuronal cells, and not restricted to the neuronal processes. In contrast to neural tissue, distinct luminal material was not commonly seen in transverse sections of microtubules from cultured rat hepatoma (HTC) cells (). The particulate luminal material was abundant in the microtubules of all examined tomograms from neuronal processes. It was also observed in microtubules from astroglial cells, which are present as a minor constituent in our dissociated hippocampal cultures (). We further examined whether it could also be detected by cryo-ET within the microtubules of stem cells using the pluripotent P19 embryonal carcinoma cell line (Fig. S1, available at ; ). The microtubules of P19 cells contained discrete internal particles (; Video 2, available at ; and ) that were very similar in size to the ones observed in neuronal cells (). However, inside P19 microtubules, the typical stretches of closely spaced particles were more frequently separated by longer particle-free intervals of up to 50 nm. To determine whether the luminal particles are a common constituent of the microtubule lumen in other cell types, we applied similar data collection and tomographic analyses to microtubules in epithelial cell lines, such as PtK2 () and HeLa (), as well as to microtubules nucleated in vitro from purified pig brain tubulin (). There was no evidence of any electron-dense material in the lumen of in vitro–nucleated microtubules (). Consistent with the observations made using CEMOVIS (), the lumen of microtubules in tomograms of epithelial cell lines was mostly free from distinct, particulate densities (); however, occasional traces of electron-dense material could be detected inside microtubules. These densities were irregularly distributed, varied in size, and displayed no obvious symmetry. Based on the known constraints on the signal and resolution, as well as the relatively high level of noise in images and tomograms of intact cells (), we can presently neither confirm nor conclusively rule out the presence of luminal material in such microtubules. In contrast, in neuronal, astroglial, and stem cells we can systematically observe and analyze luminal particles with a consistent size and distribution, irrespective of the noise levels in the tomograms. Measurements both on original tomograms and on the volumes reconstructed by cross-correlation and averaging revealed that the luminal particles had a roughly globular shape with a diameter of 6–7 nm (). A particle of such dimensions would have a molecular mass of at least 200 kD. As the mean distance between particles was 14 nm (), we estimated that ∼6–7% of the total volume available within the neuronal microtubule lumen was occupied by the observed luminal material. At depolymerizing microtubule plus ends with mean distances of 8–9 nm (), the occupied volume increased to ∼10%. The particles were also enriched by 12–21% in the proximal segments of both minor neurites and nascent axons compared with the distal regions of the same processes (). The luminal particles inside the microtubules of P19 cells were also abundant, but the average distance between them (19.3 nm) was somewhat larger than in neurons. We have demonstrated that cryo-ET and CEMOVIS are reliable methods for analyzing the macromolecular architecture of microtubules, as they occur in the cytoplasm of living cells. Currently, it is an open question as to whether the luminal particles we describe represent novel types of MAPs that would be involved in modulating microtubule stability. Another possibility is suggested by the fact that the sole acetylated residue of tubulin (α-tubulin Lys40) has been predicted to reside on the internal surface of microtubules (). It would therefore be intriguing to investigate whether the luminal particles we observed may represent the hitherto elusive tubulin acetyltransferase or a tubulin deacetylase (). The alternative hypothesis that the luminal material (protein or mRNA) may be transported inside the microtubules has already been put forward (; ). Our estimates of the size, shape, and molecular mass of the particles do not support a classical motor–cargo complex; however, novel mechanisms of transport, or even the use of the lumen as a storage space, cannot be excluded. Our work provides a basis for future studies to characterize the biological role and the structure of the luminal particles within microtubules. For primary neuronal cultures, the hippocampi of 17.5-d-old rat embryos were dissected, trypsinized, and dissociated by trituration (). 5 × 10 cells were plated in 35-mm tissue culture dishes containing either EM grids, which were processed as described in the following paragraph, or poly--lysine–coated coverslips. The cultures were first imaged at ∼30°C in HBSS containing 7 mM Hepes, pH 7.25, with an inverted light microscope (Axiovert 135 TV; Carl Zeiss MicroImaging, Inc.), with a 32× air objective (Achrostigmat, NA 0.40; Carl Zeiss MicroImaging, Inc.), using a high-performance charged-couple device (CCD) camera (model 4912; Cohu) and Scion Image 4.0.2 software (Scion Corporation). Detailed maps of the cultivated cells were recorded from every grid before preparing them for cryo-EM by rapid freezing. Rat HTC cells were grown at 37°C, 5% CO, in 50-ml tissue culture flasks (Falcon) containing D-MEM supplemented with 10% fetal calf serum, 2% -glutamine, 50 μg penicillin, and 50 U streptomycin per ml. 400-μm-thick transverse hippocampal slices were prepared from 6–7-d-old rats and maintained for 10–15 d in culture as previously described (). They were high-pressure frozen after a 5-min immersion in medium supplemented with 20% dextran (40 kD) and 5% sucrose, as previously described (). This treatment did not affect the viability of the slices. For cell cultivation and cryopreparation we used finder gold EM grids of 200 mesh, covered with a carbon support containing widely spread small holes (either self-made or obtained from Jena [Quantifoil-R5/20]). The grids were sterilized (UV light for 15 min), coated in a 1 mg/ml solution of poly--lysine, washed in water, and incubated in MEM containing 10% horse serum, which was substituted with N2 medium before plating the neurons (). During preparation, the grids were kept at a temperature of 31–36.5°C. The grids containing cells were mounted in a plunger equipped with a custom-made humidifying device (). After adding fiducial markers (3 μl protein A-gold; Sigma-Aldrich; in N2 medium) the excess liquid was removed from the grids by blotting with filter paper (Whatman Nr 4) from the underside for 30–40 s. The grids were rapidly frozen in liquid ethane slush, cooled in liquid nitrogen to a temperature of −180°C, mounted in a 70° tilt cryospecimen holder (model 626; Gatan, Inc.), and examined in a cryoelectron microscope (CM 300; FEI) equipped with field emission gun and a postcolumn GIF 2002 energy filter (Gatan, Inc.), and slow-scan CCD camera (Gatan, Inc.) with 2048 × 2048 pixels. Low electron–dose series (4,000–5,000 electrons/nm) of typically 60–70 images were recorded using the Digital Micrograph package (Gatan, Inc.) in tilt ranges of ±60 to ±70°, with 2° tilt intervals, at nominal magnifications of 43,000 (0.82 nm/pixel) or 52,000 (0.68 nm/pixel), and with objective lens defocus of 6–10 μm. The areas previously imaged in the light microscope were relocated in the electron microscope using the symbols on the finder grids. The images in tilt series were aligned using fiducial markers and merged in 3D reconstructions by weighted back-projection using the EM program package (). We used this package, as well as the TOM package (), for postprocessing the volumes. We extracted 518 luminal particles from tomograms of microtubules that were differently oriented with respect to the tilt axis (only side views; no microtubule top views were available) using the TOM tools in MatLab (The MathWorks). For the first round of alignment, an in silico–created cylindrical density model was used. The resulting mean volume was used as a reference for further iterative missing wedge–weighted correlation averaging (). For the quantitative analysis, the alignment was focused on the luminal densities using a mask excluding the neighboring luminal densities during the cross correlation. The positions of the luminal densities were then quantified based on the shifts determined by the converged single-particle alignment, providing a list of distances between neighboring particles. Based on previous work indicating that the overwhelming majority of cellular microtubules are composed of 13 protofilaments (), we imposed 13-fold symmetry for alignment and averaging of the microtubule wall. The luminal densities, on the other hand, were averaged without imposing any symmetry. The aligned single particles were sorted into three general groups, based on cross-correlation values, supported by their visual appearance. Averaged volumes within each group were separately refined by further iterative refinements (). For visualization, these averages were then merged together with the 3D density map of the microtubule wall. We used the AMIRA visualization package (Mercury Computer Systems) for surface rendering the microtubules and the luminal densities in the original reconstructions, as well as for displaying the results of particle averaging. The volumes for color displays were selected by adjusting the threshold, and then by removing the noise-dominated parts using automated procedures in AMIRA. The final threshold was set so as to match the 4-nm thickness of microtubule walls. Sample vitrification, cryosectioning, and imaging were carried out as previously described in detail (; ; ). In brief, vitreous samples obtained by high-pressure freezing were mounted in an FCS cryochamber of a microtome (Ultracut UCT; Leica). 50-nm-thick cryosections (nominal thickness; because of compression the final thickness increased to ∼75 nm) were obtained using a 45° cryodiamond knife (Diatome) with a clearance angle of 6°. The sections were observed on a cryotransmission electron microscope (CM100; FEI) at 80 or 100 kV under minimal beam exposure conditions (<1,000 electrons/nm/micrograph). The images of vitreous sections were recorded on film (SO-163 film; Kodak) at various magnifications, and the negatives were scanned on a PRO film scanner (Expression 1680; Epson) with 1600 dpi resolution. Fourier transform calculations of the images were low-pass filtered with a mask, the radius of which corresponded to the first zero of the contrast transfer function (CTF), 1/3 nm, and back projected to real space. Density profiles were determined on rectangular selections of inverted images using the Plot Profile function of ImageJ (National Institutes of Health). Fig. S1 depicts an example of P19 cells in culture. Videos 1 and 2 provide 3D representations of the tomograms shown in and , respectively. Online supplemental material is available at .

xref #text To experimentally implement this concept, we fabricated a microscope stage that permits the positioning of a soft SiN atomic force microscopy (AFM) cantilever in the path of a migrating fish keratocyte (). Particularly critical is the height adjustment: it must block the thin lamellipod without touching the substrate, which would produce a stall that occurs too quickly because of the lamella pushing a cantilever that is not free. The pyramidal tip of the cantilever, which is normally for scanning the sample, could be imaged (), and its position could be measured to subpixel accuracy as the cell deflected it. (also see Video 1, available at ) shows selected frames from the entire time course of the experiment, from initial contact with the edge at t = 0 to deformation of the lamella at t = 30 s, contact with the nuclear mound at 60 s, maximal deflection of the cantilever at 233 s, and ending with release of the cantilever after the cell moves on (Video 1, frame at 260 s). Knowing the stiffness of the cantilever and measuring its deflection (see Materials and methods) permits the load force to be calculated by Hooke's Law as the cell moves against the cantilever (). The load force increases with time until a stall of the entire cell occurs, after which the cell escapes and the load force drops abruptly to zero. The whole cell stall force (stalling the forward translocation of the cell body rather than blocking the lamellipodial leading edge protrusion) is ∼40 nN, which is consistent with previous results on keratocytes using calibrated microneedles (). One of the features in is the period of initial contact, which is highlighted. This section is expanded in . The initial contact of the lamellipod is followed by a rapid increase in load force as the lamellipod pushes the cantilever, eventually stalling after 6–8 s (, highlighted section). One issue was whether a portion of the lamellipod adjacent to the substrate actually slipped under the cantilever as opposed to the cantilever blocking the lamellipod as desired. We investigated this by rapid fixation shortly after the cantilever struck the leading edge of the cells using a protocol described by . Such a cell is seen in stained for filamentous actin by rhodamine-phalloidin. Contact-mode AFM of the same fixed cell () demonstrates that no part of the lamellipod slipped under the cantilever, as bare substratum can be seen where the cantilever indented the leading edge. (It should be noted that the indentation at that time is much larger than the indentations when the protrusion force is measured in the first 10 s after cantilever contact; such indentations are more difficult to visualize). This suggests that the lamellipod behaves as an integral unit consistent with its inherent stiffness (). In addition, the region at the base of the indentation, which actually has closed somewhat by the time fixation has occurred, does not appear to be appreciably higher. The typical height of the lamellipod of a keratocyte is only ∼140–200 nm (). This corroborates the assertion that the contact length of the cantilever where it hits the lamellipod will be ∼3 μm; this is the long dimension of the cantilever at its base (). Note that after the leading edge is stalled locally, the parts of the leading edge adjacent to the stalled region continue to advance and deform, and, on the scale of tens of seconds, the lamellipodial actin network undergoes significant remodeling so that the lamellipod “sneaks around” the cantilever. In ∼20 s, the cantilever hits the mound of the cell body and starts to deflect significantly ( and Video 2, available at ). Three representative examples of the force versus time curves are given in . Because the cantilever is assumed to maintain contact with the protruding lamellipod, differentiating the position of the cantilever as a function of time gives the velocity of that portion of the lamellipod. This enables force-velocity relations to be plotted for each force-time curve as shown in (b, d, and f). The mean stall force (where the cantilever movements stops, at least momentarily) is 1.18 ± 0.35 nN (SD; = 12). shows the variability in individual cells regarding stall force, initial cantilever velocity at the point where it contacts the cell, and velocity of the trailing edge of the cell at the moment the cantilever contacts the leading edge. These two velocities are uncorrelated, and neither of them is correlated with the stall force. A surprising element of this study was that the initial cantilever velocity, rather than being equal to the cell body forward translocation rate, was considerably (by about a factor of seven) reduced (), indicating that the protrusion rate decreases abruptly upon contact. Mechanically speaking, this decrease could be caused by initial contact and loading not visible to us (forces of the order of ≤100 pN would cause deflections so small that they would be undetectable from the images). Chemically speaking, signaling material that accumulated on the cantilever does not appear to be a factor, as the velocity reduction effect is seen with both clean cantilevers and those used multiple times. Rather, the leading edge of the cell appears to slow down just before or at the initial instant of striking the cantilever. To ask when this reduction in lamellipodial velocity occurred, we used reflection interference contrast microscopy (RICM) in which close adhesions at the cell ventral surface and the cantilever as a stationary obstacle could be simultaneously visualized. To avoid pronounced reflections from the gold-coated cantilever, which obscures events when the cantilever is in close proximity to the leading edge, we used uncoated SiN cantilevers. This setup enabled us to observe that lamellipod slowdown occurs within typically ≤2 pixels (corresponding to 232 nm; = 12) in front of the cantilever ( and Video 2). If there is a narrow nonadherent rim of the leading edge that extends beyond the most anterior close contacts visualized by RICM (), slowdown occurs even closer to the cantilever. At this juncture, we favor the idea that the cell mechanically senses the presence of the proximate cantilever and tunes the protrusion velocity and force generation mechanism accordingly (see the last two paragraphs of this section). However, remote sensing by the cell of chemical/electrochemical gradients on the tens of nanometer–length scale cannot be absolutely ruled out. Our provisional interpretation of the stall forces is as follows. If we assume that the 3-μm region of the lamellipod edge in contact with the cantilever is stalled independently of the rest of the lamellipodial edge and that ∼4 pN of force is generated per filament (the elastic Brownian ratchet model predicts ∼2–7 pN per filament; ), there must be ∼100 active filaments impinging on 1 μm of the leading edge. For comparison, V. Small estimates ∼120 filaments per micrometer from electron micrographs of the trout keratocyte leading edge (Small, V., personal communication). have estimated the number of actin filaments in the fibroblast lamellipod to be ∼240 in a frontal area of 176 nm × 1 μm. If we assume that the area in contact with the cantilever is ∼200 nm × 3 μm, effective pressures caused by the actin polymerization at stall can be calculated. For keratocytes, the lamellipodial pressure is ∼2 nN/μm (2 kPa), whereas for fibroblasts, it is ∼10 nN/μm (10 kPa; ). In comparison, the measured polymerization pressure for an actin comet tail modeling that in is ∼1 nN/μm (1 kPa; ; ). shows force-velocity relationships normalized by the unloaded velocity and the stall force. (Note that the initial sharp drop of velocity at forces of the order of ≤100 pN is not depicted). They indicate that at low force, the velocity is insensitive to the load, whereas at high loads, the velocity of the lamellipod decreases sharply similar in form to the recent measurement for in vitro –like actin networks (). The initial force-insensitive region is not caused by a geometric effect of a flat cantilever hitting a curved leading edge (the estimate in supplemental material shows that it would take only a second or so for the part of the cell not initially tangent to the cantilever to hit it, whereas the flat part of the force-velocity curve persists for 5–8 s; available at ). Interestingly, the force-velocity relation we measured conforms neither to recent theoretical models () nor to a previously measured force-velocity relation of actin comet tails. and obtained a convex (bending up) force-velocity relation in which the velocity decreases rapidly at low loads and slowly decreases at a greater force, which is in sharp contrast to our measurement. Only observed a concave (bending down) force-velocity relation; however, in contrast with our data, their velocity increased before it became insensitive to the load, possibly as a result of transient actin growth effects. The elastic polymerization ratchet (, ) as well as the elastic propulsion theory () also predicts a convex force-velocity relation. The theory of autocatalytic branching () predicts a constant protrusion rate that is completely insensitive to force () because greater load indirectly increases the Arp2/3-mediated branching and effective density increase of the actin network. Curiously, in vitro, such a force-velocity relation was measured for a bead undergoing actin-based motility in a purified protein system (), albeit for small forces that we cannot probe. The filament end–tracking motor model () assumes the existence of a motorlike molecular complex at the filamentous actin barbed end and predicts a few possible force-velocity relations, one of which is concave (bending down; ). None of the existing theories predicts the observed complex three-phase force-velocity relation: a sharp drop of velocity at a very small load, a region where velocity is insensitive to low loads, and an abrupt decrease of velocity at large loads and subsequent stall. There are several theoretical possibilities that could explain the observed force-velocity relation. The initial sharp drop of velocity at very small forces of tens of piconewtons per micrometer can be explained in the following ways: weak adhesions at the leading edge that limit the polymerization rate and either slide () or stop to assemble () at very small loads; small osmotic/hydrostatic pressure at the leading edge (); or thermal membrane undulations that can be dampened by small loads (). In principle, the first sharp drop of velocity could be explained by rapid recoil of the softer lamellipodial network when it encounters the stiffer cantilever. If this is the case, this part of the force-velocity relation is not a feature of the lamellipodial network but rather is a result of the measurement technique. These possibilities are discussed in detail in the supplemental material. The insensitivity of the velocity to low loads and its sharp drop at a greater load can be the result of a few possibilities. This could be the result of two sequential processes, one of which is force independent (for example, the chemical reactions associated with adhesion). In that case, at small loads, the force-limited process is much faster than the force-independent process, and the average duration of the step of protrusion is force independent. However, at a greater load, the force-limited process becomes slower than the force-independent one, and the average duration of the step of protrusion increases with the load. Several molecular motors (for example, RNA polymerase [] and kinesin []) and possibly myosin VI () have such force-velocity relations for this reason. Another possibility is that a strong local osmotic/hydrostatic pressure () or gel swelling pressure () is the force-generating mechanism at the leading edge, in which case the velocity would not depend on the load until the pressure at the leading edge is overcome, and then actin polymerization is rapidly stalled, leading to a concave force-velocity relation. Yet another possibility is the force-dependent reinforcement of the dendritic actin network by accelerated branching (). Finally, the elastic ratchet model can explain the concave force-velocity curve if actin filaments at the leading edge are short and rigid (). To summarize, the stall force we measure agrees well with the elastic polymerization ratchet model, but the measured convex force-velocity relation poses a challenge to models of protrusion. In vivo, the force-velocity curve, which was measured with an AFM cantilever as described above, results from an interdependent composite of multiple, possibly redundant mechanisms and limiting factors, including actin polymerization, local osmotic pressure, molecular motors, adhesion, and viscoelastic coupling to regions proximate to the cantilever, rather than a single process such as actin polymerization that can be isolated in in vitro systems (; ). (This interdependence of factors can be seen by the fact that a weak shear flow of only ∼0.01 nN/μm acting on the leading edge of keratocytes, which is much less than what we measure as a stall force, can stop protrusion by probably interfering with nascent adhesions []). Also, in vivo, mechanical contact can trigger local mechanochemical pathways that generate signals, causing the delocalization of polymerization-maintaining complexes. Nevertheless, our results provide the first direct measurements of lamellipodial protrusion force characteristics of a crawling cell and, therefore, represent a mechanical benchmark against which the adequacy of our theoretical understanding of protrusion can be judged. Fish keratocytes were cultured from the scales of rainbow trout (). The scales were removed from the freshly killed fish and transferred into 100 ml of start medium (17.5 ml RPMI 1640 without phenol red, 14 ml Fish Ringers [0.22 mM NaCl, 4 mM KCl, 4.8 mM NaHCO, 2 mM CaCl, and 2 mM Tris], 4 ml FCS, 1.2 ml penicillin/streptomycin, 1 ml of 1 M Hepes, and 1 ml Steinberg medium [0.52 M NaCl, 3 mM Ca(NO), 6 mM KCl, and 8.6 mM MgSO]) and washed several times. The scales were then sandwiched between two microscope slides (Menzel-Gläser) with 200 μl of the start medium and left overnight at 4°C. Clusters of cells grew out from the scales, and these were dissociated by treating with EDTA/trypsin for 30 s and washing with 100 ml of running medium (20 ml Fish Ringers, 1 ml Steinberg medium, and 1 ml of 1 M Hepes). Single cells began to migrate after ∼5 min. All chemicals were purchased from Sigma-Aldrich or Biochrom. Experiments were usually performed within a few minutes after cells started to migrate. Cells were kept at 4°C until they were transferred to the optical microscopes, which were at room temperature. A 200 × 200 × 20-μm piezostage (Physik Instrumente) was integrated into a thick aluminum plate that replaced the stage on an optical microscope (Axiovert 135 TV; Carl Zeiss MicroImaging, Inc.) such that a microscope slide with cells could be micropositioned in x, y, and z (Fig. S1, available at ). A second thick aluminum plate was placed on top of the Axiovert stage and supported by three adjustable points. An AFM cantilever (microlever obtained from Veeco Instruments) was mounted on a plexiglass holder and inserted in a recess of the second aluminum plate so that the cantilever was oriented perpendicular to the substrate (). The spring constant of the cantilevers was measured according to the thermal noise method () and was found to be in the range of 7 mN/m. To adjust the distance between the cantilever and microscope slide, the slide was oscillated in the y direction by driving the piezo with a sine function, and the cantilver was lowered so that it was in contact with the slide. Then the cantilever was retracted until its oscillation disappeared. To compensate for potential drift, we retracted the cantilever further by 80–100 nm. This procedure was applied just before each measurement. Cells were positioned in front of the cantilever, and its deflection after cell contact was recorded with a CCD camera (4912-5100/0000; Cohu) and a videocassette recorder (TL300; Panasonic). RICM was performed on the same Axiovert 135 TV optical microscope as the force measurement and with the same setup for holding the cantilever. A mercury lamp (HBO 50; Carl Zeiss MicroImaging, Inc.), an antiflex slider, and a 63× antiflex objective together with standard oil immersion objectives ( = 1.518; all components were purchased from Carl Zeiss MicroImaging, Inc.) were used. Cells were prepared on coverslips (Omnilab). The coverslips were attached with magnets to a stainless steel holder to allow free access from the top. In this series of experiments, uncoated cantilevers (MLCT-NONM; Veeco Instruments) were brought in contact with the glass slide and served as fixed obstacles. A green bandpass filter (D535/40; Chroma Technology Corp.) was used to avoid damage of the cells by UV light and to enhance contrast in the RICM image. The sequences were recorded with a 12-bit CCD camera (Retiga 4000R FAST Mono; QImaging) and transferred directly to a computer (MacIntosh G4; Apple) via fire wire. The cells were prepared for fluorescence as described previously (). In brief, ∼500 μl of a mixture of 1.5 ml PBS, 0.5 ml of 0.1% Triton X-100 in PBS, 100 μl of 50% glutaraldehyde in water, and ≤100 μl of 100 μg/ml rhodamine-phalloidin in MeOH was introduced shortly after the lamellipod struck the cantilever. After ∼1 min of incubation with the fixation and staining mixture, cells were washed several times in PBS and imaged later with an Axiovert 135 TV microscope. Images were recorded with a Visicam (Visitron Systems) using CamWare 1.26 (PCO AG) and a PC operated with Windows 2000 SP4 (Microsoft). Contact-mode AFM on fixed and stained cells was performed on a microscope (MFP3D; Asylum Research). Force measurements were performed as follows: the recorded video frames were digitized with a frame grabber card (AG-5; Scion Corp.) using ImageJ 1.33u software (National Institutes of Health). The time period of interest (deflection of the cantilever by the lamellipod) was recorded in 768 × 512-pixel images at 25 frames/s. For each image, we averaged five neighboring horizontal lines, which were perpendicular to the cantilever. Because the cantilever is the brightest object in this line, the position of the edge could be defined by looking at a certain gray value. By following this position, we were able to determine the deflection as a function of time. Calibration was performed with a stage micrometer (100 lines per millimeter; Leitz) to generate deflection versus time graphs. These data sets contained a significant amount of noise and were smoothed by using a spline fit option of IGOR Pro (WaveMetrics). The smoothing factor was set to one, and the SD was varied to obtain the best fit. To produce force-velocity curves, the velocity of the tip was obtained by differentiating the smoothed position versus time curve, whereas the force was calculated from Hooke's Law. Velocity versus force curves were normalized by the initial velocity of the cantilever (v) and by the force at the point where v = v/2 (f). Error bars for the force were calculated by using the SD of the deflection. Errors in velocity were not calculated because the position versus time data had to be smoothed before taking the derivative. For AFM, height profiles were obtained after flattening height images with a second order flatten option, which is available in the IGOR software. The recording and analysis were performed with IGOR Pro. For RICM, the images were recorded with a 512 × 512-pixel array and an exposure time of 1 s. The position of the adhesions at the front of the lamellipod, as detected by RICM, were recorded from ∼30 s before cantilever contact to the last visible position of the leading edge (as a result of the high reflection of the cantilever). Using IGOR Pro, kymographs were constructed from the time series of images () by extracting lines from the videos that capture the collision of the leading edge with the cantilever; spatial differentiation of such lines gave sharper edge detection. Calibration was performed with a stage micrometer. Video 1 is a typical video of a trout keratocyte deflecting an AFM cantilever until the cell passes under the cantilever, which then springs back to its initial position. Video 2 is an RICM video showing cell movement before contact with the leading edge of a keratocyte, during initial contact, and during contact with the body of the cell. Fig. S1 is a schematic of the cantilever-positioning stage. Supplemental material provides data on the biophysics of possible force generation mechanisms. Online supplemental material is available at .

Secreted signaling proteins, such as BMPs, Wnts, Hedgehog, and FGFs, play key roles in animal development. Although it is established that reception of these molecules on the cell surface is mediated by heparan sulfate proteoglycans (HSPGs), the mechanism producing selective binding of proteins to heparan sulfate (HS) in a growth factor–rich environment remains a fundamental question. HS is synthesized as disaccharide polymers, which then undergoes a series of modification events including , 2-, 6-, and 3- sulfation. A number of in vitro studies showed that interactions between HS and various growth factors require unique HS structures in which 2- and 6- sulfate groups contribute to generate specific sulfation patterns (for reviews see ; ). Crystallographic studies also supported this biochemical evidence, showing that the 2- and 6- sulfate groups form hydrogen bonds with heparin binding residues of FGFs and/or FGF receptors (FGFRs) to induce dimerization of FGFRs (). Thus, in vitro studies showed that specific sulfation patterns on HS have critical roles in its selective binding to ligand proteins. However, the in vivo importance of these sulfation events is poorly understood. FGF signaling regulates tracheal system formation in (; ). The tracheal precursor cells express Breathless (Btl), a FGF receptor, and migrate toward regions expressing Branchless (Bnl; a FGF) to form primary branches in the embryo. FGF also controls the formation of the adult tracheal system, the air sac, which develops from a group of cells called “tracheoblasts” in the wing disc (). A previous study showed that Btl-dependent activation of MAP kinase relies on (), which encodes -deacetylase/-sulfotransferase (NDST), indicating that HS has a crucial role in these processes (). Because the reaction catalyzed by NDST is the first step in HS modification and is critical for subsequent reactions, mutation of results in the production of sugar chains with no sulfation (). To determine what structural features of HS are required for regulating FGF signaling, we characterized functions of () and () genes during tracheal development. We generated and mutations by imprecise P-element excision (Fig. S1, available at ). The excision alleles, and , delete their respective coding regions, and lethality of and homozygotes was equivalent to that of their deficiency transheterozygotes, indicating that these mutants are null alleles for each gene. Despite the previous implication of 2- and 6- sulfate groups in the binding of HS to many growth factors in vitro (for reviews see ; ), and mutants showed only moderate effects on development. Zygotic and mutants survive to the adult stage without showing obvious morphological defects. We also generated embryos in which both maternal and zygotic gene activities are eliminated (“ null embryos”). Although such null mutations caused partial lethality during development, significant fractions of these null mutants survive to the adult stage without visible phenotypes. This finding demonstrated that loss of either 2- or 6- sulfation does not completely disrupt normal development. Because 2- and 6- sulfations are critical for FGF-HS binding in vitro (for reviews see ; ), we focused our efforts on the function of and in -mediated tracheal migration. The tracheal system develops from clusters of ectodermal cells that invaginate into the underlying mesoderm and form ten sacs on each side of the embryo. Each sac forms six primary branches by stereotypical cell migration. Some of these branches, such as the dorsal trunk, fuse with corresponding branches in neighboring segments to form a continuous tracheal network (). In or mutants, the tracheal cells remain clustered at the site of the tracheal pits without migration (; ). In contrast, we found that maternal and zygotic null mutations of or had only limited effects on tracheal development (). Remarkably, only 9% of null embryos exhibited a stalled migration of the dorsal branch (). A fraction (39%) of null embryos exhibited tracheal defects (). In these mutant embryos tracheal migration is incomplete, as revealed by the presence of large gaps in the dorsal trunks, as well as stalled tracheal branches. The migration defects in these embryos were observed in all primary branches, but most commonly in the dorsal branch and the dorsal trunk. Surprisingly, however, tracheal morphology was indistinguishable from that of wild-type embryos in the remaining 61% of the embryos (). Next, we examined whether and mutations affect the formation of the tracheoblast in the wing disc. Normal development of tracheoblasts was observed in all and most mutants we examined (), although the tracheoblast was slightly reduced in size in a small fraction (18%) of mutant discs (). The small tracheoblast phenotype of mutants was completely rescued by expression from a transgene (). Thus, null mutations in and do not completely block -mediated tracheal formation, showing that and mutant animals can produce HS chains that retain a considerable level of activity to mediate FGF signaling. The modest tracheal phenotypes of the and null mutants clearly challenge a current view on the role of HS fine structures: numerous biochemical analyses have demonstrated that 2- and 6- sulfate groups are critically required for the HS-growth factor interaction (for reviews see ; ). One possible reason for the restricted phenotypes of these mutants is that the sulfation patterns of mutant HS are altered to restore the growth factor signaling. To examine this possibility, disaccharide profiles of HS from and mutant animals were determined using fluorometric post-column HPLC (). In wild-type adult flies, the disaccharide composition of HS showed a similar pattern to representative vertebrate tissues (). In contrast, HS samples from or zygotic mutant adults showed a complete loss of the corresponding disaccharide units, confirming the amorphic nature of these mutant alleles. Significantly, HS disaccharides from mutants showed not only a loss of 2- sulfated disaccharide units, but also a remarkable increase of 6- sulfated disaccharides. Similarly, levels of the 2- sulfated disaccharides are strikingly elevated in mutants. As a result, the level of total sulfate groups on HS was not affected in each case, and the total charge of HS in and mutants was almost wild type (). These results strongly suggested the existence of a compensation mechanism that adjusts the levels of sulfate groups when a component of the HS-modification machinery is lacking. Importantly, similar compensation of HS sulfation has also been observed in mutant mice (), implicating this system as a general property of the HS modification machinery that is widely conserved across species. Thus, the unaltered charge levels on HS in the mutants may contribute to their mild phenotypes, and the function of the 2- sulfate group seems to be replaceable with that of the 6- sulfate group, and vice versa, in some developmental contexts. To confirm the hypothesis that the compensatory increase of sulfation weakens and mutant phenotypes, we performed several sets of experiments in which the compensation was blocked. First, we examined the tracheal phenotypes of ; double-mutant animals. In these animals, the compensation of HS sulfation would not occur due to the absence of both counterparts ( and ) that complement each other in or single mutants. In fact, despite the relatively normal development of the single mutants, the zygotic double mutants are completely lethal. In wild-type embryos, tracheal precursor cells invaginate in each hemisegment at stage 11 (), and migrate and elongate to form primary branches at stage 12 (). One of these branches, the dorsal trunk, fuses with ones in the neighboring segments at stage 14 (). Although invagination seems to occur normally in the embryos, they exhibit several characteristic defects in branching morphogenesis. First, mutant tracheal precursor cells failed to migrate to form the primary branches (). This defect resembles that of or mutants (; ). Second, clusters of mutant tracheal cells tend to extend dorsally and ventrally, forming long, skinny sacs of tracheal precursor cells of various size (). Finally, 16% of mutant embryos showed fusion of the tracheal sacs to those in the neighboring segments (). We asked whether FGF signaling is impaired in these animals using an antibody that specifically recognizes the diphosphorylated form of MAP kinase (dpMAPK; ). In wild-type embryos, dpMAPK is detected in the tracheal placodes at stage 10, reflecting activation of DER, a EGF receptor (). This dpMAPK signal was not diminished in the embryos, showing that signaling is not affected by the double mutations ('; Fig. S2 A, available at ). At stage 12, wild-type embryos show a strong dpMAPK signal in the migrating tip cells of each primary branch due to activation of FGF signaling (). In contrast, the -dependent MAPK activation in the tip cells is disrupted in the embryos ('; Fig. S2 A). In situ RNA hybridization experiments revealed that expression is not altered in the double mutant embryos (Fig. S2 C), confirming that the branching defects observed in the double mutants are caused by disruption of FGF reception but not FGF expression. These results showed that HS with neither 2- nor 6- sulfate groups lost the ability to mediate Btl signaling. Next, we examined whether simultaneous loss of both 2- and 6- sulfate groups affects tracheoblast formation in the wing disc. Because mutants die during embryogenesis, we analyzed homozygous animals bearing a transgene that expresses double-stranded RNA for ( RNAi) under a specific Gal4 driver (). Tracheoblast development was not affected either by homozygosity of the null mutation () or by expression of the RNAi construct (unpublished data). In contrast, RNAi in -expressing (tracheal) cells in homozygous mutant background completely blocked the formation of the tracheoblast (). No such effect was observed, however, when the RNAi was induced in -expressing (nontracheal) cells in the same mutant background (). Thus, HS requires either 2- or 6- sulfate groups for reception of FGF, but these modifications are not essential in the FGF-expressing cells. Collectively, tracheal development could occur in or single mutants, but not in the double mutants. These findings demonstrated redundant roles of 2- and 6- sulfate groups of HS in FGF signaling during tracheal development. As another approach to reduce 6- sulfation without inducing an increase of other sulfation events, we examined the effects of overexpressing , a extracellular sulfatase (CG6725), on FGF signaling. Vertebrate genes encode secreted HS 6- sulfatases, which remove sulfate groups from the HS on the cell surface (). Because Sulf1 seems to modify HS fine structure extracellularly, and we hypothesized that compensatory changes in sulfation occur during HS biosynthesis in the Golgi, we expected that the number of sulfate groups on HS in -expressing animals would decrease. Indeed, this was the case. Disaccharide profiling of HS from animals showed a significant reduction in the level of 6- sulfation without the compensatory increase of other sulfate groups (). As a result, the total sulfate level is reduced in these animals to 76.3% of the wild-type level. Importantly, overexpression of had stronger effects on viability and FGF-mediated tracheogenesis than mutations. - animals showed 71% lethality (unpublished data). The tracheoblast was dramatically reduced in size by expression of in -expressing (tracheal) cells (). The fact that -expressing animals show more severe phenotypes than null mutants strongly suggests that the compensatory increase of 2- sulfation in mutant HS restores the ability to mediate FGF signaling. From these findings, we conclude that biosynthesis and modification of HS show a striking flexibility. In the absence of a component of the HS-modification machinery, living cells can form HS that lacks normal fine structures but retains normal levels of sulfate groups and a considerable level of activity for growth factor signaling. Numerous in vitro studies have identified various ligand proteins that bind to specific sulfated HS sequences. Recent studies using animal models have also highlighted the importance of distinct HS sulfation patterns for HSPG functions (). Thus, it is widely accepted that a specific sequence of sulfation on HS determines a binding site for a ligand, enabling HSPGs to interact selectively with proteins. However, it is not known how strictly ligand binding sites are defined in vivo. Our study demonstrated that living cells show an unexpected level of flexibility in biosynthesis and function of HS. In vivo HS sulfation is flexible in two ways. First, HS modifications can be adjusted in response to a defect in one type of sulfation. Second, mutant HS chains thus synthesized, which do not contain normal sequences of sulfate groups but bear normal levels of sulfation, do not completely lose coreceptor activity for growth factor signaling. We found that mutations induce compensatory increases in sulfation at other positions, restoring a wild-type net charge on HS. embryonic fibroblasts did not have 2- sulfate groups, but this loss was compensated for by increased - and 6- sulfation. mice. Our study provides evidence that the HS compensation indeed contributes to the modest phenotypes of animals deficient for these HS-modifying enzymes. The ability of the mutant HS to mediate signals is achieved, at least partly, by the sulfation compensation system because HS loses this ability when the compensation is blocked. These observations suggest that some in vivo roles of HS require a sufficient amount of sulfate groups but not a strictly defined placement on HS. This idea is supported by a recent biochemical study showing that binding of FGF to HS is dictated primarily by charge density rather than by the precise positioning of various sulfate groups (; ). On the other hand, in different biological processes, specific sequences play essential roles in generating specificity of HS–protein interaction. In particular, sulfation at the 3- position of the glucosamine residue, the rarest component of HS sulfation, is critically required for the binding site for antithrombin III () and a coat glycoprotein of herpes simplex virus (). Collectively, the mechanism for in vivo HS–protein interactions may occur by several mechanisms: some proteins bind specific fine structures; some proteins are attracted to the charge on HS but have less strict structural requirements; and some proteins bind to HS based on a combination of specific sequence and charge density. Further studies will define ligand proteins in each class as well as the nature of their binding to HS. The detailed information for fly strains used is described in Flybase (), except where noted. All flies were maintained at 25°C. The following strains were used: Oregon-R, wild-type strain; (see the Gene Search Project web site: ) and , P-element insertion lines for and , respectively; and , null mutants for and , respectively (see below for mutant isolation); , a null allele of ; (breakpoints, 37D02-E01; 37F05-38A01) and (breakpoints, 92B02-03; 92C02-03), chromosomal deficiency lines; , an enhancer trap line for the () gene. The transgenic animals used were as follows: ; ; ( cDNA (SD04414, Berkeley Genome Project) was fused to a 344-bp genomic PCR fragment to complete the coding region and inserted into pUAST vector); (see below for construction of transgenic RNAi flies); ; ; ; and (strain number 2211; Genetic Resource Center, Kyoto Institute of Technology, Japan). . Their progeny were screened for loss of marker gene expression. Excision chromosomes were analyzed by PCR using flanking primers to find deletions, and the extent of each deletion was determined by sequencing PCR products that spanned the junction (see the legend to Fig. S1 for details). Lethality of and homozygotes (3.8% and 43%, respectively) was equivalent to that of their deficiency transheterozygotes (, respectively), indicating that these mutants are null alleles for each gene. Embryos lacking maternal and zygotic activity of were obtained by crossing homozygous females to males. To obtain maternal and zygotic mutant embryos, germ line clones were generated using the autosomal FLP-DFS technique (). Females carrying were mated with . The resultant maternal and zygotic mutant embryos were identified with marked balancer. Transgenic RNAi flies of were obtained as described previously (). A 500-bp-long cDNA fragment from the first methionine was amplified by PCR and inserted as an inverted repeat (IR) into a modified pBluescript vector, pSC1, which possesses an IR formation site. IR-containing fragments were subcloned into pUAST, a transformation vector, and transformation of embryos was performed using as a recipient strain. Antibody staining was performed as described previously () using rabbit anti–β-galactosidase (1:500; Cappel) and mouse anti-diphosphorylated MAP kinase (1:200; Sigma-Aldrich). The primary antibodies were detected with Alexa Fluor–conjugated secondary antibodies (1:500; Molecular Probes). For quantitative analysis of MAPK activation, the percentage of segments that show normal dpMAPK staining in tracheal precursor cells (stage 10 wild type, = 12; stage 10 , = 18; stage 12 wild type, = 27; and stage 12 , = 21) was calculated. In situ RNA hybridization was performed as described previously (). Light microscopy images were taken using a microscope (model BX50; Olympus) with a 40×/0.75 UPlanFl objective by a CCD camera (DP-50; Olympus) controlled by Studio Lite software. Confocal imaging was performed using a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with a 40×/0.75 Plan-Neofluar objective equipped with a confocal microscope system and a software (LCM5 PASCAL; Carl Zeiss MicroImaging, Inc.). Images were processed using Photoshop 7.0 (Adobe). HS disaccharide was analyzed by fluorometric post-column HPLC as described previously (). Approximately 50 mg of lyophilized adult flies was used to isolate HS. The HS sample was digested with a heparitinase mixture (Seikagaku) and subjected to a reversed-phase ion-pair chromatography. Fig. S1 shows the molecular characterization of and mutants. Fig. S2 shows the quantitative analysis of MAPK activation and in situ RNA hybridization of mRNA in wild-type and embryos. Online supplemental material is available at .

The multicopy yeast plasmid 2-μm circle is organized as a cluster of approximately three to five foci in the nucleus and segregates as a cluster during cell division (; ; ). The plasmid-coded Rep1 and Rep2 proteins together with the cis-acting locus are responsible for the nearly chromosome-like persistence of the 2-μm circle. The plasmid provides no obvious advantage to yeast but poses no significant disadvantage either at its steady state copy number. An amplification machinery based on the Flp site-specific recombination system can correct a potential drop in copy number as a result of rare missegregation events (; ; ). The amplification system is under both negative and positive controls. The negative control involves repression of the gene by the Rep1 and Rep2 proteins acting, presumably, as a bipartite repressor (). The positive control is mediated through the plasmid protein Raf1p, which appears to antagonize the action of the Rep1p–Rep2p repressor. Together, the partitioning and amplification systems can account for faithful plasmid propagation as well as maintenance of plasmid copy number. The Rep- partitioning system appears to couple 2-μm circle segregation to chromosome segregation using quite unsuspected mechanisms. The plasmid assembles the yeast cohesin complex at in a Rep1p–Rep2p-dependent fashion, presumably to pair replicated plasmid clusters during S phase (). Strong circumstantial evidence suggests that sister clusters segregate when cohesin is disassembled by separase action during anaphase. Therefore, the mechanics of plasmid segregation and chromosome segregation appear to be fundamentally similar. In contrast to chromosomes, cohesion between plasmid clusters is absolutely dependent on the integrity of the nuclear microtubules (). In the mutant, in which multiple chromosomes are detached from the spindle during a cell cycle (), the 2-μm plasmid tends to cosegregate almost always with the spindle and, thus, spindle-attached chromosomes (). These observations are consistent with a spindle-dependent and perhaps chromosome-assisted segregation mechanism for the plasmid. The partitioning locus can be divided into two halves, proximal and distal, with respect to the plasmid replication origin (). The proximal consists of a tandem array of approximately six units of a 65-bp consensus sequence. The distal contains a transcription termination signal that prevents plasmid transcription directed toward the origin from entering the repeated segment of . The SWI/SNF-related yeast chromatin remodeling complex RSC2 () is required for proper chromatin organization at (; ). A lack of functional RSC2 complex prevents cohesin assembly at and causes high plasmid loss (; ; ). In this study, we demonstrate that the histone H3 variant Cse4p (the centromere protein A homologue), which was thought to be exclusive to chromatin at yeast centromeres, is harbored by and promotes equal plasmid segregation. In wild-type [cir] yeast cells (harboring the native 2-μm circle), a fluorescence-tagged reporter plasmid resides in close proximity to the spindle pole (). This characteristic localization is lost in the absence of an intact partitioning system, as in a [cir] strain lacking the Rep1 and Rep2 proteins. The plasmid foci were almost always coincident with or partially overlapped the kinetochore marker Ndc10p (unpublished data), suggesting that the 2-μm circle shares more or less the same nuclear locale as centromeres. Consistent with these observations, the Cse4 protein and Rep proteins were found to colocalize in chromosome spreads prepared from [cir] cells (). To distinguish between the true association of Cse4p with the 2-μm plasmid and mere overlap between the plasmid cluster and congressed centromeres, we performed the chromosome spread assays in the [cir] and [cir] strains at permissive (26°C) and nonpermissive (37°C) temperatures. Instead of the compact focus of Cse4p seen at 26°C, the protein became disbursed over the DAPI staining region at 37°C, often with a punctuate pattern (). This delocalization was presumably caused by the centromeres being disorganized in the absence of Ndc10 function (). However, in the [cir] strain, a subset of these Cse4p dots showed a near perfect one to one correspondence with the foci formed by a resident reporter plasmid (). The function of the endogenous 2-μm circles in this assay was to provide the Rep1 and Rep2 proteins. As demonstrated previously, in the absence of either Rep1p or Rep2p, the plasmid failed to localize to chromosome spreads (; and unpublished data). When Ndc10p was inactivated, Cse4p was not associated with centromeres, as assayed by chromatin immunoprecipitation (ChIP; see next section; ). Furthermore, the authentic kinetochore protein Ctf19p was absent from chromosome spreads under this condition (Fig. S1, available at ; ). There is precedent for the mislocalization of Cse4p to inappropriate chromosomal loci. observed dispersed Cse4p in chromosome spreads in the background. Consistent with this observation, the inactivation of Spt4p, whose localization to kinetochores is dependent on Ndc10p, also causes the mislocalization of Cse4p to noncentromeric locales (). To probe the suspected Cse4p-plasmid association more critically, we performed ChIP in a [cir] strain expressing myc-tagged Cse4p using myc-directed antibodies. We detected the presence of DNA in the immunoprecipitate by PCR (, first row) but failed to amplify other regions of the 2-μm plasmid (, second to fifth rows). At higher inputs of template DNA, a low level of PCR product was observed for the origin region (, sixth row) but not for other plasmid loci (not depicted). This background was not unexpected for DNA fragments sheared to a mean length of 500 bp because the origin is located ∼340 bp from the proximal end of . Centromeric DNA was also immunoprecipitated, as exemplified by the amplification of (, seventh row). The Cse4p- association was dependent on both Rep1 and Rep2 proteins (). It was positive in a [cir] strain harboring an reporter plasmid and simultaneously expressing the two proteins but was negative in a [cir] strain or its derivatives expressing either Rep1p or Rep2p alone (). The Rep1 and Rep2 proteins interact with each other, and each one interacts with (; ; ; ). At least a subset of these protein–protein and DNA–protein interactions is critical for equal segregation of the 2-μm circle. A Rep1p mutant Rep1(Y43A) that interacts with but not with Rep2p () was unable to support Cse4p- association (, second row). In contrast, a second mutant, Rep1(K297Q), that interacts with Rep2p but not was active in enlisting Cse4p at (, first row). ChIP analysis has validated the inference from chromosome spread assays () that an intact kinetochore or Cse4p- association is not a prerequisite for Cse4p recruitment at . Cse4p was detected at in the mutant at both permissive and nonpermissive temperatures (, second row), whereas it was absent at under the latter condition (, compare the first and second rows). The nuclear mitotic spindle plays a key role in the clustered organization and precise nuclear localization of the 2-μm circle, cohesin assembly at , and equal plasmid partitioning (). We wished to know whether spindle integrity is essential for the acquisition of Cse4p by the plasmid. Depolymerization of the spindle using nocodazole abolished Cse4p- association, whereas Cse4p- association was not affected by this treatment (, compare the first and second rows). The incorporation of Cse4p specifically at within the 2-μm circle genome as well as the requirement of the Rep proteins and an intact spindle for this event implies a potential role for Cse4p in 2-μm circle segregation (see last section of Results). Is Cse4p an integral unit of the nucleosome core on which the chromatin is assembled? We addressed this issue using variations of two assays that were previously used to establish the presence of Cse4p in centromeric nucleosomes. One is based on the ionic strength of buffers required to extract histones out of intact nucleosome oligomers that were isolated from yeast chromatin (). The other relies on the differential sensitivity of chromatin to the DraI restriction enzyme under conditions in which Cse4p is functional or nonfunctional (). In a modified ChIP assay, we pretreated yeast spheroplasts with increasing NaCl concentrations (from 0 to 2.0 M) before performing immunoprecipitations with antibodies directed to myc-tagged Cse4p. As revealed by PCR amplification of the immunoprecipitated DNA, Cse4p association with could withstand up to 1 M NaCl; at 1.5 and 2.0 M NaCl, was denuded of Cse4p (, first row). The results for were quite similar: Cse4p- association was markedly weakened at 1.5 M NaCl, and it was undetectable at at 2.0 M NaCl (, second row). Furthermore, a Western blot analysis of the elution profiles of Cse4p and histone H3 from chromatin were in agreement with the ChIP results (). Total release of Cse4p into the supernatant fraction required 1.50 M NaCl; in comparison, H3 release was virtually complete at 1.0 M NaCl. In the restriction enzyme sensitivity assay, we followed in isolated nuclei the extent of digestion at three DdeI sites located within a PstI–XbaI region of the 2-μm circle genome that covers (). Two of the DdeI sites are harbored within the -proximal segment of , whereas the third one lies just outside of adjacent to the AvaI site. Nuclei were prepared from two genetically matched yeast strains, one containing and the other containing its temperature-sensitive allele (), after a 3-h shift to 37°C and were treated with a limiting amount of DdeI for different durations. DNA was isolated, cut completely with PstI plus XbaI, and the composite digestion profiles were displayed by gel electrophoresis and Southern blotting (). From the 10–30-min period, the extent of DdeI digestion represented by the abundance of the three lower fragments (632, 508, and 383 bp) relative to the intact PstI–XbaI fragment (1,289 bp) was clearly higher in the background compared with the wild type (). This is graphically represented by ratios of the intensities of the 383-bp fragment to those of the 1,289-bp fragment at different time points (, top). The same result is also conveyed by plotting the relative abundance of these two fragments with increasing digestion time (, bottom). Whereas the half-life of the PstI–XbaI fragment was ∼20 min for the wild-type chromatin under these conditions, it was <10 min for the mutant. This contrast was not detected when a different region of the 2-μm plasmid, namely the coding region, was probed by a similar procedure (Fig. S2, available at ). As revealed by ChIP, the mutant Cse4 protein was not dislodged, at least not completely, from or after the transfer of cells to the nonpermissive temperature (; and unpublished data). However, centromere function is compromised at the elevated temperature, leading to high rates of chromosome loss (). The present analysis suggests that the mutant protein, although present at , cannot support normal nucleosome organization at this locus. The increased DdeI accessibility of in the cells relative to wild type was not caused by differences in their cell cycle stages. The mutation triggers the spindle checkpoint, causing the cells to stall in G2/M. However, as revealed by the strain, there was little difference in the DdeI digestion patterns at in cells cycling normally at 26°C or trapped in metaphase at 37°C (Fig. S3, available at ). The sum of the outcomes from the high-salt ChIP and high-salt protein release assays combined with the DdeI susceptibility analysis strongly argues for Cse4p being an authentic nucleosome component of chromatin. Localization of Cse4p selectively to centromeres is mediated through locale-specific protection of the protein from ubiquitin-mediated proteolysis (). Does a similar protective mechanism apply to Cse4p at the locus? We followed Cse4p in wild-type cells, induced for its expression for a 2-h period from the promoter, and then treated with cycloheximide by Western blotting and ChIP. The former assay would assess the gross steady state level of Cse4p, whereas the latter would monitor its specific association with (). The cellular pool of Cse4p was depleted quite rapidly, with a half-life of ∼15 min; however, its association with was unaffected during the 60-min period of the assay. As expected from a previous study (), Cse4p-centromere association was also independent of the global destruction of the protein. We surmise that the logic of Cse4p localization at the centromeres and is the same: namely, shielding it against proteolytic degradation. During the yeast cell cycle, newly synthesized Cse4p replaces the old protein at centromeres at the time of DNA replication, as revealed by FRAP analysis (). Once incorporated, Cse4p remains stable within the centromere chromatin throughout mitosis. The timing of Cse4p exchange presumably coincides with the low frequency orientation switching of sister kinetochores between spindle halves. We have compared the cell cycle dynamics of Cse4p at and centromeres using a time course ChIP assay. Consistent with the results of the FRAP analysis (), we detected the presence of Cse4p at during G1, a brief absence during the G1→S transition, and subsequent stable Cse4p- association (, row 1) through the remainder of the cell cycle. The analysis presented in followed this association for a period of 60 min after release from G1. In contrast, Cse4p was not present at in G1-arrested cells (, row 1). However, Cse4p- association was established at the same time during the cell cycle as was Cse4p- association (, A and B; row 1; 20 min). Furthermore, this timing matched that of the reassociation of Rep1p (or Rep2p) with after its ejection from this locus at the time of G1 exit (, row 2; and not depicted; ). These observations were concordant with very short interval ChIP assays performed from 0 to 20 min with a twofold increase in the template DNA during PCR (, A and B; insets above row 1). After recruitment, Cse4p association with remained stable until late telophase (see next paragraph). To examine more carefully the difference between and in their cell cycle–dependent Cse4p association, we followed cells arrested in telophase and released into a synchronous cell cycle (). Cse4p was absent at at 15 min after release, which is at the time of spindle disassembly but before the completion of cytokinesis (). The dissociation of Cse4p from occurred later after cell division had been completed and bud emergence was just about to be initiated in the following cell cycle (). Results from ChIP assays with better time resolution in the 0–45-min interval and a twofold higher template input during PCR agreed with this inference (). When telophase-arrested cells were released in the presence of α factor to block cells in G1, Cse4p exit from was prevented (). Thus, although the de novo association of Cse4p during the cell cycle occurs coincidentally on yeast chromosomes and the 2-μm circle, the lifetime of the associated state is shorter in the case of the plasmid. The release of Cse4p from does not occur until well past the completion of plasmid segregation. Consistent with a role for Rep1 and Rep2 proteins in Cse4p recruitment by the plasmid, the cell cycle timing of renewal of the association of all three proteins with is the same. The centromeric nucleosome containing Cse4p is thought to provide a platform for the organization of the kinetochore complex to initiate the spindle attachment of chromosomes during their segregation (; ). Is equal segregation of the 2-μm plasmid dependent on the presence of a Cse4-containing nucleosome at ? We assayed the segregation of a fluorescence-tagged reporter plasmid in wild-type and cells at 26 and 37°C, respectively. Approximately half of the cells in the population at 37°C were in G2/M; the others had escaped the delay induced by the spindle checkpoint. Among the latter cells, plasmid foci were counted in the large budded subpopulation showing distinct DAPI staining zones in the two cell compartments (indicating that nuclear elongation had been accomplished and chromosome segregation had been completed, at least at a gross level; ). However, because of the mutation, a majority of this population (∼75%) showed the clear inequality of DAPI between the cell compartments. An equal number of fluorescent foci in the mother and daughter compartments (4:4 and 3:3, etc.) was scored as normal plasmid segregation. Plasmid missegregation was represented by the collective set of cells showing an unequal number of foci in each compartment. In the mutant strain, the fraction of cells missegregating the plasmid increased from 29% at the permissive temperature to 67% at the nonpermissive temperature, approaching the missegregation rates observed in a [cir] wild-type strain (lacking the Rep proteins). Complete missegregation (:0) was rarely observed in the mutant at the nonpermissive temperature. As a control, we assayed equal and unequal segregation (1:1 and 2:0, respectively) frequencies of a reporter plasmid in the background at the permissive and nonpermissive temperatures. Consistent with a role for Cse4p in chromosome segregation, the missegregation rate of the plasmid was elevated at 37°C (Fig. S4 A, available at ). Finally, to circumvent the potential effects caused by cell cycle delay or arrest, we determined plasmid missegregation frequencies in a Δ double mutant at 26 and 37°C. The results were not altered significantly by the inclusion of Δ (Fig. S4 B). According to the currently favored model for plasmid segregation, pairing of duplicated plasmid clusters by the cohesin complex assembled at and their separation by cohesin disassembly later on are critical steps in equal partitioning of the 2-μm circle (, ). Therefore, we wondered whether plasmid missegregation in the mutant could result from the failure of cohesin recruitment or maintenance at . Results of the ChIP analysis shown in demonstrate that in the mutant strain, the cohesin component Mcd1p was associated with at 26°C but not at 37°C. As pointed out earlier in the paper, the association between the mutant Cse4 protein and is retained at the nonpermissive temperature even though cohesin recruitment is blocked, and plasmid partitioning is adversely affected under this condition (). Also note that the acquisition of cohesin by the 2-μm circle absolutely requires the interaction of Rep1 and Rep2 proteins with each other and at least that of Rep1p with (). We wished to know whether the association of one or both of these proteins with might be disrupted by the mutant Cse4p. The results of ChIP assays performed in the strain using antibodies to Rep1p or Rep2p revealed the presence of Rep1p at at 26 and 37°C; however, Rep2p was not associated with at 37°C (). One of the attributes of an -containing plasmid, which appears to be functionally relevant for its normal segregation, is its ability to localize to yeast chromosome spreads (, ). This may potentially represent direct tethering of the plasmid cluster to a chromosome, which is in agreement with a possible hitchhiking model for plasmid segregation (). Alternatively, the plasmid cluster and chromosomal domains may share common anchoring points on the nuclear substratum without direct tethering between the two. Regardless, plasmid foci are not detected in the vast majority of chromosome spreads when either of the two Rep proteins is missing or when the integrity of the mitotic spindle is tampered with (, ). Consistent with the inability of the mutant Cse4p to support Rep2p- association at the nonpermissive temperature, chromosome spreads from the - strain were predominantly bereft of an reporter plasmid at 37°C (). In contrast, a reporter lacking was not detected in most of the chromosome spreads either at the permissive or nonpermissive temperature (unpublished data). The yeast chromatin remodeling complex RSC2 is required for establishing the functional chromatin state of the locus, stable association of Rep1p with , and acquisition of cohesin by the 2-μm circle (; ; ). Consistent with these observations, the plasmid loss rate becomes highly elevated in an Δ background (). It is plausible that remodeling of the chromatin by the RSC2 complex takes place only in the context of Cse4p-containing nucleosomes. To test this idea, we assayed the association of the RSC2 components Rsc2p and Rsc8p with after the inactivation of Cse4p. As shown by the ChIP results in , both proteins were associated with in the background at 26°C, but neither one was associated at 37°C. The effects of Cse4p inactivation on association with the cohesin subunit Mcd1p, Rep proteins, and components of the RSC2 remodeling complex () are not caused by the relatively large fraction (nearly 50%) of G2/M cells in the population at 37°C. Our previous study () demonstrated that Mcd1p association with lasts from S phase until anaphase. We also found that the Rep proteins are associated with throughout the cell cycle except during the window between the late G1 and early S phase (). We have shown that both Rsc2p and Rsc8p are associated with in cells arrested in metaphase at 37°C (Fig. S4 C). The significant fraction of metaphase cells in the population at 37°C should have displayed Mcd1p, Rep2p, Rsc2p, and Rsc8p at if Cse4p is not required for their recruitment/maintenance. Finally, even when the metaphase delay was relieved in the Δ host, Rsc2p associated with at 26°C but failed to do so at 37°C (Fig. S4 D). Thus, when Cse4p is inactivated, the Rep2 protein does not associate with , the 2-μm circle is not displayed in chromosome spreads, and cohesin assembly at is blocked. At least part of the effects of inactivating Cse4p is manifested through the lack of functional remodeling of the chromatin by the RSC2 complex. In a previous study, we noticed that it is the Rep1 (and not Rep2) protein that fails to associate with in the Δ background (). Overall, these observations suggest that before RSC2-mediated remodeling, the distinct states of organization conferred by wild-type Cse4p and its mutant counterpart exclude Rep1p and Rep2p, respectively, from occupying this locus. It is only the remodeled chromatin containing functional Cse4p that can nucleate the stable tripartite Rep1p––Rep2p interactions required for faithful plasmid segregation. xref #text The yeast strains and plasmids used in this study are listed in Table S1 (available at ). The reporter plasmids containing Lac operator arrays has been described previously (; ; ). Synchronized cell populations were obtained by α-factor arrest in G1 phase followed by release as described previously (). Cell cycle arrest at telophase was affected by shifting log phase cultures of the mutant strain from 26 to 37°C for 2.5 h. The majority of ChIP analyses were performed essentially as described previously (; ). After immunoprecipitation, the beads were washed five times at room temperature as follows: twice with 1 ml of immunoprecipitation wash buffer I (50 mM Hepes-KOH, pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.1% sodium deoxycholate, and 1% Triton X-100) for 5 min each, once with 1 ml of immunoprecipitation wash buffer II (50 mM Hepes-KOH, pH 7.5, 500 mM NaCl, 1 mM EDTA, 0.1% sodium deoxycholate, and 1% Triton X-100) for 5 min, and once with 1 ml of immunoprecipitation wash buffer III (10 mM Tris-HCl, pH 8.0, 250 mM LiCl, 1 mM EDTA, 0.5% sodium deoxycholate, and 0.5% NP-40) for 5 min. Finally, the beads were washed with 1 ml Tris-EDTA for 10 min at room temperature. The additional washing steps resulted in a particularly clean PCR background in the negative samples even though there was some loss of signal in the positive ones. Serial dilutions of the template DNA were used in PCR reactions to verify that the yield of the amplified DNA was in the linear range with respect to the input template. For ChIP assays at the nonpermissive temperature, log phase cultures of , , and strains grown at 26°C were preincubated at 37°C before performing formaldehyde cross-linking at this temperature. The durations of thermal shift were 3 h for and and 2.5 h for . In assays designed to test the effect of high salt on Cse4p association with chromatin, the ChIP protocol was modified as follows. Exponentially growing yeast cells were harvested and washed once with spheroplasting buffer (1 M sorbitol and 0.1 M KPO, pH 7.4). They were resuspended in the same buffer containing 70 mM β-mercaptoethanol and 50 μg/ml 100T zymolyase (U.S. Biological) and incubated at 30°C for 60 min. The spheroplasts were harvested and treated with various salt concentrations (0–2.0 M NaCl) in spheroplasting buffer for 1 h at 4°C. They were collected by centrifugation, washed, resuspended in spheroplasting buffer containing 1% formaldehyde, and carried through further steps of ChIP (; ). For probing Cse4p-chromatin association, ChIP assays were performed with antibodies to the myc epitope using a yeast strain that harbors an engineered gene expressing myc-tagged Cse4p (). In the ChIP assays shown in , antibodies to the HA epitope tag present on the Mcd1 protein were used. Antibodies against native Rep1 and Rep2 proteins were used for monitoring their association with the DNA. For detecting Rsc2p and Rsc8p at , antibodies to the myc epitope harbored by them were used. Nuclei were isolated according to and were extracted with high salt as described by . After resuspending nuclei in 10% Ficoll, 20 mM phosphate buffer (potassium as counter ion), pH 6.5, 0.5 mM MgCl, and protease inhibitors (Complete Mini; Roche), they were treated for 30 min on ice with 3 vol of extraction buffer (0.5 M sucrose, 20 mM potassium phosphate, pH 6.5, 0.5 mM MgCl, 1% Triton X-100, and protease inhibitors) that yielded a final concentration of 0–2.0 M NaCl. The samples were spun at 16,000 for 30 min at 4°C. The pellet proteins and trichloroacetic acid–precipitated supernatant proteins were fractioned by 10–15% SDS-PAGE in preparation for Western blotting. Cse4p fused to the myc epitope was detected by the same antibodies as those used in the ChIP experiments. Histone H3 was detected using a polyclonal antiserum. Restriction endonuclease susceptibility of intact chromatin was assayed using modifications of the procedure outlined by . The wild-type and mutant cells were grown to early to mid-log phase at 26°C and were shifted to 37°C for 3 h. Cells were harvested, and nuclei were isolated as described previously (). These nuclei were resuspended in NEbuffer 3 (New England Biolabs, Inc.) and treated with DdeI at 50 U/ml for various times. The nuclease-digested samples were subsequently deproteinized, and DNA extracted from them was digested to completion using XbaI plus PstI or XbaI and PvuI. After electrophoresis in agarose gels, DNA was transferred to Hybond-XL membrane (GE Healthcare) and hybridized to specific radiolabeled probes. Bands were visualized using a phosphorimager (Molecular Imager FX; Bio-Rad Laboratories), and individual band intensities were quantitated using Quantity One software (Bio-Rad Laboratories). Similar assays were performed in the strain using cells from a mid-log phase culture at 26°C and those shifted to 37°C for 3 h. Antibodies against the myc epitope (9E10) and HA epitope (HA.11) were obtained from Covance. Polyclonal antibodies against histone H3 and the HA epitope were purchased from Abcam and Sigma-Aldrich, respectively. All other antibodies used were described previously (). The preparation of chromosome spreads, fluorescence microscopy of reporter plasmids, and indirect immunofluorescence assay of microtubules have been described previously (; ). For obtaining chromosome spreads at the nonpermissive temperature, and cultures grown at 26°C were preincubated at 37°C for 3 h. Observations of live or fixed cells and chromosome spreads was performed using a microscope (BX-60; Olympus). The images were captured with a camera (Photometrics Quantix; Roper Scientific) and were processed with MetaMorph software (Universal Imaging Corp.). Table S1 provides a list of yeast strains and plasmids used in this study. Fig. S1 shows the localization of Cse4p and Ctf19p in chromosome spreads in the strain. Fig. S2 shows that chromatin integrity within the locus is not affected by the inactivation of Cse4p. Fig. S3 shows the accessibility of chromatin in cycling and metaphase-arrested cells probed by sensitivity to DdeI. Fig. S4 shows the segregation of centromere and 2-μm reporter plasmids in the background as well as the association of Rsc2 and Rsc8 with in and Δ strains. Online supplemental material is available at .

Oocytes arrest at metaphase of the second meiotic division (MetII) before fertilization because of an activity termed cytostatic factor (CSF; ; ; ). Sperm break this arrest via a Ca signal (; ; ), and in so doing, oocytes complete the second meiotic division before entering the embryonic cell cycles. CSF activity, a terminology that was first defined several decades ago (), is now known to constitute an inhibitor of the anaphase-promoting complex/cyclosome (APC; ). The APC is an E3 ubiquitin ligase whose activity is required for the metaphase–anaphase transition to polyubiquitinate key cell cycle proteins, thereby earmarking them for immediate proteolysis through association with its key coactivator cdc20 (; ; ; ). The reduced APC activity in MetII oocytes prevents the destruction of both M-phase (maturation)–promoting factor (MPF) activity (CDK1/cyclin B1) and cohesin, which holds sister chromatids together (; ). Resumption of meiosis in mammalian oocytes is achieved by a sperm-borne phospholipase C activity (; ), which generates an oscillatory Ca signal, switching on APC (; ) through a signaling pathway involving calmodulin-dependent protein kinase II (CamKII; , ; ). This signaling process is conserved and was first demonstrated in frog eggs (, ). Activation of the APC in MetII oocytes induces the destruction of MPF and sister chromatid cohesion through the polyubiquitination of cyclin B1 and securin, respectively (; ; ). Loss of cyclin B1 causes a reduction in MPF, and the loss of securin frees separase to act on the kleisin component of cohesin (; ; ; ). Many proteins have been associated with the establishment and/or maintenance of CSF activity. Factors responsible for setting up a second meiotic spindle after completion of meiosis I do not, a priori, have to be the same as those that are responsible for maintaining arrest. Indeed, proteins have been described that are involved in establishing MetII, but not in maintaining arrest once it has been achieved (). The mechanism of CSF is most well characterized in the frog, where various groups have firmly defined the c-Mos–MAPK–90-kD ribosomal protein S6 kinase (p90rsk)–budding uninhibited by benzimidazole 1 (Bub1) pathway in establishing CSF activity (; ; ; , ; ). Other activities that are fundamentally involved in the establishment of CSF in frog include cyclin E/Cdk2 () and mitotic arrest deficient 2 (Mad2; ). However, once established, p90rsk, Mad2, Bub1, and cyclin E/Cdk2 are all dispensable for the maintenance of CSF activity (; , ). So, how is CSF activity maintained in the frog? Current evidence suggests it is through early mitotic inhibitor 2 (Emi2)/Emi-related protein 1 (; ; ; ; ). Emi2, which acts to inhibit the APC accumulated during oocyte maturation, is present and stable in CSF frog egg extracts, but is rapidly degraded on Ca addition (). Degradation of Emi2 is induced by phosphorylation through CamKII (; ; ) and, thus, would be predicted to occur ahead of APC activation and cyclin B1 degradation, although this has not been tested. In mouse oocytes, the mechanism of CSF arrest is less well understood. As the mouse Emi2 homologue appears to have a similar function in maintaining CSF activity (), it would be logical to predict that the mechanism of CSF establishment is also conserved between frog and mouse. However, this is not so. Oocytes from a triple Rsk knockout mouse arrest normally at MetII (), demonstrating that p90Rsk is not involved in mouse CSF arrest. Furthermore, using dominant-negative mutants confirmed that the spindle checkpoint proteins Bub1 and Mad2 are not required for either the establishment or the maintenance of mouse CSF. Although c-Mos is known to be involved in CSF maintenance, it does not appear to be involved in the establishment of CSF arrest because c-Mos −/− oocytes remain at MetII for 2–4 h before extruding a second PB (PB2; ). As there are no firm candidates suggested to be involved in the establishment of CSF arrest in mouse, we have investigated the potential of mouse Emi2 in this process. We find that Emi2 has activity consistent with CSF, and its degradation in real time is ahead of any change in cyclin B1. We also demonstrate that a function of Emi2 is in restabilizing cyclin B1 upon exit from anaphase I and that, in this study, it contributes to the formation a bipolar spindle. Oocytes matured without Emi2 do not assemble MetII spindles, and, in the absence of cyclin B1, eventually decondense their chromatin. To measure real-time changes in Emi2 levels in oocytes we generated cRNA to mouse Emi2 coupled to Venus fluorescent protein (Emi2-V), which is a yellow variant of GFP. MetII mouse oocytes were microinjected with this construct at a dose of either 0.15 or 0.5 pg, and then cultured for a few hours to allow for Emi2 expression. We Western blotted oocytes expressing Emi2-V with a polyclonal antibody against Emi2 to determine the amount of Emi2-V expression relative to endogenous protein. More than one band was detected on oocyte blots using this antibody; however, one band at ∼85 kD migrated at the same molecular mass as in vitro–translated Emi2 (and this band was later knocked down by Emi2 morpholino [MO]). At the 0.15-pg dose, Emi2-V levels were less than endogenous protein after 2 h, whereas the 0.5-pg dose was expressed to about the same level as endogenous protein (). Emi2-V was very stable in MetII-arrested oocytes, but became rapidly unstable when cytosolic Ca increased. With either 0.15 pg () or 0.5 pg (not depicted) injections, we observed no degradation of Emi2-V in MetII oocytes after blocking further synthesis by washing into cycloheximide-containing media. However, when this experiment was repeated, but by washing into Sr-containing media to induce spermlike Ca spiking (), we observed dose-dependent effects on Emi2-V. At the higher 0.5-pg dose, we observed no loss in Emi2-V signal; instead, Emi2-V levels steadily increased (). However, at the lower 0.15-pg dose Emi2-V was rapidly degraded (). The high dose of Emi2 maintained oocytes in a MetII arrest. Oocytes injected with 0.5 pg Emi2-V cRNA showed no morphological signs of meiotic resumption, which is consistent with the maintained Emi2-V levels in these oocytes (). In contrast, oocytes that had been injected with 0.15 pg Emi2-V cRNA extruded a PB2 (). This suggests that Emi2-V has physiological CSF activity and that the oocyte has a finite capacity to degrade Emi2. At the lower dose Emi2 injection, where we observed Emi2-V degradation, the minimum in its degradation profile after activation with Sr media was reached tens of minutes before PB2 extrusion (, T). However, we have previously reported that cyclin B1 degradation, visualized by coupling to GFP, is only completed at the time of PB2 formation (). This suggests that the Emi2 and cyclin B1 degradation profiles may not fully overlap. When cyclin B1 was expressed in mouse oocytes with the same fluorescent protein tag as Emi2 (cyclin B1-Venus; cyclin B1-V), we observed the same degradation profile as that previously found for cyclin B1-GFP, such that a minimum in the cyclin B1-V profile was reached within minutes of PB2 extrusion (, T). Comparing the degradation profiles of Emi2-V and cyclin B1-V suggests that Emi2 degradation is initiated ahead of cyclin B1. Sr-induced Emi2 degradation begins immediately (), whereas that of cyclin B1 begins 20 min later (). This would have to occur if MetII arrest is being mediated by Emi2-induced inhibition of APC activity. Therefore, we wanted to confirm the immediate loss of endogenous Emi2 signal in oocytes, which is especially important given that had reported very little loss in Emi2 signal at a 6-h time point after activation with Sr. Oocytes were activated by washing into Sr media, and samples from a pool of oocytes were removed at various time points and probed for either cyclin B1 or Emi2 by Western blotting (). In these experiments, it was apparent that loss in Emi2 protein was rapid and complete by 30 min (in agreement with the rapid loss of Emi-V shown in ). Interestingly, Emi2 levels increased again in pronucleate embryos (6-h time point), which is in agreement with and suggests that it may have a further mitotic function (see Discussion). Similar to Emi2, we observed the loss of cyclin B1 at 30 min, as well as increased levels in pronucleate embryos (), which is consistent with our observation that the APC is switched off at this time (). Because of the numbers of oocytes needed for Western blots and the tens of minutes of asynchrony in timing at which Ca spiking starts with Sr media, (), we could not reproducibly resolve Emi2 degradation ahead of cyclin B1 by Western blotting groups of oocytes. Therefore, to examine with more accuracy the immediate degradation profiles of cyclin B1 and Emi2, we decided to measure their simultaneous degradation in the same oocyte. Cyclin B1 was coupled to Cerulean fluorescent protein (cyclin B1-C), which is a cyan variant of GFP. There was no overlap in Venus and Cerulean signals, showing that both Emi2-V and cyclin B1-C, with appropriate filters, could be imaged simultaneously in the same oocyte (). In these experiments, it was evident that the introduction of Emi2-V delayed cyclin B1 degradation ( and ), which is consistent with Emi2-inhibiting APC activity. However, these coexpression studies revealed that Emi2 degradation began tens of minutes before that of cyclin B1 (). Emi2-V levels were degraded by at least 50% before the start of cyclin B1-C degradation ( = 7). Degradation of cyclin B1 is dependent on APC activity, and in mouse oocytes it can be blocked by the induction of a spindle checkpoint (; ). In contrast, Emi2 degradation should be checkpoint-independent because its degradation is independent of APC involvement. Incubation of Emi2- and cyclin B1-expressing oocytes with the spindle poison nocodazole blocks mouse oocytes from exiting MetII arrest when washed into Sr media. As expected, the addition of nocodazole completely stabilized cyclin B1 levels ( = 15/15; ). However, nocodazole had no effect on the rate of Emi2-V degradation ( = 12/12; ). Such an observation is consistent with cyclin B1, but not Emi2 degradation, being dependent on the APC. Therefore, in summary, we have obtained data that are entirely consistent with a model of MetII arrest achieved by Emi2-mediated inhibition of cyclin B1 degradation. Also, the Emi2-V construct generated is a physiologically active, useful tool for both establishing CSF activity and measuring its loss in real time after a Ca signal. Emi2 levels are low in both and mouse oocytes before they are matured. This would be predicted, as high Emi2 levels during maturation may be deleterious and arrest oocytes at MetI. The increased Emi2 expression during oocyte maturation makes it highly likely that Emi2 expression can be knocked out by an antisense approach. Therefore, we designed an antisense MO to the 5′UTR immediately adjacent to the start codon of mouse Emi2 (Emi2 MO) and used an additional two MO's as controls (); a 5-base mispair MO (5mp-MO), in which five bases have been altered from the complementary sequence, and an inverted MO (Inv-MO). To explore the role of Emi2 in the establishment of MetII arrest, we injected Emi2 MO into germinal vesicle (GV) oocytes, which were then matured in vitro. Oocytes were held at the GV stage in milrinone-containing media for 2 h after MO injection; they were then released from GV arrest and allowed to mature for 16 h. Blotting of GV-stage oocytes, Emi2 MO–matured oocytes, and uninjected control matured oocytes demonstrated that Emi2 protein levels increase between GV and MetII stage, and confirmed the Emi2 knockdown in Emi2 MO–injected oocytes (). During maturation, oocytes were scored for the morphological events of oocyte maturation, which are GV breakdown (GVBD) and PB1 extrusion (). Both GVBD () and PB1 extrusion () occurred with normal timings in Emi2 MO–injected oocytes. However, after maturation, we observed marked differences in the morphology of control oocytes and those injected with Emi2 MO. When control oocytes (uninjected, 5mp-MO, and Inv-MO–injected) were stained for chromatin, oocytes were morphologically normal. They had a PB1 containing chromatin, which was produced on completion of the first meiotic division, and a fully formed MetII spindle (100%; uninjected oocytes, = 60; 5mp-MO, = 32; Inv-MO, = 25; ). However, although oocytes injected with Emi2 MO did have a PB1 containing chromatin, the chromatin in the oocyte was decondensed inside a nucleus (93%; = 110; ). Despite the lack of effect of a 5mp-MO, it remained possible that we had been extremely unlucky in the MO design, such that the observed effects of Emi2 MO were caused by its ability to block the expression of an unrelated protein involved in MetII arrest. We thought this unlikely, given that a similar morphology of Emi2 knockdown oocytes has been reported recently using a double-stranded RNAi approach (). However, we decided it was important to confirm the specificity by a rescue to the control phenotype in Emi2-MO–injected oocytes by expression of exogenous Emi2. To recover Emi2, Emi2-V cRNA was microinjected into oocytes 2 h after microinjection of Emi2 MO and immediately before release from GV arrest. This rescue is made possible because Emi2-V lacks the 5′UTR recognized by the MO. Injection of Emi2-V cRNA alone into GV oocytes that were matured induced a MetI arrest ( = 40; ), which is consistent with Emi2-V having CSF activity and the need to keep Emi2 levels low until completion of the first meiotic division. Importantly, Emi2-V cRNA expression could rescue the effects of MO knockdown. Rescue oocytes could progress through meiosis I and arrest as controls with a fully formed MetII spindle. A high dose of Emi2-V (0.5 pg; = 45) rescued the Emi2 MO phenotype, with an equal mix of either MetI or MetII arrest, whereas oocytes microinjected with the lower Emi2 dose showed a rescue with a predominantly MetII arrest ( = 45; ). By scoring oocytes for decondensed chromatin at only one time point (16 h), we were unable to pinpoint at which stage of meiosis oocytes undergo chromatin decondensation. However, we noted that the vast majority of oocytes with decondensed chromatin extruded only a single PB and contained only one nucleus (88%; = 102). This observation suggests either decondensation of chromatin occurred after anaphase I, such that oocytes did not form a MetII spindle (, i), or, alternatively, that after establishment of a MetII spindle there was no sister chromatid disjunction before decondensation (, ii). We determined whether Emi2 MO knockdown oocytes were able to build a MetII bipolar spindle by staining oocytes for both chromatin and tubulin at various times after PB1 extrusion. Oocytes were fixed at 0.5, 1, and 2 h after PB1 extrusion. In control uninjected oocytes, central spindle microtubules were observed at 0.5 h after PB extrusion (), and over the next 1.5 h a MetII spindle formed, such that by 2 h after PB1 a fully formed MetII spindle was found in all oocytes (). In Emi2-MO–injected oocytes, at 0.5 h after PB1 there was no difference from controls, with central spindle microtubules evident between the chromatin in the ooplasm and the PB1 (). However, in contrast to control oocytes, in Emi2-MO–injected oocytes at both the 1 and the 2 h time points we observed no MetII spindle; instead, the chromatin had remained essentially unaltered from the time of PB1 extrusion. The residual central spindle microtubules remained between the chromatin in the oocyte and the PB, and by 2 h the chromatin appeared to be beginning to decondense (). As the chromatin started to decondense, spindle microtubule structure was lost, which is consistent with entry into interphase. By 6 h, all Emi2-MO–injected oocytes had a single nucleus containing fully decondensed chromatin. We did not assess if these oocytes underwent S-phase; however, they showed no obvious signs of degeneration over the 6-h time period from PB1 extrusion. Thus, we failed to observe the formation of a bipolar spindle in any Emi2-MO–treated oocytes; instead, oocytes underwent full chromatin decondensation. Because the formation of a MetII spindle requires an increase in the levels of CDK1 activity after PB1 extrusion, we reasoned that in the absence of Emi2, levels of CDK1's regulatory partner cyclin B1 may not be reestablished. To test this directly, Emi2 MO or control GV oocytes were microinjected with cyclin B1-V cRNA and matured in vitro. Cyclin B1-V levels were monitored in real time during maturation, along with brightfield microscopy to assess progression through meiosis I. As previously reported (), in control oocytes there was a period of cyclin B1 degradation that lasted a few hours and was terminated on PB1 extrusion (). Immediately after PB1 extrusion in these oocytes, cyclin B1 levels rose to 24.0 ± 1.1% ( = 14) of the peak level before PB1 extrusion (). In Emi2-MO–injected oocytes, there was no difference from controls in the timing of initiation or in the rate of cyclin B1-V degradation during meiosis I, which is in agreement with the observed lack of effect of the MO on the timing of PB1 (). However, in these Emi2 MO oocytes there was no reelevation of cyclin B1 after PB1 extrusion; cyclin B1-V levels increased to just 1.6 ± 0.3% ( = 15) of the peak level before PB1 extrusion (). In keeping with the real-time effects on exogenous cyclin B1-V expression, we also observed a loss of endogenous cyclin B1 after Emi2 knockdown. In vitro–matured oocytes injected with Emi2 MO at the GV stage showed much lower cyclin B1 levels than control in vitro–matured oocytes (). Furthermore, cyclin B1 levels in individual oocytes were measured by immunofluorescence at 16 h after release from GV arrest in control and Emi2-MO–injected oocytes (). We found that cyclin B1 immunofluorescence levels in Emi2 MO–treated oocytes was about one third that of control oocytes (), suggesting a relatively uniform level of knockdown. Because the formation of a MetII spindle requires an elevation in the levels of cyclin B1, we reasoned that the Emi2 MO spindle defect may be recovered by the addition of cRNA to nondegradable cyclin B1. Furthermore, it should also be rescued by inhibiting the APC by addition of Mad2. Cyclin B1 with deletion of 90 N-terminal amino acids (Δ90 cyclin B1) lacks a D-box and is therefore not a substrate of the APC (; ). Emi2-MO–injected oocytes either received no further treatment or were microinjected with 1.5 pg Δ90 cyclin B1 or Mad2 cRNA within 15 min of PB1 extrusion. This injection had to be done immediately after PB1 extrusion because nondegradable cyclin B1 addition to oocytes before PB1, as predicted by the sustained MPF activity, blocks PB extrusion (). A few hours of Δ90 cyclin B1 cRNA expression generates cyclin B1 at a similar level to endogenous cyclin B1 in a MetII oocyte (). This dose of Mad2 causes a metaphase I arrest in maturing oocytes () and prevents completion of meiosis in MetII oocytes (). At 2 h after PB1 extrusion, all oocytes were fixed, stained, and scored for the presence of a bipolar spindle (). As observed in , oocytes microinjected with Emi2 MO gave decondensing chromatin with a lack of any bipolar spindle structure. However, as with uninjected oocytes, we observed a fully formed bipolar MetII spindle in oocytes microinjected with Emi2 MO and either Δ90 cyclin B1 cRNA or Mad2 (). We have demonstrated in live oocytes and in real time that upon a Ca signal, Emi2 degradation occurs ahead of cyclin B1, independent of APC inhibition. The current study confirmed that Emi2 constitutes CSF activity in mouse MetII oocytes and is responsible for the maintenance of MetII arrest (). More importantly, we find that Emi2 is necessary and sufficient in stabilizing cyclin B1 levels at the completion of meiosis I, and such stabilization is essential in establishing a bipolar spindle. By definition, CSF activity must accumulate during oocyte maturation, inhibit APC activity at MetII, and be inactivated on a Ca signal at fertilization. Although only recently identified, the F-box protein Emi2 is very likely to constitute CSF in frog eggs (; ; ; ; ). Emi2 inhibits the ubiquitin ligase activity of the APC in vitro, and in keeping with the nature of CSF, it accumulates during oocyte maturation, is present in CSF extracts, and is rapidly degraded by Ca (). l c h e m i c a l s w e r e o b t a i n e d f r o m S i g m a - A l d r i c h , u n l e s s o t h e r w i s e s t a t e d , a n d w e r e o f t i s s u e c u l t u r e o r e m b r y o - t e s t e d g r a d e w h e r e a p p r o p r i a t e .

In many cells, agonists trigger biphasic Ca signals in which release of Ca from the ER is followed by a sustained influx of Ca through store-operated channels (SOCs) in the plasma membrane. SOCs are activated by the depletion of Ca from the ER, and their activity is essential for a diversity of functions in electrically nonexcitable cells, including secretion, motility, and gene expression. In T cells, antigen binding to the T cell receptor (TCR) elevates inositol 1,4,5-trisphosphate (IP), depleting intracellular Ca stores and thereby activating a highly Ca-selective SOC called the Ca release–activated Ca (CRAC) channel (). CRAC channels play a critical role in promoting gene expression during T cell activation, as shown by spontaneous mutations that abrogate Ca influx and cause severe combined immunodeficiency in human patients (; , ). Despite persistent research efforts over the past 15 yr, there is still a lack of consensus as to the molecular mechanism underlying store-operated Ca entry (for reviews see ; ). A major stumbling block in this search has been the lack of identified proteins that could be shown to be unequivocally required for the process. In a major advance, two groups recently used RNAi-based screens to identify stromal interacting molecule 1 (STIM1) as an essential, conserved regulator of store-operated Ca entry (; ). Current evidence suggests that STIM1 is the Ca sensor for store depletion. STIM1 has an EF-hand domain predicted to face the ER lumen, and mutations of residues that are expected to reduce STIM1's affinity for Ca constitutively activate store-operated Ca entry in cells with full Ca stores (; ). Interestingly, store depletion causes wild-type STIM1 to redistribute from a diffuse ER localization into puncta located at the cell periphery. This redistribution is thought to be an important step in SOC activation for two reasons. First, STIM1 molecules bearing EF-hand mutations are constitutively localized in these puncta, even when stores are full, consistent with their ability to constitutively activate Ca entry (; ). Second, redistribution brings STIM1 toward the plasma membrane, suggesting a means by which STIM1 could transmit a store depletion signal to CRAC channels. Several issues must be addressed to determine whether and how the redistribution of STIM1 is involved in opening CRAC channels. First, does STIM1 accumulate in puncta rapidly enough to be causally linked to CRAC channel activation? The kinetics of STIM1 redistribution has been reported in HeLa cells and Jurkat cells using total internal reflection fluorescence (TIRF) and confocal microscopy (; ). Although puncta formation is generally slow, like CRAC current (I) activation, quantitative comparisons have not been made between redistribution and I activation in single cells, which is essential for testing causality. Another critical area of uncertainty is the precise location of STIM1 puncta within the cell. concluded that puncta formed in the ER near the plasma membrane. They based their conclusions on the fact that YFP-labeled STIM1 was within the evanescent TIRF field of 100–200 nm but could not be detected by anti-GFP antibodies applied to the cell exterior. On the other hand, concluded that store depletion causes STIM1 to enter the plasma membrane, based on immuno-EM and surface biotinylation studies. The precise location of STIM1 after store depletion is central to understanding how store depletion is linked to SOC activation. Over the past decade, several classes of models have been proposed to describe how this communication might occur (; ). These include conformational coupling between proteins in the ER and plasma membranes, possibly between STIM1 and the CRAC channel itself. In an extension of this model, referred to as “induced-coupling” or “secretion-like coupling,” the depleted ER would move to the cell periphery to establish these physical coupling sites (). Store depletion may also promote vesicle fusion and insertion of CRAC channels or STIM1 into the plasma membrane. Finally, depletion could activate CRAC channels at a distance through the release of a diffusible messenger from the ER. Despite extensive studies by many groups, no single model has achieved widespread acceptance. Resolving the precise location of STIM1 in the cell and, in particular, its relationship to the ER and the plasma membrane is needed to more fully define STIM1's role in the store-operated Ca entry process. Unfortunately, neither light microscopy nor immuno-EM provide sufficient spatial resolution to determine whether STIM1 puncta form in the plasma membrane or in ER directly beneath it. We therefore approached this question in two ways. First, we examined the quenching of a GFP-labeled STIM1 using TIRF microscopy as a real-time assay for plasma membrane–localized STIM1. Second, we used HRP-labeled STIM1 to visualize STIM1 at the EM level. HRP-labeled proteins have been used to study subcellular localization with greater spatial resolution and better preservation of membrane structure than is possible with immuno-EM (; ). We report that STIM1 puncta formation slightly precedes CRAC channel opening in single cells, strongly supporting the idea that STIM1 redistribution is a critical early step in channel activation. Store depletion causes STIM1 to redistribute within the ER to form puncta in tubules near the plasma membrane without any detectable insertion of STIM1 into the plasma membrane on the time scale of CRAC channel activation. At the EM level, STIM1 puncta are found in both preexisting and newly docked junctional ER located 10–25 nm from the plasma membrane. The distance between STIM1 puncta and the plasma membrane is short enough to allow for the possibility of direct interactions between STIM1 and proteins at the cell surface. In Luik et al. (this issue, p. ), we demonstrate that CRAC channels and Ca entry occur exclusively at these junctional sites containing STIM1, thereby establishing a structural and functional basis for the elementary unit of store-operated Ca entry. The redistribution of STIM1 from a diffuse ER localization into puncta at the cell periphery is widely assumed to be an important step in opening CRAC channels. Although STIM1 redistribution measured in cell populations has been shown to occur on average with roughly the same time course as CRAC channel activation, a direct comparison of STIM1 puncta formation and I activation has not been done (; ). If the formation of STIM1 puncta is necessary for CRAC channel activation, it must precede channel opening. Because the kinetics of CRAC channel activation and STIM1 redistribution vary among cells (and will vary independently if they are not coupled), the relative timing of these two processes can only be unequivocally determined in single-cell measurements. We therefore applied TIRF microscopy and whole-cell patch-clamp recording to monitor STIM1 redistribution and I activation simultaneously in individual Jurkat T cells transiently expressing a GFP-STIM1 chimera. GFP-STIM1 was made by inserting GFP into the lumenal domain of STIM1, between the N-terminal signal sequence and the EF-hand. GFP-STIM1 behaved similarly to wild-type STIM1 and to the fluorescent protein–labeled STIM1 constructs used by others (; ) in that GFP-STIM1 was localized to the ER in resting cells ( and Figs. S1 A and S2 A, available at ) but redistributed into bright puncta located at the cell periphery upon store depletion (, Fig. S1 A, Fig. S2 B, and Video 1). 1 mM EGTA was introduced into the GFP-STIM1–expressing cell through the recording pipette to deplete Ca stores and activate I. The time course of I development was monitored during brief voltage steps to −122 mV delivered every 3 s. During each voltage step, a TIRF image was acquired to monitor GFP-STIM1 accumulation at sites within ∼200 nm of the plasma membrane. As reported previously in intact cells (; ), passive depletion of Ca stores during whole-cell recording led to the appearance of GFP-STIM1 puncta near the plasma membrane ( and Video 2, available at ). Inward current was activated in parallel and was identified as I on the basis of its characteristic slow kinetics, dependence on extracellular Ca, and inwardly rectifying current–voltage relation (). The mean density of I was increased by 78 ± 30% (transfected, = 6; control, = 14) in cells expressing low to moderate levels of GFP-STIM1 (or Cherry-STIM1; ), suggesting that the abundance of endogenous STIM1 may normally limit the extent of CRAC channel activation. GFP-STIM1 redistribution and I activation occurred over a period of ∼60 s after break-in to the whole-cell configuration (). A plot of each parameter on a normalized scale illustrates that GFP-STIM1 accumulated near the plasma membrane in this cell before CRAC channels began to open (). The relative timing of STIM1 accumulation and I induction was assessed from the times for each parameter to reach 20% of its maximal value. In five cells examined in this way, GFP-STIM1 fluorescence began to increase 10 ± 2.5 s before I increased, and in every cell, GFP-STIM1 redistribution preceded the initial increase in I (). Similar results were obtained using perforated-patch recording to avoid washout of intracellular constituents and a more physiological mode of stimulation in which an anti-CD3 mAb was used to cross-link the TCR and deplete stores via IP-induced Ca release ( and Fig. S1). Under these conditions as well, STIM1 redistribution always preceded CRAC channel activation (mean latency of 6.2 ± 1.8 s; = 4 cells). These results support the idea that STIM1 redistribution is required for CRAC channel activation. Determining the location of STIM1 after store depletion is critical to understanding how it activates CRAC channels. A recent study reported that store depletion causes STIM1 to translocate from the ER into the plasma membrane, where it might directly interact with or form CRAC channels (). To test whether STIM1 is inserted into the plasma membrane, we performed fluorescence-quenching experiments with GFP-STIM1. GFP fluorescence is pH sensitive, decreasing by >80% from pH 7.0 to 5.0 (). If store depletion triggers GFP-STIM1 insertion into the plasma membrane, then the GFP will be exposed extracellularly and therefore be susceptible to quenching at low pH. We used TIRF imaging to compare the acid-induced quenching of GFP-STIM1 to that of cytosolic GFP and a glycosylphosphatidylinositol (GPI)-anchored GFP targeted to the outer leaflet of the plasma membrane. and Video 3 (available at ) show TIRF images of Jurkat cells expressing GPI-GFP, GFP-STIM1, or cytosolic GFP. Ca stores were depleted by incubating the cells in 0-Ca Ringer's solution containing thapsigargin (TG), an ER Ca pump inhibitor, for 10–15 min before the start of the experiment. Store depletion caused GFP-STIM1 to accumulate in bright puncta at the cell periphery, whereas the distributions of GPI-GFP and cytosolic GFP were unaffected ( and Video 3). Acidifying the extracellular solution from pH 7.3 to 5.2 immediately quenched the fluorescence of the extracellular-facing GPI-GFP but had no effect on the fluorescence of GFP-STIM1 or cytosolic GFP (). GFP-STIM1 and cytosolic GFP fluorescence were quenched by low extracellular pH only after exposure to the ionophores nigericin and monensin, which equilibrate intracellular and extracellular pH (). Because significant quenching of GFP-STIM1 only occurred in the presence of ionophores (), we conclude that virtually all of the GFP-STIM1 is intracellular. Similar GFP-quenching experiments using wide-field epifluorescence microscopy confirmed the TIRF results (Fig. S2 and Video 4). To address the possibility that the large (239 amino acid) GFP tag may have interfered with trafficking to the plasma membrane, we performed similar experiments using STIM1 labeled with a much smaller (41 amino acid) N-terminal HA tag. After store depletion, HA immunofluorescence was detected in permeabilized cells but not in intact cells, demonstrating that HA-STIM1 is not inserted into the plasma membrane to any significant degree (Fig. S3). Therefore, we conclude that the STIM1 puncta that form after store depletion are intracellular and located near but not in the plasma membrane. Given that the STIM1 puncta are intracellular, where do they form? To address this, we compared the localization of GFP-STIM1 to that of an ER marker (DsRed2-ER) in resting and store-depleted cells using TIRF microscopy. In Jurkat cells with full Ca stores, GFP-STIM1 colocalized with the ER marker (), consistent with previous work (; ). After store depletion, GFP-STIM1 accumulated in puncta close to the plasma membrane (), and these puncta also overlapped with the ER marker (). These results suggest that rather than exiting the ER upon store depletion, GFP-STIM1 moves from being diffusely distributed within the ER to accumulate in puncta in areas of the ER near the plasma membrane. The formation of STIM1 puncta in areas of the ER close to the plasma membrane led us to consider the “induced coupling” model of store-operated Ca entry previously proposed (; ). This model states that the ER moves to the plasma membrane to establish protein–protein interactions necessary for SOC activation. We used TIRF microscopy to image cells coexpressing GFP-STIM1 and DsRed2-ER to test this hypothesis. Because the intensity of the evanescent field decays exponentially with distance from the coverslip, any oriented movement of the ER toward the plasma membrane would be marked by increased DsRed fluorescence. After store depletion with TG, there was no detectable increase in DsRed2-ER fluorescence over the time period when GFP-STIM1 fluorescence increased dramatically (). The quantitation of results from 14 cells () suggests that STIM1 accumulation near the plasma membrane does not involve bulk movements of the ER toward the plasma membrane. Thus, at the resolution of the light microscope, store depletion appears primarily to induce a redistribution of STIM1 within the ER. The TIRF imaging experiments suggest that store depletion causes GFP-STIM1 to redistribute from a diffuse localization in the ER to accumulate locally in the ER close to the plasma membrane, whereas the overall structure of the ER itself does not change. However, the resolution of light microscopy imposes certain limits on conclusions about localization. Even with the narrow optical sectioning afforded by TIRF microscopy, small rearrangements in ER structure may be difficult to detect against a large volume of fluorescent ER. Immuno-EM offers much better spatial resolution but often low contrast and poor preservation of membrane structure if cells are permeabilized to provide access to antibodies. Alternatively, the localization of HRP fusion proteins in the ER, Golgi, and transport vesicles has been achieved with very high contrast and without permeabilization, using peroxidase cytochemistry and conventional transmission EM (; ). Because the HRP reaction product is restricted to the lumen of the HRP-containing compartment, ER-localized STIM1 should be clearly distinguishable from STIM1 in the plasma membrane. Therefore, we examined the subcellular distribution of STIM1 and the ER at the ultrastructural level using HRP-labeled STIM1 (HRP-STIM1) and an ER-targeted HRP (HRP-ER). In a previous study, an ER-targeted HRP in transfected fibroblast and epithelial cells appeared in the nuclear envelope and in serpentine cytoplasmic tubules, consistent with localization in the ER (). In transfected Jurkat cells with full Ca stores, both HRP-ER () and HRP-STIM1 () displayed a similar distribution, producing electron-dense HRP reaction product in the nuclear envelope as well as in tubules throughout the cytosol. These results confirm observations with light microscopy indicating that GFP-STIM1 is localized in the ER in cells with full stores (; ; ) and confirm that the N terminus of STIM1 faces the ER lumen. Store depletion with TG did not grossly alter the ER morphology visualized with HRP-ER (, compare A and B with C and D), in agreement with our conclusions from TIRF measurements (). In contrast, store depletion dramatically changed the distribution of HRP-STIM1. After treatment with TG, HRP-STIM1 shifted from being located in tubules throughout the cytoplasm () to being concentrated in tubule segments that were closely associated with the plasma membrane (, arrows). HRP-STIM1 was clearly located in tubules rather than vesicles, and these tubules were similar in width and location to plasma membrane–associated ER tubules visualized with HRP-ER (, arrows). Consistent with a redistribution toward plasma membrane–associated ER, the residual HRP reaction product in cytosolic ER tubules after store depletion (, arrowheads) was greatly reduced compared with resting cells (, arrowheads). Importantly, in both resting and store-depleted cells, we failed to detect any extracellular HRP reaction product indicative of HRP-STIM1 in the plasma membrane, providing further evidence that STIM1 is not inserted into the plasma membrane in response to store depletion (, Fig. S2, and Fig. S3). Together, these results show that the peripheral STIM1 puncta seen by light microscopy in store-depleted cells correspond to the accumulation of STIM1 in plasma membrane–associated ER at the ultrastructural level. Interestingly, we consistently observed in resting cells a small fraction of both HRP-ER and -STIM1 tubules in close apposition to the plasma membrane (, A and B; and , arrows). In terms of width and proximity to the plasma membrane, these ER–plasma membrane contact sites resemble the “junctional ER” described in many cell types (; ; ; ). The presence of junctional ER in resting Jurkat cells raises the question of whether STIM1 puncta form at these preexisting sites of ER–plasma membrane contact. To begin to address this question, we asked whether store depletion affects the number or length of junctional ER structures, which we identified as HRP-ER tubules located within 50 nm of the plasma membrane (). Store depletion increased the total length of junctional ER by 55% (, blue vs. black), and this effect resulted from a 62% increase in the number of junctional ER contacts () without a significant change in the lengths of individual contacts (). In cells overexpressing HRP-STIM1, depletion also increased the total length of junctional ER (, red vs. green); however, in these cells, the effect was attributed to a combined increase in both the number () and the length of individual contacts (). Together, these results indicate that store depletion promotes the formation of new contact sites between the ER and the plasma membrane and causes STIM1 to accumulate at these sites as well as in preexisting regions of junctional ER. The minimum distance between HRP-containing tubules and the plasma membrane was quantified in cells expressing HRP-ER and -STIM1 before and after store depletion. For every 200-nm length of tubule located within 50 nm of the plasma membrane, we measured the minimum distance from the proximal side of the tubule to the plasma membrane. Because there was no significant difference in these values among the different cells and conditions, the measurements were pooled, giving a mean minimum distance between junctional ER tubules and the plasma membrane of 17 ± 10 nm (mean ± SD; = 293 measurements from 61 micrographs). sub xref #text Jurkat E6-1 cells (American Type Culture Collection) were grown as previously described (). Cells were transiently transfected with a gene pulser (Bio-Rad Laboratories) 48–72 h before experiments. For each transfection, 10 cells were centrifuged, resuspended in 300 μl of Jurkat medium, and mixed with 10 μg of DNA in a 0.4-cm gap cuvette. Cells were incubated with DNA for 10 min at room temperature, electroporated at 0.24 kV and 960 μF, incubated again for 10 min, and resuspended in 10 ml of fresh medium. All salts and chemicals were obtained from Sigma-Aldrich unless otherwise stated. Control Ringer's solution contained 155 mM NaCl, 4.5 mM KCl, 2 mM CaCl, 1 mM MgCl, 10 mM -glucose, and 5 mM Hepes, pH 7.3 with NaOH. 0-Ca Ringer's solution contained 155 mM NaCl, 4.5 mM KCl, 3 mM MgCl, 10 mM -glucose, 5 mM Hepes, and 1 mM EGTA, pH 7.3. For pH 5.2 0-Ca Ringer's solution, CaCl was replaced with 1 mM EGTA, and Hepes was replaced with MES. Nigericin/monensin solution contained 70 mM NaCl and 70 mM KCl (to equilibrate extracellular and intracellular H), 1.5 mM KHPO, 2 mM CaCl, 1 mM MgSO, 10 mM glucose, 10 mM Hepes, 10 mM MES, 0.01 mM nigericin, and 0.01 mM monensin, pH 5.2. TG was obtained from LC Laboratories; for TCR cross-linking experiments, 10 μg/ml OKT3, a murine mAb against human CD3 (eBioscience), was added to the extracellular solution. Patch-clamp experiments were conducted in the standard whole-cell recording and perforated-patch configurations as previously described (; ). The internal solution for whole-cell recording contained 140 mM Cs aspartate, 3 mM MgCl, 1 mM EGTA, and 10 mM Hepes, pH 7.2 with CsOH. For perforated-patch experiments, the internal solution contained 115 mM Cs aspartate, 1 mM CaCl, 5 mM MgCl, 10 mM NaCl, 10 mM Hepes, pH 7.2 with CsOH, and 100 μg/ml amphotericin B. For all experiments, currents were filtered at 1 kHz and sampled at 2 kHz without series resistance compensation. Unless indicated, all data were leak subtracted using currents collected in nominally Ca-free bath solution plus 10 μM LaCl. Transiently transfected Jurkat cells were plated onto poly--lysine–coated coverslip chambers and imaged with epifluorescence or through-the-objective TIRF microscopy using a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with a Fluar 100×, 1.45 NA oil-immersion objective (Carl Zeiss MicroImaging, Inc.). Laser light (488 or 514.5 nm) from a 100-mW argon ion laser (Enterprise 622; Coherent) was coupled to the microscope via optical fiber (Point Source) and focused at the back focal plane of the objective using a TILL-TIRF condenser (Till Photonics) that employs a translatable 45° prism to control the incident angle of the laser beam. For epifluorescence microscopy, cells were excited with 488-nm (GFP) or 540-nm (rhodamine) light using a monochromator (Polychrome II; Till Photonics). GFP and DsRed/rhodamine images were acquired by switching between GFP (z488/10 excitation, Q505LP dichroic, and 535/30 emission) and DsRed/rhodamine (HQ535/45 excitation, Q570LP dichroic, and D660/50 emission) filter cubes (Chroma Technology Corp.). For simultaneous patch-clamp and TIRF imaging experiments, a 1.0 neutral density filter was added in front of the GFP excitation filter and the 535/30 emission filter was replaced with a 510LP. Images were acquired with a cooled CCD camera (ORCA-ER; Hamamatsu). Filter switching and image acquisition were controlled by scripts written in MetaMorph (Molecular Devices). All fluorescence images were acquired with 2 × 2 pixel binning except for those in (1 × 1). Video 2 was acquired at 3-s frame intervals, and other videos were acquired at 20-s frame intervals. All experiments were performed at room temperature (22–25°C). After transient transfection with HRP-STIM1 or -ER plasmids, Jurkat cells were plated onto poly--lysine–coated coverslips and fixed with 2% glutaraldehyde (Electron Microscopy Sciences) in 0.1 M Na cacodylate buffer (Electron Microscopy Sciences) in a microwave oven (EM3450; Pelco) equipped with a cold spot set at 15°C. Fixed cells were amplified with a TSA-biotin system (PerkinElmer) and ABC kit (Vector Laboratories) for 30 min each before being prereacted with 1 mg/ml DAB (Sigma-Aldrich) in Tris-buffered saline for 10 min and then reacted with DAB with 0.01% HO for 30 min. After postfixation with 1% OsO and en bloc stain with 1% uranyl acetate, cells were further processed as previously described () before embedding in Embed 812 (Electron Microscopy Sciences). Glass coverslips were removed, cells were located by light microscopy and punched out, and 50–90-nm sections were cut and mounted on formvar-coated grids and viewed with an 80-kV transmission electron microscope (model 1230; JEOL) equipped with a slow-scan cooled CCD camera (model 967; Gatan). HA-STIM1 and untransfected Jurkat cells were plated onto poly--lysine–coated coverslips for 45 min at 37°C, treated with either control Ringer's or 0-Ca Ringer's plus 1 μM TG for 15 min at room temperature, fixed with 4% paraformaldehyde (Electron Microscopy Sciences) in 10 mM PBS (Sigma-Aldrich) for 30 min at room temperature, and washed with PBS plus 20 mM glycine. Selected coverslips were permeabilized with 0.1% Triton X-100 for 5 min before blocking for 1.5 h in PBS plus 2% BSA, 20 mM glycine, and 75 mM NHCl. Cells were incubated with each antibody diluted in blocking buffer (1 μg/ml 12ca5 mouse anti-HA [a gift from A. Venteicher, Stanford University, Stanford, CA] and 30 μg/ml rhodamine red-X goat anti–mouse IgG [Jackson ImmunoResearch Laboratories]) for 1 h each at room temperature and washed thoroughly with blocking buffer in between incubations. Coverslips were mounted onto glass slides with Vectashield (Vector Laboratories). Image analysis was performed using NIH ImageJ software; figures were prepared using Photoshop and Illustrator (Adobe). All raw fluorescence images were dark-noise subtracted before analysis. Images in and Fig. S1 A were thresholded to three to four times the remaining background. To quantitate fluorescence intensities of GFP- and DsRed-labeled proteins, images were thresholded, and only data from pixels above the threshold were quantified. The threshold was set at the minimum pixel intensity + 10% of the range of pixel intensities for that image (threshold = min + 10%[max – min]). Because the footprints of Jurkat cells occasionally increased during an experiment, we limited intensity measurements to an outline drawn around the cell footprint in the first frame of a time-lapse experiment. For each GFP-quenching video, the mean fluorescence intensity of a cell in each frame of the video was normalized to the mean fluorescence intensity of that cell in the frame just before the switch from pH 7.3 to 5.2 (, t = 100 s; Fig. S2, t = 540 s). In , the mean fluorescence intensity of a cell in each frame of a video was normalized to the mean fluorescence intensity of that cell in the first frame of the movie. Any adjustments made to brightness and contrast levels were made across the entire fluorescence or EM image or the entire video. To measure lengths of HRP-tubules in close contact with the plasma membrane (≤50 nm; ), we selected only micrographs where the nucleus and entire cell circumference were visible (taken at 8,000–25,000× magnification), to restrict our analysis to sections cut through the middle rather than the edges of cells. To estimate the mean minimum distance between HRP-tubules and the cell membrane, we measured the distances between the proximal edge of HRP-tubules and the cell membrane using micrographs taken at 25,000–150,000×. We measured the minimum distance from the plasma membrane for every 200 nm of HRP-containing tubule located within 50 nm of the plasma membrane. Means are expressed ±SEM unless otherwise stated; statistical significance was measured by the test or analysis of variance (ANOVA). Fig. S1 shows that STIM1 accumulation precedes I activation during IP-induced Ca release. Fig. S2 shows fluorescence-quenching measurements using wide-field epifluorescence microscopy to assay for plasma membrane–localized GFP-STIM1. Fig. S3 shows immunofluorescence staining of HA-tagged STIM1 in Jurkat cells before and after store depletion. Video 1 shows GFP-STIM1 redistribution in an intact cell after store depletion with TG. Video 2 shows STIM1 accumulation near the plasma membrane during I activation. Video 3 shows assaying for plasma membrane–localized GFP-STIM1 puncta by fluorescence quenching and TIRF microscopy. Video 4 shows assaying for plasma membrane–localized GFP-STIM1 puncta by fluorescence quenching and wide-field epifluorescence microscopy. Online supplemental material is available at .

A wide variety of cell-surface receptors produce Ca signals resulting from the release of Ca from intracellular stores followed by Ca entry through channels in the plasma membrane. This Ca entry process, termed capacitative or store-operated Ca entry (SOCE), is evoked by the depletion of Ca from the lumen of the ER and drives several critical cellular functions, including growth, motility, secretion, and gene expression (). The best-studied store-operated channel (SOC) is the highly Ca-selective Ca release–activated Ca (CRAC) channel found in lymphocytes and other hematopoietic cells, which is absolutely required for T cell activation by antigen (; ). The structural and mechanistic basis of SOCE has remained a mystery since its first proposal, by . The central issue is how the depletion of Ca from the lumen of the ER controls the activation of SOCs in the plasma membrane. From the outset, it was assumed that the ER was physically in close apposition to the plasma membrane in order for it to communicate with channels in the plasma membrane. ) and could occur without a detectable rise in intracellular free Ca concentration ([Ca]). These results implied an intimate relationship between ER and plasma membrane, in effect creating a private pathway for store refilling. Although the initial concept of direct coupling between the ER and plasma membrane was later abandoned, indirect evidence continued to support the idea that the ER is close to Ca entry sites in the plasma membrane (). Consistent with this view, in oocytes and astrocytes, fluorescence signals from store-operated Ca influx were highest in regions having the highest density of ER (; ). There has been much speculation that the connection between the ER and plasma membrane may resemble the dyad and triad junctions underlying excitation–contraction coupling in muscle. In these cells, the sarcoplasmic reticulum (SR) is positioned within 10–20 nm of the T-tubule (plasma membrane) to enable Ca channels in the plasma membrane to trigger the release of Ca from nearby ryanodine receptors in the SR; in this way, the dyad/triad forms the elementary structural unit of Ca release that triggers muscle contraction (). By analogy, the ER and closely apposed plasma membrane could be considered the elementary unit of SOCE, but the predicted structures, consisting of ER closely coupled to sites of Ca influx, have never been identified. The recent identification of stromal interacting molecule 1 (STIM1) as the probable Ca sensor for SOCE offers new strategies in the search for the elementary unit of SOCE. STIM1, a type I ER membrane protein, has a predicted EF-hand domain facing the ER lumen, and mutation of conserved residues likely to be involved in Ca binding activates Ca influx in cells with full stores (; ). Store depletion causes STIM1 to redistribute from a diffuse localization throughout the ER into puncta near the plasma membrane. This redistribution is likely to be involved in CRAC channel activation, because the constitutively active EF-hand mutants of STIM1 show this distribution even when stores are full (; ). Moreover, puncta formation slightly precedes the opening of CRAC channels, consistent with a causal role in SOC activation (see Wu et al. on p. of this issue). Interestingly, ultrastructural analysis shows that puncta correspond to the accumulation of STIM1 in discrete regions of junctional ER lying within 10–25 nm of the plasma membrane (). The proximity of STIM1 puncta to the plasma membrane is close enough to permit local interactions with proteins in the plasma membrane, including Orai1, which has recently been identified as an essential part of the CRAC channel (; ; ). These new studies raise important questions about how STIM1 transmits the activation signal to CRAC channels in the plasma membrane. Do junctional ER structures containing STIM1 provide the proximal stimulus for the activation of CRAC channels by store depletion? Are CRAC channels activated only locally at sites of STIM1 accumulation or more generally throughout the surrounding plasma membrane? What constitutes the basic unit of SOCE? To answer these questions and better understand how Ca store content regulates CRAC channel activity, the spatial relationship between STIM1, Orai1, and open CRAC channels must be determined. In this study, we have combined total internal reflection fluorescence (TIRF) microscopy and patch-clamp recording to determine the location of active CRAC channels relative to that of STIM1 in the ER. We find that CRAC channel activation in store-depleted cells is tightly restricted to regions directly apposed to STIM1 puncta. These results show for the first time that Ca influx through CRAC channels is not widely dispersed throughout the cell but, rather, is highly concentrated in areas juxtaposed to junctional ER that comprise only a small fraction of the cell surface. Store depletion also causes Orai1 to accumulate at these sites, providing a physical basis for the local activation of CRAC channels by STIM1. These results identify the elementary unit of SOCE and show that it is a dynamic assembly, arising from the coordinated migration of the Ca sensor and its target channel to closely apposed sites in the ER and the plasma membrane. The location and density of open CRAC channels was determined by measuring local steady-state [Ca] gradients at the plasma membrane after depleting Ca stores. For this purpose, we adapted a method that had been used previously to localize Ca entry through ryanodine receptors in cardiac myocytes () and voltage-gated Ca channels in excitable cells (). Jurkat T cells were dialyzed through the whole-cell recording pipette with 200 μM fluo-5F, a low-affinity, fast-binding Ca indicator, together with an excess (10 mM) of the higher affinity, slower binding Ca chelator, EGTA. EGTA binds Ca too slowly to significantly diminish Ca binding to fluo-5F close to channels, but the higher affinity and concentration of EGTA enables it to outcompete fluo-5F for Ca binding at distances >1 μm from the source (), thus restricting the fluo-5F signals to the vicinity of open channels. Further restriction of the signals in the z dimension is afforded by TIRF microscopy, which restricts excitation to distances within ∼200 nm of the membrane. Our protocol for localizing Ca influx sites is described in . Intracellular Ca stores were depleted by intracellular dialysis with EGTA to ensure maximal CRAC channel activation. Voltage clamp was used to control the driving force for Ca entry and, thus, the flux through open channels. At +38 mV, the driving force is minimal, and control images at this potential show dim baseline fluorescence (, F and F). Hyperpolarization to −122 mV elicits a rapid fluorescence increase as Ca enters the cell, and depolarization back to +38 mV rapidly terminates the signal. The baseline fluorescence of cells was generally nonuniform, presumably because of spatial variations in dye concentration and illumination. Therefore, to isolate the Ca-dependent signal, the fluorescence change in each image (ΔF = F − F, where F is the mean of control images F and F) was divided by F. Typically, the Ca-dependent fluorescence signal, ΔF/F, develops fully within 10–60 ms of hyperpolarization and dissipates equally quickly upon depolarization (), leaving behind a small, slowly decaying signal that reflects a small increase in global [Ca] (see the following section). Given that Ca influx through CRAC channels has never been visualized before at a microscopic scale, it is important to confirm that these signals do in fact reflect CRAC channel activity. that are typical of I (). Further evidence that the signals are due to CRAC channels comes from pharmacological experiments. The currents and the fluorescence signals evoked by hyperpolarization were both inhibited by La and 2-aminoethyldiphenyl borate (2-APB), known blockers of CRAC channels (; ; ). Finally, fluorescence changes were minimal in Jurkat CJ-1 cells (), a mutant line defective in CRAC channel activity (). In all cases, the ΔF/F signal was strongly correlated with the level of I measured electrophysiologically, confirming that the fluo-5F fluorescence signals result from Ca entry through CRAC channels. Several observations provide evidence that fluo-5F monitors predominantly local Ca signals near CRAC channels. First, the rapid rise in ΔF/F after membrane hyperpolarization is consistent with the rapid generation of a steady-state [Ca] elevation near open channels, whereas the rapid dissipation of the signal upon depolarization is expected from the kinetics of diffusion out of this microdomain (). The rapid dissipation of the signal upon depolarization was incomplete; a small residual fraction of the signal (typically <10% of the peak) decayed back to baseline over tens of seconds. This time course is similar to the rate of Ca clearance from Jurkat T cells by the plasma membrane Ca-ATPase operating near resting [Ca] (), suggesting that the residual signal is due to a small rise in global cell [Ca]. Consistent with this interpretation, the residual signal was relatively uniform across the cell, whereas the rapidly changing (local) signal was heterogenous ( and ). A second, more stringent test of the ΔF/F signal as a local detector of open CRAC channels is to compare the time course of fluorescence and I as CRAC channels undergo Ca-dependent inactivation. Because of the speed of Ca diffusion into the cell, local [Ca] close to channels is expected to vary in proportion to the current amplitude, whereas the global [Ca] should more closely follow the time integral of the current. In response to hyperpolarization, CRAC channels undergo rapid inactivation because of a local effect of incoming Ca (; ). The size of the current and the speed and extent of inactivation increase as [Ca] is increased from 2 to 20 mM (). Simultaneous fluorescence measurements also indicate a rapid decline in ΔF/F during the 150-ms hyperpolarization, which is enhanced by increasing [Ca] (). To compare the time course and amplitude of the current with those of ΔF/F, we first corrected ΔF/F for the buildup of global [Ca], assuming for simplicity that the global ΔF/F component increases linearly during the hyperpolarization. . The ΔF/F signal closely tracks both the time-dependent decline in I because of inactivation as well as the reduced amplitude of I at lower [Ca]. These results provide strong confirmation that the rapidly developing component of ΔF/F is a valid indicator of local [Ca] near open CRAC channels. The ability to resolve Ca influx sites is limited by the diffusional spread of Ca bound fluo-5F molecules. This can be measured empirically from the spatial ΔF/F profile produced by a point source of Ca influx. Serendipitously, we found that exposure of cells to unattenuated 488-nm laser illumination for >40 s evoked a light-activated increase in a nonselective leak conductance, which was associated with “hotspots” of fluo-5F fluorescence (). During maintained high-intensity illumination, these hotspots fluctuated, blinking on and off at various locations. These events appeared to be unrelated to CRAC channels, as they were insensitive to 100 μM 2-APB and 10 μM La and arose with similar frequency in control and CRAC-deficient CJ-1 Jurkat cells (unpublished data). The dimensions of the hotspots were relatively uniform, suggesting that they represent influx sites smaller than the overall spatial resolution of our Ca-detection system and therefore approximate a point source. The mean intensity profile of the hotspots followed a Gaussian function that decayed by 50% within 377 nm and by 90% within 644 nm (). This is significantly broader than the resolution of the microscope optics (the Airy disc decays by 50% within 86 nm), indicating that the measured ΔF/F profile is mostly determined by diffusion of Ca and fluo-5F. The spatial spread of the ΔF/F signal agrees reasonably well with that predicted by a simple reaction-diffusion model for a point source of Ca flowing into 200 μM fluo-5F and 10 mM EGTA (50% decay within 490 nm; see Materials and methods). Together with the results of and , these data demonstrate that the ΔF/F signal can be used to map the location of active CRAC channels in the cell footprint with submicrometer resolution. The spatial relationship between STIM1 puncta and active CRAC channels was determined by simultaneously visualizing STIM1 and Ca influx sites in individual cells. For separation of fluorescently labeled STIM1 and fluo-5F fluorescence, we labeled STIM1 with a monomeric red fluorescent protein (Cherry-STIM1; ). Jurkat cells transfected with Cherry-STIM1 were loaded with fluo-5F and imaged by TIRF microscopy; during the period between break-in and the start of recording, Cherry-STIM1 accumulated in puncta (). Upon hyperpolarization to −122 mV, transfected cells exhibited robust Ca influx through CRAC channels (); to minimize contributions from global [Ca], fluorescence data were collected only at short times (10–60 ms) after hyperpolarization. The distributions of both Ca influx density and Cherry-STIM1 puncta were nonuniform, with the highest densities of each near the center of the cell footprint (). To facilitate comparisons of the two maps, contour lines of Ca influx density in pseudocolor are overlaid on the image of Cherry-STIM1 fluorescence in . The spatial distributions of Ca influx and Cherry-STIM1 overlap but are not identical. Peaks of influx are not always centered on Cherry-STIM1 puncta, and some Cherry-STIM1 puncta are not associated with peaks of Ca influx (). Similar results were obtained in eight other cells. Unfortunately, the close spacing of the puncta complicates interpretation of these images; if clusters of active channels are separated by less than the spatial resolution of the [Ca] measurement (∼400 nm), superposition of their Gaussian fluorescence profiles will generate a maximum signal between the clusters. Because most of the Cherry-STIM1 puncta are separated by <400 nm, individual Ca influx sites cannot be unequivocally assigned to specific puncta. To clarify more precisely the relationship between STIM1 and active CRAC channels, we sought conditions that would increase the separation of STIM1 puncta. We tried several approaches that were unsuccessful; for example, partial depletion of stores with cyclopiazonic acid reduced the intensity of puncta but not their number. Depolymerization of microtubules with nocodazole caused the ER to coalesce into a clump near the center of the footprint and caused cells to detach from the coverslip (unpublished data). However, the actin depolymerizing agent cytochalasin D, added together with or after thapsigargin (TG), dramatically increased the space between STIM1 puncta. 5 μM cytochalasin D caused Cherry-STIM1 puncta to coalesce into larger structures within several min ( and Video 1, available at ) without affecting footprint adherence as monitored by interference reflection microscopy (IRM; ). A similar fusion of ER tubules into larger structures was noted in cells expressing ER-targeted fluorescent proteins after exposure to either 5 μM cytochalasin D or 20 μM latrunculin A (unpublished data). Several independent lines of evidence indicate that cytochalasin D treatment after store depletion preserves the normal CRAC channel activation mechanism in Jurkat cells. First, cytochalasin D did not disrupt the close association of STIM1 puncta with the plasma membrane, as determined at the ultrastructural level using Jurkat cells expressing the HRP-STIM1 fusion protein. Cytochalasin D treatment after store depletion caused HRP-STIM1 to accumulate in ER tubules in close proximity to the plasma membrane (), similar to what was seen in the absence of the drug (). Second, cytochalasin D did not affect the total near-membrane Cherry-STIM1 fluorescence (), suggesting that the drug reorganizes existing STIM1 puncta without altering the amount of STIM1 near the plasma membrane. Finally, I density was also unaffected by cytochalasin D (), showing at a functional level that the drug and the ensuing reorganization of STIM1 puncta leaves the CRAC channel activation mechanism intact. By reducing the number of puncta, cytochalasin D pretreatment allowed better resolution of the spatial relationship between Cherry-STIM1 and Ca influx sites (). In seven cells treated in this way, Ca influx sites were all tightly associated with STIM1 puncta, although, as in control cells (), not all puncta were associated with influx (; and Fig. S1, available at ). The spatial profiles of the Ca signal and STIM1 fluorescence were compared for 23 individual influx sites from seven cells, with one example shown in . Typically, the width of the ΔF/F signal was greater than expected from a single point source (), suggesting that each influx site represents a cluster of open CRAC channels. The spatial profile of the Ca signal was similar to but consistently wider than that of the associated STIM1 punctum. To estimate how much of this difference might be attributable to diffusion of Ca and fluo-5F from CRAC channels, we convolved the STIM1 distribution with a Gaussian function describing the spread of the fluo-5F signal. The resulting curve overlaps quite closely with the observed ΔF/F distribution, with the widths of the two curves differing by <200 nm (). The similarity of the fluo-5F and convolved STIM1 fluorescence profiles indicates that CRAC channels only open in extremely close proximity to STIM1. In principle, the local activation of CRAC channels could arise if STIM1 produces a local signal, i.e., one that only activates channels in the immediate vicinity; alternatively, the channels themselves may be confined to the vicinity of STIM1 puncta. Orai1 has recently been identified in several studies as an essential component of the CRAC channel (; ; ). To visualize the distribution of CRAC channels relative to STIM1, we transfected cells with Cherry-STIM1 and GFP-myc-Orai1. At rest, there is little overlap between Cherry-STIM1 and GFP-myc-Orai1. TIRF imaging shows that Cherry-STIM1 is localized to the ER, whereas GFP-myc-Orai1 is distributed across the cell footprint (). Wide-field epifluorescence images taken through the cell center show that Orai1 in resting cells is diffusely distributed at the cell perimeter and in membrane ruffles, suggesting a location in the plasma membrane (), in marked contrast to the ER distribution of STIM1 (Fig. S2 A, available at ). Remarkably, after store depletion, both Cherry-STIM1 and GFP-myc-Orai1 redistribute into a punctate pattern with a high degree of colocalization (, right; and Fig. S2 B). Subsequent treatment with cytochalasin D caused puncta to coalesce, and STIM1 and Orai1 remained colocalized (), further supporting our conclusion that cytochalasin D reorganizes the CRAC activation machinery without disrupting it. Most important, the accumulation of Orai1 near STIM1 puncta suggests a physical mechanism for the local activation of CRAC channels and Ca influx at these sites. Together, these results reveal the structure and dynamic nature of the elementary unit of SOCE. xref #text Jurkat E6-1 cells (American Type Culture Collection) and the CJ-1 mutant Jurkat cell line () were maintained as described previously (). 2-APB and cytochalasin D were obtained from Sigma-Aldrich, TG was purchased from LC Laboratories, and fluo-5F was obtained from Invitrogen. Cherry-STIM1 plasmid was constructed by inserting mCherry (a gift from R. Tsien, University of California, San Diego, La Jolla, CA) after the signal sequence of human STIM1 (Origene). mCherry was amplified by PCR from mCherry-pRSET-B () with primers 5′-ttattaggtaccatggtgagcaagggc-3′ and 5′-ataataggtacccttgtacagctcgtccat-3′ to append KpnI sites onto each end. After KpnI digestion, mCherry was ligated into a unique engineered KpnI site in STIM1 (Quickchange XL kit; Stratagene) at 1188–1193 bp. The HRP-STIM1 plasmid was constructed as described by . N-terminally myc-tagged WT Orai1 (a gift from S. Feske, Harvard Medical School, Boston, MA) in the Gateway entry vector pENTR11 () was inserted into the Gateway destination vector pDS_GFP-XB by recombination reaction using enzyme mix (Gateway LR Clonase; Invitrogen) to generate GFP-myc-Orai1. For transient transfection, 10 cells were electroporated (Gene Pulser; Bio-Rad Laboratories) with 10 μg plasmid DNA 48–72 h before imaging. Cells were allowed to adhere to poly--lysine–coated coverslip chambers and were bathed in extracellular solution containing 155 mM NaCl, 4.5 mM KCl, 2 or 20 mM CaCl, 1 mM MgCl, 10 mM -glucose, and 5 mM Na-Hepes, pH 7.4. 0-Ca Ringer's solution was nominally Ca free except in cytochalasin D experiments, when 2 mM MgCl plus 1 mM EGTA was added. Illumination at 488 and 514.5 nm was supplied by a 100-mW argon-ion laser (Enterprise 622; Coherent, Inc.) coupled via a single-mode fiber (Point Source) and a TILL-TIRF condenser (Till Photonics) to a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) equipped with a 100× α-Plan Fluar objective (NA 1.45; Carl Zeiss MicroImaging, Inc.). For 488-nm excitation (fluo-5F and GFP), the filter cube contained a 1.0 neutral density in series with a Z488/10 excitation filter, a Q505LP dichroic mirror, and a HQ535/50M emission filter (all filters were obtained from Chroma Technology Corp.). A 510 LP emission filter was used when imaging fluo-5F in untransfected cells. For 514.5-nm excitation (Cherry-STIM1), the filter cube contained a 0.5 neutral density in series with a HQ535/45 excitation filter, a Q570LP dichroic mirror, and an E610LPV2 emission filter. Neutral densities were omitted when imaging GFP-myc-Orai1 and Cherry-STIM1. Images were acquired with a cooled charge-coupled device camera (ORCA-ER; Hamamatsu). Cherry-STIM1 images were captured with 250–500-ms camera exposures at 2 × 2 camera pixel binning. In GFP-myc-Orai1 and Cherry-STIM1 colocalization experiments, images were captured at 1 × 1 pixel binning 3–4 s apart to allow time for filter cube switching. All imaging functions were controlled by MetaMorph software (Molecular Devices). All experiments were performed at room temperature (22–25°C). Cells were imaged with epi-illumination at 460 nm using a Polychrome II monochromator (Till Photonics); a 455DCLP allowed collection of 460-nm light reflected at both the glass–solution and solution–cell interfaces. If the distance between these interfaces is <230 nm, an interference pattern is generated where destructive interference is maximal at the smallest separation between the cell and the coverslip (). Electron microscopy was performed as described by . Patch-clamp experiments were conducted in the standard whole-cell recording configuration as previously described (). The standard internal solution contained 140 mM Cs aspartate, 3 mM MgCl, 10 mM EGTA, 10 mM Hepes, and 0.2 mM fluo-5F, pH 7.3 with CsOH. In Cherry-STIM1 experiments, fluo-5F concentration was reduced to 0.1 mM to minimize contamination of the Cherry fluorescence signal. Currents were filtered at 1 kHz and sampled at 2 kHz without series resistance compensation. Voltage stimuli consisted of a 100-ms step to −122 mV followed by a 100-ms ramp from −122 to +50mV. Unless indicated, all data were leak subtracted using currents collected in nominally Ca-free bath solution plus 10 μM LaCl. Three to four sweeps were averaged for each displayed current. Simultaneous imaging and patch-clamp measurements were performed using in-house software developed on the Igor Pro platform interfaced with an input/output board (ADAC 1300; Molecular Devices) that simultaneously controlled the charge-coupled device camera, the laser shutter, and the amplifier (Axopatch 200B; Molecular Devices). The voltage was alternated between +38 and −122 mV every 150 ms to generate five hyperpolarizing steps. 50-ms camera exposures were collected 10 and 85 ms after each voltage step. Camera pixels were binned to 8 × 8 when imaging fluo-5F in untransfected cells or to 4 × 4 in cells transfected with Cherry-STIM1. Image analysis was performed using NIH ImageJ software. All raw fluorescence images were dark-noise subtracted and thresholded to three to four times the remaining background. For fluo-5F experiments, the F image was generated by averaging two images collected at +38 mV. F was subtracted from each image, and the result was divided by F to yield ΔF/F. Unless indicated, each displayed ΔF/F image is a mean derived from four voltage stimulus presentations. The predicted radial spread of Ca–fluo-5F from a point source was calculated for 0.2 mM fluo-5F and 10 mM EGTA (). As Ca enters the cell, the fraction of Ca bound to fluo-5F (R) depends on the binding rate constants for fluo-5F (k = 2.36 × 10 Ms) and EGTA (k = 2.5 × 10 Ms): Initially, upon Ca entry, 65% of Ca is bound to fluo-5F. The spread of Ca–fluo-5F is determined by the unbinding rate constant for fluo-5F (k = 543 s) and its diffusion coefficient (D = 40 μm/s). On average, fluo-5F will diffuse 379 nm before unbinding Ca, and 65% of this released Ca will again be bound by fluo-5F. This reaction–diffusion cycle results in a 50% decay of the Ca–fluo-5F signal within 490 nm. The radial spread of Ca–fluo-5F was directly measured from Ca influx hotspots in CJ-1 cells evoked by constant unattenuated 488-nm laser light. 50-ms images were acquired every 150 ms at 2 × 2 pixel binning; two control images were collected at +18 mV, and nine consecutive images were collected at −112 mV. This stimulus was repeated at 10-s intervals until hotspots of light-activated Ca influx were detected. Fig. S1 shows that active CRAC channels colocalize with STIM1 after cytochalasin D treatment. Fig. S2 shows that STIM1 and Orai1 colocalize after store depletion. Video 1 shows that cytochalasin D reorganizes STIM1 puncta in Jurkat cells. Online supplemental material is available at .

The formation of long-term memory requires both new RNA and protein synthesis, whereas short-term memory requires only covalent modifications of constitutively expressed preexisting proteins (; ; ; ; ; ). On the cellular level, transcriptional regulation is thought to be the starting point for a series of biological amplifications, which are steps necessary for the induction and maintenance of long-term facilitation (LTF). Thus, during consolidation of LTF, a cascade of gene activation is tightly regulated by various transcription factors, including positive and negative regulators (). The core molecular features of the transcriptional regulation involved in long-term memory seem to be evolutionarily conserved in , , and the mouse (). A growing body of evidence indicates that gene regulation by different combinations of transcriptional factors may be involved in specific forms of long-term memory (; ). Moreover, the activation and inactivation of transcription factors by various kinases can regulate the transactivational potency of preexisting transcription factors and the recruitment of other factors. For example, the activation of CREB by PKA, MAPK, or CaMKIV is particularly important for its contribution to transcriptional activation and long-term memory in and in the mammalian hippocampus (; ; ; ). In the marine snail , the molecular mechanisms of long-term memory have been extensively studied in the sensory neuron–to–motor neuron synapses of the gill-withdrawal reflex. Five pulses of 5-hydroxytryptamine (5-HT) spaced at 15 min produce LTF that lasts >24 h and depends on transcription and translation. These pulses of 5-HT up-regulate the levels of cAMP within the sensory cell via G protein–coupled receptors and activate protein kinase A (PKA) and MAPK (; ; ; ; ; ; ). Both kinases then translocate into the nucleus, where they can activate transcription factors expressed in sensory neurons, such as ApCREB1 and ApCREB2. PKA can activate the transcriptional activator ApCREB1 (; ) and relieve the repression of ApCREB1-mediated transcription by the repressor ApCREB2 (; ). PKA-mediated activation of ApCREB1 can recruit CBP, and this leads to the activation of immediate-early genes, such as ApC/EBP, which are involved in the gene expression required for the consolidation and stabilization of LTF (; ). Although several transcription factors are important for the consolidation of LTF, phosphorylated ApCREB1 is sufficient (). This implies that the CRE-driven downstream genes activated by CREB should also be essential for the consolidation of LTF. Indeed, induction of ApC/EBP by activated ApCREB1 is necessary for LTF (; ). However, even though ApC/EBP is necessary, it does not seem to be sufficient because overexpression of ApC/EBP alone does not induce LTF (). This implies that other transcription factors activated by 5-HT must also be required for the consolidation of LTF. Therefore, suggested that the constitutively expressed transcription factor activating factor (ApAF) may be a potential transcriptional cofactor in LTF via its interaction with ApC/EBP or ApCREB2, but not with ApCREB1. However, the transcriptional regulation initiated by these interactions, and the identity of the core transcriptional component that is important for the consolidation of LTF, have not been characterized. We demonstrate that ApAF relieves the repression mediated by ApCREB2 by interacting with ApCREB2 and induces LTF by forming heterodimers with ApC/EBP. We further show that PKA-activated ApAF–ApC/EBP can induce LTF without CRE-driven gene expression. We further characterize the critical phosphorylation of ApAF by PKA at Ser-266. Thus, our study suggests that PKA-activated ApAF–ApC/EBP heterodimer is both necessary and sufficient in the consolidation of LTF. In the aforementioned study, found that ApAF could interact with ApC/EBP or ApCREB2 in vitro and might function as a transcriptional cofactor by interacting with other transcription factors. However, it was unclear whether ApAF can actually interact with ApC/EBP or ApCREB2 in neurons. To directly examine interactions between memory-related transcription factors at the cellular level, we used an two-hybrid system (). We measured β-galactosidase activity in neurons as an indication of the interactions among ApAF, ApCREB2, and ApC/EBP. We found the following types of interaction: (a) those between full-length ApAF and the basic leucine zipper (bZIP) domain of ApAF; (b) those between full-length ApAF and the bZIP domain of ApC/EBP; and (c) those between full-length ApAF and the bZIP domain of ApCREB2 (). The ApAF–ApC/EBP or ApAF–ApCREB2 heterodimers showed higher β-galactosidase activities than ApAF or ApC/EBP homodimers, indicating that ApAF may heterodimerize better with ApC/EBP or ApCREB2 than it homodimerizes with itself. These were consistent with both yeast two-hybrid and in vitro binding assay data () showing that ApAF formed stable heterodimers with both ApC/EBP and ApCREB2 proteins, although ApAF formed homodimers inefficiently. In contrast, neither the pair alone nor the Gal4DB-bait (bZIP of ApC/EBP or ApCREB2) on its own activated the reporter gene (). These results indicate that nuclear protein ApAF actually interacts with both a transcriptional activator ApC/EBP and a repressor ApCREB2 in the nucleus, implying that ApAF can function in the sensory cells as a transcriptional cofactor during LTF in . Because ApAF interacts both with ApC/EBP and with ApCREB2, it raises the question: what is the functional role of the interaction of ApAF with ApC/EBP in the consolidation of LTF? Because the consensus-binding sequences of ApAF from were previously reported to contain CAAT box sequences (), we investigated whether ApAF regulates reporter gene expression through enhancer response element (ERE) sequences, which contain the CAAT core box sequence, and whether 5-HT can activate the reporter gene expression in sensory neurons (; ). Accordingly, we microinjected pNEXδ-ApAF together with ERE-luciferase reporter into sensory neurons. 1 d after this microinjection, we found that reporter gene expression was increased by approximately fivefold by the ApAF construct, as compared with the ERE-luciferase reporter that was introduced as a control (). We next asked whether reporter gene expression induced by ApAF introduction is activated by 5-HT. To address this question, we microinjected pNEXδ-ApAF with the ERE-luciferase reporter into sensory neurons. 1 d later, we exposed sensory neurons to one pulse of 5-HT (10 μM), which by itself only produces short-term facilitation (STF) and does not induce gene expression (; ). We found that 1 d after this one pulse of 5-HT, reporter gene expression in neurons into which ApAF had been introduced was increased by approximately twofold compared with untreated cells (). ApC/EBP overexpression combined with one pulse of 5-HT also activated ERE reporter gene expression (). These results show that ApAF as well as ApC/EBP can activate ERE-driven gene expression by 5-HT in sensory neurons. We next investigated the consequences of the ApAF–ApC/EBP interaction on ERE-mediated gene expression. For this purpose, pNEXδ-ApAF and pNEXδ-ApC/EBP with ERE-luciferase reporter were microinjected into sensory neurons. 1 d after DNA microinjection, sensory neurons were treated with one pulse of 5-HT. We found that 24 h after 5-HT treatment, ERE-mediated reporter gene expression by ApAF–ApC/EBP overexpression was further increased by approximately twofold compared with that by either ApC/EBP or ApAF alone (). Furthermore, we also found that reporter gene expression by ApAF–ApC/EBP was increased by approximately fourfold in 5-HT–treated cells versus non–5-HT–treated cells (). These results indicate that ApAF can enhance the ERE-mediated gene expression by cooperating with ApC/EBP and that ApAF can be activated by 5-HT treatment. Because ApAF could activate reporter gene expression through ERE by interacting with ApC/EBP, we next investigated the actual role in long-term synaptic facilitation of ApAF heterodimerized to ApC/EBP. We previously reported that ApC/EBP is essential for the consolidation of LTF by showing that blocking ApC/EBP blocks the facilitation (), and that overexpression of ApC/EBP combined with only one pulse of 5-HT converted STF to LTF (). We also showed that LTF produced by ApC/EBP overexpression and one pulse of 5-HT was comparable to that produced by five pulses of 5-HT (; ). Therefore, we considered the possibility that the ApAF–ApC/EBP heterodimer may be essential for the consolidation of LTF because overexpressed ApC/EBP interacts with constitutively expressing endogenous ApAF. To test this idea by blocking the ApAF expression, we used the RNAi technique (). First, to confirm the effectiveness and the specificity of ApAF double-stranded (ds) RNA, we injected ApAF dsRNA into sensory neurons. 1 d later, we examined the expression level of ApAF mRNA by in situ hybridization. We injected the luciferase dsRNA as a control. The expression of ApAF mRNA was significantly knocked down in ApAF dsRNA–injected neurons compared with luciferase dsRNA–injected neurons (). In contrast, ApAF dsRNA did not affect the expression level of ApCREB1 mRNA compared with luciferase dsRNA (). These results show that ApAF dsRNA specifically impairs the expression of ApAF. Given these results, we examined the effect of the inhibition of ApAF synthesis on LTF induced by five pulses of 5-HT. The cells injected with dsApAF failed to produce LTF induced by five pulses of 5-HT without effecting STF and the basal synaptic strength (, C and D, and not depicted). These results are consistent with the earlier finding () that ApAF is required for LTF in sensory-to-motor synapses. Next, we coinjected pNEXδ-ApC/EBP with ApAF dsRNA into a sensory neuron. Cells injected with ApAF dsRNA also failed to produce the LTF induced by pairing one pulse of 5-HT with ApC/EBP overexpression, whereas cells injected with luciferase dsRNA showed normal long-term synaptic facilitation (). These results suggest that ApAF–ApC/EBP heterodimer is essentially required for the consolidation of LTF, and indicate that the ApC/EBP homodimer is insufficient to convert STF to LTF. If ApAF–ApC/EBP heterodimer is important for LTF, can overexpression of ApAF enhance the LTF induced by ApC/EBP overexpression when combined with one pulse of 5-HT? To address this question, we overexpressed ApAF, ApC/EBP, or ApAF + ApC/EBP in the sensory neurons of sensory-to-motor synapses, and then exposed cocultures to one pulse of 5-HT. We found that ApAF overexpression enhanced LTF induced by ApC/EBP overexpression combined with one pulse of 5-HT (), without affecting basal synaptic transmission (unpublished data). However, ApAF overexpression alone did not produce LTF in response to one pulse of 5-HT (), suggesting that ApC/EBP is also critical in LTF. If ApAF plays an important role in LTF by cooperating with ApC/EBP, how is the functional activation of ApAF regulated by 5-HT to induce LTF? Our reporter assays and electrophysiological recordings showed that a signaling pathway activated by 5-HT is required to activate ERE-mediated gene expression and to induce ApAF-mediated LTF. However, as is the case with ApCREB2, ApAF is a constitutive protein that is not induced by 5-HT, whereas ApC/EBP is rapidly induced (; , ; unpublished data). Therefore, we considered the possibility that ApAF might be regulated by kinases activated by 5-HT. Because 5-HT can activate the cAMP–PKA pathway and ApAF has two possible PKA phosphorylation sites at Ser-175 and -266 (Fig. S1, available at ), we explored whether a PKA-signaling pathway is involved in ApAF-mediated LTF by blocking the PKA pathway using the PKA inhibitor KT5720. Incubation with 10 μM KT5720 was found to completely block LTF consolidated by ApC/EBP overexpression and 5-HT. Moreover, the ApAF-mediated enhancement of LTF was completely impaired by KT5720 (). These results suggest that PKA signaling is essentially involved in ApAF–ApC/EBP–mediated LTF. Because 5-HT can also activate MAPK, we examined whether MAPK activity is required for the activation of the ApAF–ApC/EBP complex using PD98059, which is an inhibitor of the MAPK kinase MEK. PD98059 did not block LTF consolidated by ApAF and ApC/EBP overexpression (). Although PD98059 treatment slightly attenuated ApAF–ApC/EBP–mediated LTF, the degree of facilitation was not significantly different from that without drug treatment (P > 0.05; two-tailed Mann-Whitney test; ). This result suggests that MAPK signaling is not critically involved in ApAF–ApC/EBP–mediated LTF. To clarify whether ApAF is an endogenous substrate of PKA, we examined ApAF phosphorylation by using cell lysate as a source of the kinases. As shown in , KT5720 dramatically blocked the phosphorylation of ApAF, whereas either PD98059 or chelerythrine (PKC inhibitor) did not (). Moreover, to examine whether ApAF is directly phosphorylated by PKA at Ser-175 or -266, we incubated the proteins with the catalytic subunit of bovine PKA. We found that PKA phosphorylation was detected in wild-type (WT) ApAF and in ApAF (S175A), but not in ApAF (S266A) or ApAF (S175/266A; ). Thus, these results clearly suggest that PKA specifically phosphorylates ApAF at Ser-266. Next, we examined whether PKA is required for transcriptional activation by ApAF–ApC/EBP. To address this issue, ERE-mediated gene expression regulation by ApAF mutants (S175A, S266A, or S175/266A) was first tested by reporter gene assay. ApAF (WT, S175A, S266A, or S175/266A) and ApC/EBP constructs were injected together with the ERE-reporter construct into sensory neurons. The day after microinjection, sensory cells were treated with one pulse of 5-HT, and the following day luciferase assays were performed. Consistent with previous results (), reporter gene expression regulated by ApAF–ApC/EBP (S266A) or by ApC/EBP–ApAF (S175/266A) was found to be attenuated by ∼4.6- and ∼7-fold compared with that regulated by ApAF–ApC/EBP (WT; ). However, reporter gene expression regulated by ApAF–ApC/EBP (S175A) was not reduced compared with that regulated by ApAF–ApC/EBP (WT; ). In addition, we found that both WT ApAF and ApAF (S266A) form a heterodimer with ApC/EBP with similar affinity in GST pull-down assay and two-hybrid assay ( and Fig. S2, available at ). These data suggest that ApAF (S266A) is likely to act as a dominant-negative mutant through interfering with the transcriptional machinery, such as RNA polymerase II complex, that may recognize only a phosphorylated ApAF–ApC/EBP heterodimer. Collectively, our results suggest that ApAF phosphorylation by PKA at Ser-266 essentially regulates the ERE-mediated gene expression via cooperation between ApAF and ApC/EBP. Because Ser-266 phosphorylation is important for ERE-mediated gene expression (), we examined whether phosphorylation at Ser-266 is also required for LTF. We injected both pNEXδ-ApAF constructs (WT, S175A, S266A, or S175/266A) and pNEXδ-ApC/EBP into sensory neurons and treated them with one pulse of 5-HT the next day. ApC/EBP-mediated LTF was completely blocked in cells expressing pNEXδ-ApAF mutants (S266A or S175/266A), which cannot be phosphorylated, without affecting basal synaptic strength (). However, cells injected with pNEXδ-ApAF (WT) or pNEXδ-ApAF (S175A) showed enhanced ApC/EBP-mediated LTF (). To further confirm whether ApAF phosphorylation at Ser-266 is required for LTF produced by five pulses of 5-HT, we overexpressed ApAF (S266A) in sensory neurons. We found that ApAF (S266A), functioning as a dominant-negative mutant, completely blocked 5-HT–induced LTF (). Collectively, these results suggest that ApAF phosphorylation by PKA at Ser-266 is essential for the LTF induced by cooperation between ApAF and ApC/EBP. First, we examined whether ApAF could regulate CRE-driven gene expression by interacting with ApCREB2 using a transcriptional reporter gene assay. ApCREB1 binds to the CRE sequence and activates CRE-mediated gene expression, whereas ApCREB2 inhibits CRE-mediated gene expression by repressing ApCREB1 (, ; ). Therefore, we microinjected into neurons pNEXδ-ApCREB1, pNEXδ-ApCREB2, and pNEXδ-ApAF with CRE-luciferase reporter and found that ApCREB2 overexpression completely blocked the CRE-driven gene expression activated by ApCREB1 (). But remarkably, both ApAF mutants and ApAF WT relieved the repression by ApCREB2 of CRE-driven reporter gene expression (). In addition, we found that both ApAF WT and ApAF mutant (S266A) form a heterodimer with ApCREB2 with similar affinity in GST pull-down and two-hybrid assays ( and Fig. S2). Collectively, these results indicate that ApAF can activate ApCREB1 indirectly by direct interaction with ApCREB2, and that ApAF phosphorylation by PKA at Ser-266 is not required to relieve the repression of ApCREB2. If this is so, then is Ser-266 phosphorylation by PKA essential for the relief of LTF repressed by ApCREB2? To address this question, ApAF constructs (WT, S175A, S266A, or S175/266A) were coinjected with ApCREB2 into a sensory neuron in a sensory-to-motor coculture. One day later, the cultures were exposed to five pulses of 5-HT. We found that ApAF mutants, which cannot be phosphorylated by PKA, and the ApAF (WT) restored the LTF repressed by ApCREB2 () without affecting basal synaptic transmission (not depicted). The apparent loss of the dominant-negative effect of ApAF (S266A) on ApC/EBP is likely to be caused by a large quantity of the overexpressed ApCREB2. ApAF (S266A) and ApCREB2 can inhibit each other by direct interaction. Thus, overexpressed ApCREB2 may prevent ApAF (S266A) from interfering with ApC/EBP in a dominant-negative fashion. Control cells and cells injected with pNEXδ-ApAF produced normal 5-HT–induced LTF (). Collectively, these data suggest that ApAF relieves the repression of LTF by ApCREB2 through a direct binding with ApCREB2, and indicate that ApAF phosphorylation on Ser-266 is not necessary to relieve LTF repressed by ApCREB2. If Ser-266 phosphorylation in ApAF is essential for induction of LTF, is PKA-activated ApAF–ApC/EBP sufficient for LTF? To test whether PKA-activated ApAF induces LTF by interacting with ApC/EBP without CRE- and CREB-mediated gene expression, we blocked CRE- and CREB-mediated gene expression by microinjecting CRE oligonucleotide () and by overexpressing dominant-negative “killer” CREB (K-CREB ; ). The injection of CRE oligonucleotide blocked the LTF produced by five pulses of 5-HT (). However, CRE injection did not affect the LTF produced by ApAF–ApC/EBP heterodimer in the presence of one pulse of 5-HT (). Moreover, the injection of CRE mutant containing the randomly scrambled CRE sequences did not affect the LTF produced by 5 pulses of 5-HT or the LTF produced by ApAF–ApC/EBP heterodimer in the presence of one pulse of 5-HT (unpublished data). To further demonstrate whether PKA-activated ApAF–ApC/EBP heterodimer is sufficient for the consolidation of LTF, we next blocked CREB-mediated gene expression using dominant-negative K-CREB. The cells injected with pNEXδ-K-CREB failed to produce a normal 5-HT–induced LTF. However, the cells overexpressing K-CREB did not block the LTF produced by ApAF and ApC/EBP in the presence of one pulse of 5-HT. Collectively, these results suggest that the PKA-activated ApAF–ApC/EBP heterodimer is sufficient to induce LTF, independent of CREB activity, and imply that ApAF–ApC/EBP is a downstream effector of ApCREB. To examine whether the PKA-activated ApAF–ApC/EBP heterodimer produces LTF through ERE-mediated gene expression, we injected ERE oligonucleotides into the sensory neurons overexpressing ApAF and ApC/EBP. As shown in , ERE oligonucleotide injection impaired the LTF induced by ApAF–ApC/EBP overexpression combined with one pulse of 5-HT. ERE oligonucleotide injection also blocked the LTF induced by five pulses of 5-HT (). In contrast, cells injected with ERE mutant oligonucleotides affected neither the ApAF–ApC/EBP-mediated LTF nor the 5-HT induced LTF (unpublished data). Collectively, these data suggest that the ApAF–ApC/EBP heterodimer, once activated, induces LTF through ERE-mediated gene expression. t h i s s t u d y , w e d e m o n s t r a t e d a n o v e l , f u n c t i o n a l c o o p e r a t i o n b e t w e e n A p A F a n d A p C / E B P ( a n a c t i v a t o r ) o r A p C R E B 2 ( a r e p r e s s o r ) f o r L T F . M o r e o v e r , w e f o u n d t h a t t h e m e t h o d s o f A p A F i n t e r a c t i o n w i t h t h e s e t w o f a c t o r s w e r e q u i t e d i f f e r e n t . O u r s t u d y s u g g e s t s t h a t P K A - a c t i v a t e d A p A F – A p C / E B P i s a k e y t r a n s c r i p t i o n a l c o m p o n e n t t h a t i s n e c e s s a r y a n d s u f f i c i e n t f o r t h e c o n s o l i d a t i o n o f L T F . To clone ApAF from , we used the ∼850-bp N-terminal fragment, which was obtained using NdeI–BbsI–digested pNEX-ACT-ApAF () as a probe. We screened ∼1.5 × 10 clones from an cDNA library, and five positive signals were obtained after the second screening. Five clones were analyzed by DNA sequencing. One of them had an open reading frame of 1,083 bp encoding a polypeptide of 360 amino acids, and it also contained a bZIP domain (Fig. S1). Moreover, the amino acid sequence of ApAF from shares ∼80% similarities with that of . A search for potential phosphorylation sites revealed a common consensus sequence for the phosphorylation by PKA (Fig. S1), PKC, and CaMKII, indicating that ApAF may be regulated by these kinases. The ApAF obtained by PCR using specific primers (sense primer, 5′-CCCAAGCTTGCCACCACCATGATATCCAGCATTTCC-3′; antisense primer, 5′-CGGGATCTTTAGGCTGTACCTGCCAT-3′) was separately subcloned into HindIII–BamHI–digested pNEXδ and SalI–BamHI–digested pNEXδ-EGFP () to create pNEXδ-ApAF or pNEXδ-EGFP-ApAF, respectively. The mutant fragments of ApAF (S175A, S266A or S175/266A) were generated by recombinant PCR using specific sense or antisense primers (sense primer containing S175A, 5′-CAAGAAAGCTGCCAAATCACCTG-3′; antisense primer containing S175A, 5′-CAGGTGATTTGGCAGCTTTCTTGG; sense primer containing S266A, 5′-CAAGAGGATTGCTTCCACAGCTTCTG-3′; and antisense primer containing S266A, 5′-GCTGTGGAAGCAATCCTCTTG-3′). The PCR products containing mutations (S175A, S266A, or S175/266A) in ApAF amino acid residues were subcloned into HindIII–KpnI–digested pNEXδ vector. The full-length of ApAF was subcloned into BamHI–SacI–digested pNEX-ACT () to create pNEX-ACT-ApAF (Gal4-AD-ApAF). The bZIP domain of ApAF and the bZIP domain of ApCREB2 were subcloned into BamHI–SacI–digested pNEX-AS () to create pNEX-AS-ApAF (bZIP) (Gal4-DB-ApAF [bZIP]) and pNEX-AS-ApCREB2 (bZIP) (Gal4-DB-ApCREB2 [bZIP]), respectively. To generate pLitmis28i-luciferase, luciferase was subcloned into BamHI–KpnI–digested pLitmis28i vector (New England Biolabs, Inc.). To create pLitmus28i-ApAF, ApAF was subcloned into StuI–BamHI–digested pLitmus28i. The full length of ApC/EBP was inserted into pNEXδ-EGFP () to examine its cellular localization. The construction of pNEXδ-EGFP-CREB2 was previously described (). WT GST-ApAF, its mutant GST fusion protein (GST-ApAF S266A), and GST-ELAV1 (an RNA-binding protein with a protein size similar to ApAF) as a negative control were purified from using a general purification method (GE Healthcare). ApCREB2-HA and ApC/EBP-HA fusion proteins were translated in the TNT rabbit reticulocyte lysate (Promega). 5 μl of the lysates were mixed with 20 μl of saturated GST protein–bound beads in 200 μl of PBS containing 0.5% NP-40; these binding mixtures were incubated for 3 h at RT. The bound complexes were washed in 600 μl of TNET (25 mM Tris, pH 7.8, 150 mM NaCl, 0.2 mM EDTA, and 0.2% Triton X-100) and 300 μl of NET (25 mM Tris, pH 7.8, 150 mM NaCl, and 0.2 mM EDTA) serially, and then eluted by 2× LDS sample buffer (Invitrogen). The eluted samples were loaded in 12% SDS-PAGE gel and detected with anti-HA mouse monoclonal antibody (Sigma-Aldrich). In situ hybridization experiments were basically performed as previously described (). In brief, cultured sensory neurons were fixed with cold 4% paraformaldehyde in PBS and then permeabilized using 0.1% Triton X-100. After prehybridization with a hybridization solution (50% formamide, 5× SSC, 5× Denhardt's reagent, 0.25 g/ml yeast tRNA, and 0.5 g/ml salmon sperm DNA), hybridization was performed using 1 ng/μl of ApAF- or ApCREB1-specific, DIG-labeled antisense probe at 58°C for 12–18 h in a humidified chamber. After washing with 5× SSC at 58°C for 1 h, the sample was incubated with 10% heat-inactivated goat serum in PBS at RT for 1 h. After overnight incubation with the anti-DIG antibody (Roche), the sample was washed three times with PBS at RT for 30 min, followed by 10 mM Tris-Cl, pH 9.5, containing 0.5 mM MgCl for 5 min at RT. Development was performed using NBT/BCIP (Roche) for 24–36 h. The resulting cell images were acquired using a microscope (Diaphot; Nikon) attached to a digital camera system (Coolpix 995; Nikon) with a 20× objective at 18 ± 1°C. The hybridization signal in each cell was measured by outlining the cell body using the histogram function of Photoshop software (Adobe). The mean pixel intensity in the cell bodies was calculated by subtracting the background intensity from the cell body intensity. WT and mutant ApAF (S175A, S266A, and S175/266A) coding sequences were amplified by PCR with primers containing restriction enzyme sites (sense primer, CGATGATATCCAGCATTTCC; antisense primer, GCTTAGGCTGTACCTGCCAT; restriction enzyme sites are underlined). Amplified coding sequences were serially digested with BamHI and XbaI and ligated into pGEX-KG vector (Pharmacia, Inc.) for induction in . The mutation of each construct was confirmed by sequencing. Each construct was transformed into , and protein expression was induced by 0.2 mM IPTG for 3 h at 37°C. Protein induction was confirmed by SDS-PAGE and purified by a GST purification kit (M-Biotech) as recommended by the manufacturer. To examine whether ApAF is an endogenous substrate of PKA, the crude tissue extract preparation from pedal-pleural ganglia was performed as previously described (). The reaction was performed at 18°C for 20 min, containing 1 μg GST bead–binding GST-ApAF, 10 μg tissue extract, and 1 mCi γ [P]ATP in extraction buffer. To confirm the specificity of phosphorylation, the crude tissue extracts were previously incubated for 10 min with inhibitors of specific kinases, 40 μM KT5720 (PKA inhibitor; ), 20 μM PD98059 (MEK inhibitor), and 10 μM chelerythrine (PKC inhibitor). GST pull-down assay was performed as previously described (). The [P]phosphate incorporation was analyzed by SDS-PAGE and phosphoimager analysis. To examine whether ApAF is directly phosphorylated by PKA at Ser-175 or -266, PKA kinase assay was performed at 30°C for 30 min in a final volume of 25 μl, containing 1 μg substrate, 200 μM ATP, 1 mCi γ[P]ATP, and 5 U PKA catalytic subunit (New England Biolabs, Inc.). The reaction solution contained 50 mM Tris-Cl and 10 mM MgCl, pH 7.5. Reactions were stopped by adding SDS sample buffer and boiling at 100°C for 5 min. [P]phosphate incorporation was analyzed by SDS-PAGE and phosphoimager analysis. To confirm the specificity of phosphorylation by PKA, either 40 μM KT5720 (A.G. Scientific, Inc.) or DMSO was added. Microinjection into neurons was performed using air pressure, as previously described (). To investigate the interaction of ApAF with other transcription factors in neurons, such as ApC/EBP or ApCREB2, ∼20 giant cells of the abdominal ganglion were injected with 70 μg/ml of each activator DNA construct (Gal4AD-ApAF [full], Gal4AD-ApC/EBP [full], Gal4BD-ApAFbz, Gal4BD-ApC/EBPbz, and Gal4BD-ApCREB2bz), 760 μg/ml 4×gal4-lacZ, and 30 μg/ml pNEX2-luciferase, as previously described (). Luciferase and β-galactosidase assays were carried out as previously described (). To investigate the effect of ApAF on CRE-driven gene expression, pNEXδ-ApAF (WT or mutants), pNEXδ-ApCREB1, or pNEXδ-ApCREB2 was coinjected with CRE-luciferase () into ∼30 neurons of abdominal ganglion. To examine transcriptional activity of ApAF (WT or mutants) or ApC/EBP, pNEXδ-ApAF (WT, S175A, S266A, or S175/266A) or pNEXδ-ApC/EBP was coinjected with ERE-luciferase () into ∼30 sensory cells of pleural ganglion. ERE (; ) and CRE sequences were inserted into the BglII site of the pGL3-promotor vector (Promega), producing ERE- and CRE-luciferase, respectively. The inserted sequences are 5′-CGCGCGCATGCGGGGCCCAGATC -3′_ (ERE) and 5′-TGGGCCCCGCGCGGATC-3′_ (CRE) (core enhancer sequences are underlined; ). ERE sequences in this reporter construct contain a core CAAT motif, which is homologous to the binding sequences for ApAF homodimer and ApAF–ApC/EBP heterodimer (). As an internal control for gene expression, pNEX2-lacZ was included in the injection solution (). The DNA concentrations (mg/ml) in the injection solution of activators (pNEXδ-ApAF [WT or mutant], pNEXδ-ApC/EBP, pNEXδ-ApCREB1, or pNEXδ-ApCREB2), reporter (CRE- or ERE-luciferase), and pNEX2-lacZ were 0.1, 0.7, and 0.05, respectively. 24 h after microinjection, the injected ganglia were exposed to one pulse (5 min) of 10 μM 5-HT to examine the effect of ApAF regulated by 5-HT on transcriptional regulation. Ganglion (sensory cluster of pleural ganglion) cell extracts were made 48 h after microinjection using 30 μl of reporter lysis buffer per ganglion (Promega) and luciferase, and β-galactosidase assays were performed as previously described (). To examine the effect of dsRNA of ApAF on long-term synaptic facilitation, the sensory neurons of sensory-to-motor synapses were injected with the injection buffer containing 500 μg/ml of pNEXδ-EGFP and 500 μg/ml of dsRNA (ApAF or luciferase as a control). To make dsRNA of ApAF or luciferase, each cDNA template was linearized by HindIII–BamHI digestion of pLitmus28i-ApAF or BamHI–KpnI digestion of pLitmus28i-luciferase. dsRNAs were generated using a MEGAscript RNAi kit (Ambion) as previously described (). To investigate the effect of ApAF (WT or mutants) related to ApC/EBP or ApCREB2 on synaptic facilitation, sensory cells of sensory-to-motor synapses were injected with 500 μg/ml pNEXδ-ApAF (WT or mutants), 500 μg/ml pNEXδ-ApC/EBP, and 500 μg/ml pNEXδ-ApCREB2. 500 μg/ml pNEXδ-hrGFP () was used as an injection marker. When only either pNEXδ-ApAF or pNEXδ-ApC/EBP was injected, each plasmid was injected at concentration of 1 mg/ml instead of 500 μg/ml. To inhibit CRE-mediated gene expression using CRE oligonucleotide () or K-CREB (), the sensory neurons of sensory- to-motor synapses were injected with the injection buffer containing 50 μg/ml of CRE (or ERE) or 500 μg/ml of K-CREB before 5-HT treatment. The oligonucleotides used were as follows: CRE, TGGCATCTACGTCAAGGCTT; ERE, GATCATATTAGGACATGCGG (). Culture dishes and media were prepared as previously described (). Cultures of sensory neurons and sensory-to-motor cocultures were made as previously described (; , ). The motor cell was then impaled intracellularly with a glass microelectrode filled with 2 M K-acetate, 0.5 M KCl, and 10 mM K-Hepes (10–15 MΩ), and the membrane potential was held at 40 mV below its resting value. The excitatory postsynaptic potential (EPSP) was evoked in an LFS motor cell by stimulating the sensory neurons with a brief depolarizing stimulus using an extracellular electrode. The initial EPSP value was measured 24 h after microinjection. The cultures then received one or five pulses of 5-HT (10 μM) for 5 min at 15-min intervals to induce LTF. We applied one pulse of 5-HT to cells overexpressing ApAF and/or ApC/EBP to examine whether ApAF and/or ApC/EBP overexpression could convert STF to LTF. Some cultures were exposed for 2 h starting 30 min before the first application of 5-HT to membrane-permeable PKA inhibitor (10 μM KT5720; Calbiochem; ). In the MAPK experiments, 30 μM PD98059 was applied to the cultures 1 h before the application of 5-HT and during the 5-HT treatment. Fig. S1 shows the amino acid sequence of ApAF in Fig. S2 shows the interaction of ApAF with ApC/EBP and ApCREB2. Fig. S3 shows the lack of ApC/EBP mRNA induction by one pulse of 5-HT. Online supplemental material is available at .

In most eukaryotic cells, the C terminus of α-tubulin is subject to a cycle of detyrosination–tyrosination, in which the C-terminal tyrosine residue of α-tubulin is sequentially cleaved from the peptide chain by tubulin carboxypeptidase (TCP) and readded to the chain by the tubulin-tyrosine ligase (TTL; ). In both animal models and human cancers, TTL is often suppressed during tumor growth, indicating that TTL suppression and resulting tubulin detyrosination represent a strong selective advantage for proliferating transformed cells (; ). In whole animals, TTL is essential for neuronal organization: TTL-null mice die within hours after birth because of the disorganization of vital neuronal circuits (). In , where the tyrosination cycle does not occur but where the structure of the α-tubulin C terminus is conserved, removal of the C-terminal aromatic residue of α-tubulin (a phenylalanine in yeast instead of a tyrosine) disables the interaction of microtubule (MT) plus ends with Bik1p, the yeast equivalent of the mammalian MT plus-end tracking protein cytoplasmic linker protein (CLIP) 170 (). Consistent with a role of tubulin tyrosination in CLIP-170 localization, CLIP-170 is mislocalized in TTL-null neurons, being absent from the distal part of neurites and from growth cones (). In this study, we have used TTL-null fibroblasts, in which, in contrast with neurons, individual MTs are distinct, to probe the influence of tubulin tyrosination on the recruitment of CLIP-170 and of other MT plus-end proteins (+TIPs) at MT ends. We find that tubulin tyrosination is important for proper localization of +TIPs such as CLIP-170, CLIP-115, or p150 Glued, which comprise at least one cytoskeleton-associated protein glycine-rich (CAP-Gly) MT binding motif (; ), whereas other +TIPs, such as EB1, EB3, CLIP-associating protein (CLASP), or mitotic centromere-associated kinesin (MCAK), interact similarly with tyrosinated or detyrosinated polymers. We provide evidence that TTL suppression and resulting tubulin detyrosination induce abnormalities in spindle positioning and in cell morphology. TTL-null cells contain massive amounts of detyrosinated (Glu) tubulin but also variable amounts of tyrosinated (Tyr) tubulin originating from tubulin synthesis (). Here, when double stained for Tyr and Glu tubulin, interphasic WT cells displayed extensive tyrosination of cytoplasmic MTs, with only a small subset of Glu MTs (), known to correspond to poorly dynamic polymers (; ; ). Interphase TTL-null fibroblasts contained extensive arrays of Glu MTs and variable amounts of Tyr tubulin (). In four independent experiments, 11–15% of TTL-null fibroblasts displayed only background signal when stained with Tyr tubulin antibody (, Tyr−), 50–60% of the cells showed partial staining of the MT network (, Tyr+−), and 30–40% of the cells showed distinct staining of the whole MT network (, Tyr+). All Tyr− cells were also negative for the G2/S marker cyclin A, indicating that all were G1 cells (). When MT stability was probed using cell exposure to nocodazole (unpublished data), the bulk of cytoplasmic MTs in all WT or TTL-null cells were nocodazole sensitive, indicating that in interphase TTL-null cells, MT detyrosination was not due to increased MT stability. In WT mitotic cells, Tyr tubulin antibody stained both the central region of the spindle and the cell periphery, which contains a free tubulin pool and astral MTs. The Glu tubulin signal was low, with a distinct staining of centrioles (). In TTL-null mitotic cells, the Tyr tubulin signal was evident in prophase and remained strong through the whole mitotic process (). In metaphase or anaphase cells, spindle MTs were stained with both Tyr and Glu tubulin antibody. Interestingly, in all metaphase cells examined, Tyr tubulin staining was apparently enriched in the core region of the spindle, with no detectable signal at the periphery of the cell, where Glu tubulin staining yielded both a diffuse signal corresponding to the soluble tubulin pool and a distinct staining of astral MTs. Collectively, these results indicate that tubulin detyrosination is maximal in interphase TTL-null fibroblasts, where G1 cells can lack detectable Tyr tubulin. Tyr tubulin is present in G2/S and throughout mitosis. Additionally, in metaphase TTL-null cells, Tyr tubulin seems to be preferentially recruited to core spindle MTs, which include the slowly treadmilling kinetochore- to-pole fibers (), whereas the soluble tubulin pool and the dynamic astral MTs are essentially composed of Glu tubulin. Previous evidence has suggested that CLIP-170 association with growing MT plus ends is inhibited by tubulin detyrosination, whereas EB1 localization is unaffected (; ). To assess CLIP-170 localization as a function of tubulin tyrosination, WT or TTL-null cells were triple labeled with EB1 antibody, CLIP-170 antibody, and Tyr tubulin antibody (). In WT cells, CLIP-170 colocalized with EB1 in comet-like structures at MT ends. Similar colocalization was observed in Tyr+ TTL-null cells. In contrast, Tyr− TTL-null cells were essentially devoid of CLIP-170 comets. For quantitative analysis of CLIP-170 localization on growing MT plus ends, 30 EB1-labeled MT ends were selected in each individual cell and examined for CLIP-170 labeling, and the percentage of CLIP-170–positive MT ends (CLIP-170+) was determined. All measurements were performed blind to genotype. In interphase WT cells, >87% of the EB1+ MT ends were also positive for CLIP-170 (). In TTL-null cells, a strong relationship between tubulin tyrosination and CLIP-170 localization was evident, with only 14% of the EB1-labeled MT ends being also positive for CLIP-170 in cells classified as Tyr−, compared with 80% of positive ends in cells classified as Tyr+ (). CLIP-170/EB1 fluorescence ratios were higher in WT cells than in Tyr+ TTL-null cells and close to background levels in Tyr− TTL-null cells ( a). The few CLIP-170 comets visible in Tyr− cells had low CLIP-170 signal compared with CLIP-170+ comets in Tyr+ or WT cells ( b). In a series of control experiments, we checked that CLIP-170 expression levels were not different in WT cells compared with TTL-null cells (). We also checked by cyclin A staining that a similar proportion of G1 (75%) and G2/S cells were present in WT or TTL-null cell populations and that in WT cells, CLIP-170 localization was not detectably different in cyclin A− cells compared with cyclin A+ cells (unpublished data). CLIP-170 localization at growing MT plus ends could be influenced by MT growth rates. We used video microscopy and fluorescent protein constructs to test whether MT growth rates differed as a function of the tubulin tyrosination status. In cells double transfected with GFP–CLIP-170 and EB3-RFP, GFP–CLIP-170 behaved as endogenous CLIP-170, associating with MT ends in WT cells and in Tyr+ TTL-null cells but not in Tyr− TTL-null cells, indicating that tyrosination affects CLIP-170 localization within a large range of protein expression levels (Fig. S1, available at ). We measured MT growth rates in double-transfected cells by video microscopy tracking of EB3-RFP–labeled MT ends. Interestingly, we observed similar MT growth rates in WT cells compared with TTL-null cells, whether or not CLIP-170 was correctly localized ( and Videos 1–3), indicating that CLIP-170 mislocalization is not due to impaired MT growth in TTL-null cells but likely reflects an influence of tubulin detyrosination by itself. In WT mitotic cells, many CLIP-170 and EB1 comets were visible (). In TTL-null mitotic cells, which contain abundant Tyr tubulin, CLIP-170 comets were also conspicuous (). However, interestingly, in TTL-null metaphase cells, whereas CLIP-170 staining of the core spindle was similar to that observed in WT cells (and as found in previous work []), CLIP-170 comets were undetectable at the cell periphery, compatible with a lack of CLIP-170 interaction with the detyrosinated astral MTs (see and ). Comets reappeared at the cell periphery in telophase, presumably when Tyr tubulin from the disassembling core spindle MTs redistributed among astral MTs. Collectively, our results indicate that in TTL-null cells, MT tyrosination is critical for CLIP-170 association with MT ends. The evidence we have shown indicates a strong correlation between MT tyrosination and CLIP-170 localization. We used experimental manipulation of the tubulin tyrosination level in TTL-null cells to test the causal nature of this relationship. For Tyr tubulin depletion, we made use of previously described α-tubulin siRNAs, which suppress α-tubulin synthesis and thereby the Tyr tubulin pool in TTL-null cells (). WT controls or TTL-null fibroblasts were transfected with GFP–CLIP-170 and exposed to α-tubulin siRNA. These siRNAs have no effect on tubulin tyrosination in WT cells, where tubulin is tyrosinated by TTL. In these cells, the MT network was somewhat disorganized (; ), but MT ends remained uniformly positive for GFP–CLIP-170. In contrast, α-tubulin siRNAs induced extensive suppression of Tyr tubulin in TTL-null cells (). Accordingly, we did not observe CLIP-positive cells among siRNA-treated TTL-null fibroblasts (). Conversely, cotransfection of TTL-null cells with TTL cDNA together with GFP–CLIP-170 dramatically increased both the level of MT tyrosination and the proportion of cells with CLIP-170 at MT ends (). We observed similar results with endogenous protein as with transfected GFP–CLIP-170 (unpublished data). In control experiments, no change in the proportion of G1 versus G2/S cells was detectable after the various cell treatments. These results strongly indicate that CLIP-170 mislocalization is causally related to tubulin detyrosination in TTL-null cells. Is the influence of tubulin tyrosination on CLIP-170 interaction with MTs direct or indirect? To approach this question, we examined the interaction of the CLIP-170 HD, which contains the MT binding domain CAP-Gly (; ), with Tyr or Glu MTs, both in vivo and in vitro. When WT fibroblasts were transfected with CLIP-170 HD fused with YFP (CLIP-170–HD–YFP), CLIP-170–HD–YFP formed comets at MT ends and decorated MTs lengthwise (). Interestingly, the comets and the lengthwise MT signal were both lacking in TTL-null Tyr− cells (). We then tested whether a differential binding of CLIP-170 HD to Tyr or Glu MTs could be reconstituted in purified systems in vitro. CLIP-170 HD (His–CLIP-170 H1; ) was added in substoechiometric amounts to solutions of affinity-purified Tyr or Glu tubulin (). Subsequently, tubulin polymerization was initiated and taxol was added to stabilize MTs. At low ionic strength, CLIP-170 HD was quantitatively absorbed on both Tyr and Glu MTs. We then tested the effect of increasing NaCl concentrations. Taxol-stabilized MTs were then exposed to increasing NaCl concentrations before the cosedimentation assay of CLIP-170 with MTs (). His–CLIP-170 HD showed a double band with a molecular mass of 40 kD as observed in . Interestingly, CLIP-170 HD began to be eluted from Glu MTs at 100 mM NaCl () and was further dissociated from Glu MTs at 200 mM NaCl (), whereas at similar NaCl concentrations, the bulk of CLIP-170 HD was still associated with Tyr MTs (). Interestingly, the differential binding of CLIP-170 HD to Glu or Tyr MTs occurred at NaCl corresponding to the intracellular NaCl concentration (150 mM). CLIP-170 HD binding to Tyr and Glu MTs was then tested at different tubulin concentrations at 150 mM NaCl. CLIP-170 HD was present in supernatants at all concentrations with Glu MTs, whereas it only appears in supernatants at close to saturating concentrations (5 μM CLIP-170 HD for 6 μM tubulin) with Tyr MTs (). At the same NaCl concentration, a His-EB1 construct was still quantitatively absorbed on both Tyr and Glu MTs () and began to be eluted only at 500 mM NaCl, similarly from Tyr or Glu MTs (not depicted). Collectively, these results indicate that tubulin tyrosination interferes directly with CLIP-170 interaction with MTs but does not affect other +TIPs with different MT binding domain, such as EB1. The CAP-Gly MT binding domain of CLIP-170 is also present in the MT end-tracking proteins CLIP-115 and p150 Glued. The HDs of CLIP-115 and -170 are homologous, and in this study, in all qualitative and quantitative cellular assays, endogenous CLIP-115 or the GFP–CLIP-115 construct behaved like CLIP-170, binding to Tyr but not to Glu MTs ( and not depicted). Interestingly, endogenous p150 Glued or the EGFP–p150 Glued construct also behaved like CLIP-170 in similar assays ( and not depicted). In in vitro MT binding assays, a differential binding of p150 Glued to Tyr or Glu MTs was always detectable, although less evident than in the case of CLIP-170 (). We have shown that, in contrast, EB1 and EB3, two highly related proteins whose MT binding domain does not comprise a CAP-Gly motif, are unaffected by MT tyrosination in vivo ( and Videos 1–3) and in vitro (). MCAK, which has recently been shown to end track MTs (), and CLASPs, a previously identified +TIP (), also decorated MT plus ends in Tyr− TTL-null cells (). We conclude that tubulin tyrosination affects MT interactions with CAP-Gly +TIPs in the absence of obvious effects on the behavior of other +TIPs. TTL-null cells then offered an original system in which a single amino acid deletion in α-tubulin, not affecting protein folding, specifically disrupted the interaction of MTs with CAP-Gly +TIPs. The most obvious cellular consequences of such a disruption are shown in the next sections. CAP-Gly +TIPs are involved in MT interactions with the cell membrane (; ; ; ; ; ; ; ), and there is evidence that tubulin detyrosination affects cell morphogenesis in neurons (). In nonneuronal cells, it has been suggested that the presence of CLIP-170 at MT ends is important for MT-dependent regulations of actin assemblies, such as lamellipodia (; ). Upon examination, WT fibroblasts often displayed a distinct polarity by extending a single lamellipodium (), as expected from the behavior of a normal fibroblast, which extend a lamellipodium in the direction of cell migration (). Similar polarized lamellipodium extension was never observed in Tyr− TTL-null cells, which showed a more or less regular round shape (), whereas Tyr+ TTL-null fibroblasts often displayed numerous extensions with several large lamellipodia, resulting in a conspicuously irregular cell shape (). We used a standard shape factor (defined by 4πA/P, where A is the area and P the perimeter) for quantitative analysis of cell shape. This shape factor varies from 0 to 1, for elongated or circular shapes, respectively. Compared with WT cells, Tyr− TTL-null cells had an increased shape factor, indicative of a more regular, round shape (). Among WT cells, shape factors were similar between G1 in G2/S cells (unpublished data). Tyr+ TTL-null cells had a mean shape factor comparable to that of WT cells, although there was an excess population of cells with a low shape factor (), indicative of irregular and elongated cell shape. In a further quantitative analysis, the ratio of the perimeter of large lamellipodia extensions to the total cell perimeter was dramatically diminished in Tyr− TTL-null cells compared with other cells ( a), and the number of large lamellipodia extensions per cell was increased in Tyr+ TTL-null cells, compared with WT cells ( b). We conclude that cell morphology is perturbed in TTL-null cells, apparently because of abnormalities in cell polarization, which seems inhibited in Tyr− cells and disorganized in Tyr+ cells. Spindle positioning is dependent on +TIP-mediated interactions of astral MTs with the cell cortex (; ; ; ). We used printed micropatterns of fibronectin to study spindle positioning in WT or TTL-null fibroblasts. Micropatterns constrain extracellular matrix organization and thereby organization of both actin assemblies and cell adhesions, which are important for spindle positioning (). We observed similar organization of actin assemblies or of cell adhesions in TTL-null cells and WT cells placed on micropatterns (unpublished data). Spindle positioning then depends crucially on astral MT interactions with the metaphase cell cortex (). These interactions involve CAP-Gly +TIPs (), which fail to associate with astral MT ends in TTL-null cells. The majority of mitotic spindles of WT cells grown on L-shaped patterns were oriented along the hypotenuse (). In contrast, there was a wide dispersion of spindle orientations in TTL-null cells (), with ∼70% of spindles deviating by >15° from the median angle compared with only 30% in WT cells (), indicating an impaired control of spindle positioning in TTL-null cells. In this study, we show that tubulin tyrosination is central for MT interaction with CAP-Gly +TIPs and that in cells, TTL suppression and resulting tubulin detyrosination affect spindle positioning and cell morphology. Tyr tubulin is not fully suppressed in TTL-null cells (), where Tyr tubulin arises from synthesis of new tubulin molecules. Tyr tubulin levels vary in interphase cells that can exhibit complete detyrosination. In contrast, mitotic cells always contain Tyr tubulin. The constant presence of Tyr tubulin in mitotic TTL-null cells compared with interphase TTL-null cells is intriguing. We do not know whether this uniform presence of Tyr tubulin reflects an increased tubulin synthesis when cells enter mitosis, an inhibition of TCP, or both. Based on Tyr and Glu tubulin distributions in metaphase TTL-null cells, our data raise the possibility of a preferential recruitment of Tyr tubulin compared with Glu tubulin in kinetochore-to-pole MTs, with a corresponding enrichment of Glu tubulin in astral MTs and in the soluble tubulin pool. Such a preferential recruitment cannot be explained by a difference in Tyr or Glu MT polymerization properties (). However, recent work in indicates that tubulin incorporation in the treadmilling kinetochore-to-pole MTs during metaphase does not depend only on the intrinsic polymerization properties of tubulin but also on specific regulations that, in , require CLASP (). Possibly, tubulin incorporation in kinetochore-to-pole fibers also requires CAP-Gly proteins or other unknown proteins whose interaction with MTs depends on tubulin tyrosination. We do not know whether mitotic cells could tolerate the complete suppression of Tyr tubulin. We have observed extensively detyrosinated mitotic cells in siRNA experiments (unpublished data). In these cells, the spindle was conspicuously disorganized, but we found it hard to know whether spindle anomalies were due to tubulin detyrosination by itself or to other dysfunctions related to the inhibition of α-tubulin synthesis. In TTL-null cells, tubulin detyrosination affects the recruitment of CAP-Gly proteins at MT ends, whereas EB1 localization is unaffected. Additionally, we observe that tubulin tyrosination affects the interaction of MTs with CAP-Gly+TIPs in purified systems in vitro. Clearly, as suggested by previous studies (; ; ), CLIP-170 and p150 Glued localization at MT ends depends crucially on direct interactions between these +TIPs and MTs, even if interactions between both proteins and their interaction with EB1 are also involved (; ; ; ). The remarkable influence of the C-terminal tyrosine of tubulin on tubulin interaction with CAP-Gly +TIPs is reminiscent of a recent structural study of the interaction of EB1, which has the same C terminus as α-tubulin, with p150 Glued (). In this study, the terminal tyrosine of the EB1 C terminus had a crucial contribution in EB1 interaction with the CAP-Gly domain of p150 Glued (). Very recently, a similar role of the C-terminal tyrosine of EB1 has been reported in the case of EB1 interaction with CLIP-170 (). Such a conservation of the C-terminal tyrosine function is remarkable because the interaction of p150 Glued with tubulin involves several domains () and the interaction of EB1 with CLIP-170 also involves protein domains other than the EB1 C terminus (). It is also remarkable that the role of the C-terminal aromatic residue of EB1 and of α-tubulin in regulating interactions with CAP-Gly +TIPs is conserved among organisms, from yeast to mammals, and may concern additional CAP-Gly proteins (; ). This may explain the conservation of the α-tubulin extreme C-terminal sequence, which has been intriguing, given that this tubulin sequence is not important for the 3D structure of the protein. TTL-null cells offer a new and attractive system to evaluate the function of +TIPs' interaction with MT. Within the framework of this study, we have restricted our analysis to the most obvious cellular phenotypes. We find that TTL suppression in fibroblasts compromises spindle positioning in cells placed on a micropatterned matrix. A recent study has shown that metaphase spindle positioning in cells grown on micropatterns depends both on the organization of the extracellular matrix, which is controlled by the micropattern, and on astral MT interactions with cues on the cell cortex (). Such interactions are mediated by molecular complexes involving CLIP-170 (or CLIP-115) and p150 Glued (; ). It is likely that the inhibition of CAP-Gly +TIPs' interaction with detyrosinated astral MTs in TTL-null cells accounts for the impaired control of spindle positioning. TTL suppression also affects the control of cell shape and of cellular extensions in fibroblasts. In fully detyrosinated cells (Tyr−), where CAP-Gly proteins do not interact with MT ends, cell polarization is diminished, and this agrees with previous work indicating a central role of CLIP-170 in MT-dependent regulations of actin assemblies (). In TTL-null cells containing enough Tyr tubulin to localize CAP-Gly +TIPs at MT ends, the cell morphology was still perturbed, with multiple cell extensions and an irregular shape. It may be that there are perturbations of CAP-Gly +TIPs' function in cells where MTs are still extensively detyrosinated and where MT ends associate with diminished levels of CLIP-170 or that tubulin detyrosination affects proteins other than+TIPs. These possibilities are under examination in our laboratories. What is the relationship of the phenotypes observed in TTL-null fibroblasts with the role of TTL in tumor progression and in brain development? Impaired control of spindle positioning has been proposed as one factor favoring tumor invasiveness (). In TTL-null tissues in mice, putative defects in spindle positioning due to tubulin detyrosination are apparently compensated, possibly through the action of geometrical constraints (). The situation may be different in the brain, where TTL-null mice display a variable degree of ventricle enlargement, indicative of cell loss (). Spindle positioning in neuronal progenitors is a complex and highly regulated process crucial for the control of progenitors' proliferation/differentiation in the neuronal epithelium (; ). Spindle positioning could be perturbed in TTL-null mice with resulting abnormalities in neuronal differentiation/proliferation equilibrium. With regard to the control of cell morphology, our data are consistent with studies that indicate an important role of CAP-Gly +TIPs for the control of cell shape (; ; ). Such defects in cell shape control are apparently compensated in TTL-null nonneuronal tissues, which are apparently normal, whereas they are probably deleterious in neurons, which exhibit an erratic time course of neurite extensions (). Neuronal cells may be especially sensitive to defects in cell shape controls, given the extreme size and complexity of neurite extensions. Finally, our data suggest that defects in cell shape may be associated with alterations of cell polarity. Such defects may affect cell motility as well as cell–cell or cell–matrix adhesions, which could be involved in the facilitating effect of tubulin detyrosination on tumor growth. Our data demonstrate that tubulin tyrosination is required for MT interaction with CAP-Gly proteins but do not give definite clues to understand why a great number of eukaryotic cells developed a tyrosination cycle by introducing a detyrosination reaction. The bulk of tubulin is detyrosinated in differentiated cells (), and it may be that tubulin detyrosination is used to disconnect MT–membrane interactions in terminally differentiated cells that do not need to change shape or to divide any longer. The identification of TCP and subsequent TCP suppression will be necessary for a full understanding of the tyrosination cycle. Clearly, though, this cycle is an important aspect of MT physiology, as it is involved in MT functions that are conserved, vital, and important for tumor progression. The following primary antibodies were used: Glu tubulin (), Tyr tubulin (clone YL1/2; ); EB1 (BD Biosciences), GFP (Invitrogen), cyclin A (clone CY-A1; Sigma-Aldrich), CLIP-115– or CLIP-170–specific antisera (numbers 2238 and 2360, respectively; ; ), and p150 Glued (clone 1; BD Biosciences). Mouse embryonic fibroblasts (MEFs) were prepared from embryonic day 13.5 embryos, as previously described (). Cells were maintained at 37°C with 5% CO and 3% O. The experiments shown in the present paper were done using pooled fibroblasts from four different embryos, either WT or TTL null. GFP–CLIP-170 cDNA was provided by F. Perez (Institut Curie, Paris, France); EB3-RFP was provided by V. Small (Institute of Molecular Biotechnology, Vienna, Austria); and GFP–CLIP-115 (); EGFP–p150 Glued (), GFP-CLASP1 (), and GFP-MCAK () have been described. Mouse TTL coding sequence was inserted in pcDNA3 (Invitrogen). CLIP-170–HD–YFP (aa 1–278) was generated in pCLink vector. MEF cultures were transfected using 1 μg of DNA and Lipofectamine Plus (Invitrogen). For cotransfection experiments using RFP and GFP constructs, we used 1 μg of total DNA, and pilot transfections were performed to determine the respective proportion of each DNA, which led to transfected cells positive for both RFP and GFP. For rescue experiments, we performed cotransfection with GFP–CLIP-170 and TTL cDNA (ratio 1:4); the efficiency of the cotransfection was evaluated by the presence of both GFP–CLIP-170 protein and of Tyr tubulin (as a result of the presence of TTL). siRNA specific for α-tubulin were used as previously described (). L-shaped fibronectin micropatterns that were 35 μm long were made as described previously (). MEFs were resuspended in DME 10% FBS and deposited on the printed coverslip at a density of 10 cells per cm. To analyze MT growth rates, we cultured MEFs in 35-mm glass Petri dishes (Iwaki). 48 h after cotransfection with GFP–CLIP-170 and EB3-RFP cDNAs, cells were maintained in DME 10% FBS and placed in a humidified incubator at 37°C with 5% CO inside the video microscopy platform. Fluorescent images were captured every 3 s with a charge-coupled device camera (CoolSNAP HQ; Roper Scientific) using a 100×/1.3 Plan-Neofluar oil objective in an inverted motorized microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) controlled by MetaMorph software (Universal Imaging Corp.). To measure MT growth rate, we quantified EB3-RFP velocity in >30 cells for each phenotype. To analyze mitotic spindle position, MEFs were plated on L-shaped fibronectin micropatterns and placed inside the video microscopy platform. Time-lapse images were collected every 3 min in multiposition in transillumination. To quantify mitotic angles, we designed a line perpendicular to the metaphase plate and measured the angle between this line and the micropattern hypotenuse using MetaMorph software. Affinity-purified Tyr or Glu tubulin were prepared as previously described (). For storage (−80°C), both forms of tubulin were transferred to PEM buffer (pH 6.65, 100 mM Pipes, 1 mM EGTA, and 1 mM MgCl) made with DO instead of HO, containing 1 mM GTP and 10 mM MgCl. His–CLIP-170 HD (H1 fragment; aa 1–350; ) cloned into pET19b vector was provided by F. Perez. His–p150 Glued HD (aa 46–253) and His-tagged EB1 (aa 1–268), both cloned into pET28 vector, were as described (; ). All constructs were expressed in bacteria, and proteins were purified as previously described. Tubulin was centrifuged for 10 min at 200,000 , and polymerization was performed by incubating Tyr or Glu tubulin with His–CLIP-170 HD, with His–p150 Glued HD or with His-EB1 in PEM with 1 mM GTP, 5 mM MgCl, and 20% glycerol at 37°C for 20 min. 20 μM taxol was added for 10 min, and the presence of MTs was routinely checked using immunofluorescence microscopy or EM. Taxol-stabilized MTs were then exposed to increasing concentrations of NaCl (0, 100, 150, 200, and 300 mM in PEM 20% glycerol) for 10 min at 37°C. The reaction mixture was ultracentrifuged in a PEM 60% glycerol cushion for 30 min at 250,000 at 30°C, and supernatants and pellets were analyzed by SDS-PAGE gels. More than five independent experiments were achieved with His–CLIP-170 HD and His–p150 Glued HD, Coomassie blue–stained gels were scanned with a scanner (PowerLook 1120; Umax), and bands were quantified using Image Quant software. Fig. S1 shows GFP–CLIP-170 localization in WT or TTL-null fibroblasts. Video 1 shows MT growth rate in a WT fibroblast. Video 2 shows MT growth rate in a TTL-null CLIP+ fibroblast. Video 3 shows MT growth rate in a TTL-null CLIP– fibroblast. Online supplemental material is available at .

The portion of the cytoskeleton that is closely associated with the cytoplasmic surface of the plasma membrane is often called the membrane skeleton (MSK; ; ; ; ). The term MSK is useful partly because this part of the cytoskeleton is expected to differ from the bulk cytoskeleton in terms of its structure and protein composition, for its interactions with the plasma membrane in general and with specific molecules in the plasma membrane, and also because it plays important roles in a variety of membrane functions. It is involved in the localization of transmembrane proteins at specific sites in the cell membrane (; ) and in endocytosis and exocytosis (; ) in various cell types. It also provides the plasma membrane with the mechanical strength and resilience to withstand the stress and shear forces from the outside environment, which is well established in the thick cortical actin layers in immune cells () and in the spectrin–actin network in red blood cells (; ). Therefore, the MSK works as a part of the plasma membrane as well as a part of the cytoskeleton. It is a truly interfacial structure between the bulk cytoskeleton and the 2D bilayer of the plasma membrane. Recently, a new function of the MSK has become apparent. It was proposed that a part of the MSK is directly and closely associated with the cytoplasmic surface of the plasma membrane, and this part induces partitioning of the cell membrane with regard to the translational diffusion of membrane molecules based on high speed single-particle tracking data on membrane proteins and lipids (; ). In the short-time regime, these membrane molecules are temporarily confined within the compartments delimited by the MSK mesh, and, in the long-time regime, they undergo macroscopic diffusion by hopping between these compartments (MSK fence model). In the fence model, as a result of the collision of the cytoplasmic domains of transmembrane proteins with the MSK, transmembrane proteins are temporarily confined in the MSK mesh (; ; ; , ; , ; ; ; ; ; ; ). Lipid molecules also undergo hop diffusion, which might be explained by the anchored protein picket model (; ; ). In this model, various transmembrane proteins anchored to the actin-based MSK might effectively act as rows of pickets against the free diffusion of all of the molecules incorporated in the cell membrane as a result of steric hindrance and circumferential slowing (a hydrodynamic frictionlike effect, which propagates quite far from the immobile protein surface; without this effect, pickets will not be effective for blocking diffusion; , ) of the immobile picket proteins anchored to and lined up along the MSK. Lipid movement is affected only by pickets, whereas both pickets and fences would act on transmembrane proteins. These MSK picket-fence effects would be dramatically enhanced when the membrane receptor molecules form signaling complexes upon ligand binding as a result of receptor oligomerization and/or binding of the cytoplasmic signaling molecules to the receptor, leading to the trapping of signaling complexes in the MSK mesh, where the extracellular signal is received. This would enable spatial confinement and regulation of the downstream signaling events (; ). Despite the importance of the MSK functions and the long history of its study using EM (; ; ; ; ; ), our knowledge of its structure and the overall distribution over the plasma membrane has been very limited. For example, we do not know whether the MSK exists everywhere on the cytoplasmic surface of the cell membrane, how extensive the spatial variations of MSK mesh size is, and whether and how MSK interacts with other structures in the plasma membrane such as clathrin-coated pits (CCPs), caveolae, and cell adhesion structures. Even the structure of the MSK of the human red blood cell ghost, a traditional paradigm for MSK studies, is not satisfactorily understood (; ; ; ; ; ). In this study, to further advance our understanding of the MSK structure and function, we observed the undercoat structure on the cytoplasmic surface of the plasma membrane of cultured mammalian cells using rapid-freeze, deep-etch, immunoreplication EM. We paid special attention to the following three points. First, we tried to consistently prepare and observe large plasma membrane fragments (>10 μm in diameter) to facilitate inspections of very large plasma membrane areas. Almost all of the previous MSK studies, including those cited above, investigated the ultrastructural features of the structure of interest, but within a very limited view field. By observing these large membrane surfaces, the spatial variations of the MSK mesh size and of the number density of CCPs and caveolae can be reliably examined. Second, the 3D reconstruction of the undercoat structure within 100 nm from the cytoplasmic surface of the plasma membrane was performed using electron tomography for the platinum-replicated samples: 97–141 images for a specimen tilted at different angles (every 1°) with respect to the incident electron beam in the range of ±48–70° were obtained and then converted to 100–121 sliced images of every 0.85–1.34 nm for the 3D reconstructed images (; ; ; ; ). Third, using the 3D reconstructed images of the MSK structure within 13.6 nm (16 slices of 0.85-nm thickness) from the cytoplasmic surface, the MSK mesh size distribution on the cytoplasmic surface of the plasma membrane was determined. This part of the MSK, which is closely associated with the cytoplasmic surface of the plasma membrane, might form the compartment boundaries for partitioning of the plasma membrane for the diffusion of membrane molecules, thus determining the compartment size. Therefore, it is interesting to compare the distribution of the MSK mesh size on the membrane determined this way and that of the compartment size sensed by membrane molecules. Because the compartment size distributions for membrane molecules are very different between normal rat kidney fibroblast (NRK; median = 230 nm) and fetal rat skin keratinocyte (FRSK; median = 41 nm) cell lines (; ), the distribution of the MSK mesh size on the membrane surface was examined using these two cell lines. Although the compartment size is very different between these cell lines, within each cell type, the histogram for the MSK mesh size on the membrane surface is very similar to that for the diffusion compartment size. This strongly supports the MSK fence and MSK-anchored transmembrane protein picket models. is a typical electron micrograph providing a bird's-eye view of the cytoplasmic surface of a large area of the upper cell membrane of a cultured NRK cell. Many such EM images showing the cytoplasmic surfaces of large cell membrane fragments were obtained for NRK and FRSK cells, suggesting that the entire (upper) plasma membrane, except for the places where CCPs and caveolae exist, is coated with the filamentous netlike structure. (A and B), which was obtained for an NRK cell () and an FRSK cell (), shows the magnified images of the cytoplasmic surface of the plasma membrane, exhibiting extensive filamentous netlike structures, which are the MSK. The presence of clathrin-coated structures shows that this is indeed the cytoplasmic surface. The striped banding patterns with a 5.5-nm periodicity on individual filaments are characteristic of actin filaments and, thus, indicate that these are actin filaments (; ; ; ). Because almost all of these filaments contain this striped pattern, it is concluded that the MSK is predominantly composed of actin filaments. This was also confirmed by immunogold staining (see and related text). The electron micrograph shown in the inset in indicates the spatial resolution: because each band in the striped pattern with a 5.5-nm periodicity is visibly separated, the effective resolution is thought to be ∼2 nm (both the thickness of the platinum coating and the platinum granule size are ≤2 nm; ; ). The MSK structure observed here on the upper cell membrane is similar to that on the bottom cell membrane (the part of the cell membrane facing the coverslip) observed previously (). These results suggest that the cytoplasmic surface of a portion of the upper cell membrane >10 μm in diameter was visualized with a spatial resolution of ∼2 nm, which is much smaller than the width of a single actin filament or the repeat distance of the stripes. As shown in and (A and B), the MSK is likely to cover the entire cytoplasmic surface of the upper cell membrane except for the places where CCPs and caveolae are present in both NRK and FRSK cells. Such a notion of the complete coverage of the cytoplasmic surface of the plasma membrane by actin filaments might have existed for >30 yr in a part of the EM community (; for review see ), but the data specifically indicating that the actin filaments of the MSK may cover the entire cell membrane has not been presented in the literature, as done here, nor shared in the cell biology community. The EM observations shown in this study are consistent with the MSK fence and anchored transmembrane protein picket models, in which the entire plasma membrane except for the specific membrane domains is partitioned into many small compartments with regard to lateral diffusion of the molecules incorporated in the plasma membrane. To further examine whether the MSK is predominantly composed of actin filaments (and partly because the 5.5-nm periodicity of the banding pattern is somewhat difficult to discern in some of the filaments), we examined it using an indirect immunolabeling method with 5-nm-diameter colloidal gold particles (see Materials and methods; ). On the filaments with striped patterns, the enlarged images () show the presence of many colloidal gold actin probes, which appear as distinct white spots surrounded by somewhat blurred white halos, reflecting the platinum shadow over the antibody molecules attached to the gold particle. The electron micrographs in revealed that almost all of the colloidal gold probes are bound to the filaments located on the cytoplasmic surface (yellow dots). Therefore, it is concluded that actin is the main constituent molecule of the MSK. The 3D structure of the undercoat within 100–134 nm from the cytoplasmic surface of the plasma membrane, which includes CCPs, caveolae, and the actin-based MSK, was reconstructed using electron tomography for the platinum-replicated samples. Based on the 97–141 tilt images acquired in the range of ±48–70° every 1° step for a single EM view field, 100–121 sliced images of every 0.85–1.34 nm perpendicular to the z axis (parallel to the image obtained at 0° of the tilt angle) were calculated by a computer (long wavelength [≥∼500 nm] undulations of the cell membrane were corrected by the 3D reconstruction software IMOD). The 3D image was reconstructed based on these serial thin slices. Representative images obtained for an EM view field are shown in Video 1 (131 tilt images; an anaglyph produced from images taken at ±12° is shown in ) and Video 2 (showing the 3D image by rotating the 3D reconstructed undercoat structure; a typical view is shown in ; videos are available at ). Throughout the present research, this protocol was used to obtain 3D images. In these images, because of their 3D representation, it is especially clear that the MSK, which is mostly composed of actin filaments, generally spreads along the membrane, covering almost the entire cytoplasmic surface of the upper membrane except for the places with caveolae and CCPs. In addition, CCPs and caveolae are very closely associated with the actin filaments in the MSK, as seen in these images and also in (A and B) and 3. These results are consistent with , , and , but in NRK cells studied here, many more actin filaments were found to be associated with each CCP or caveola. Furthermore, 92 and 93% of CCPs and caveolae ( = 200) were bound by the actin filaments. These results are consistent with the requirement of filamentous actin for CCP internalization (; ). Many short, thin filaments protrude toward the cytoplasm, mostly perpendicularly, from the membrane surface (they were short probably because they were broken at the time of the membrane rip off; , arrows). Note that these perpendicular filaments are almost always connected to the MSK network lying on the cytoplasmic surface (see the tips of the arrows; ). Thus, the part of the MSK that is located on the cytoplasmic surface is connected three dimensionally to the cytoskeleton. Together, they will provide mechanical support for the membrane and the force for deforming the membrane. The part of the actin-based MSK that is in contact with the cytoplasmic surface of the cell membrane has been proposed to partition the cell membrane into 30–230-nm compartments by the fence and picket effect (; ; ). If these fence and picket models are correct, the distribution of the mesh size of the MSK on the cytoplasmic surface of the plasma membrane would be practically the same as that of the compartment size determined by diffusion measurements of membrane molecules. To carry out this examination, the 3D reconstruction of MSK by electron tomography provides a unique opportunity because the obtained images provide quantitative data on how far the individual filaments are located from the membrane surface. In , a typical MSK structure quantitatively analyzed in this study is shown in an anaglyph, and its 8.5-nm–thick sections (created by superimposing 10 0.85-nm sections) of the MSK of an NRK cell, starting from the cytoplasmic side toward the membrane, are shown (; a series of the original tilt images is shown in Video 3, and a series of sliced images of every 0.85 nm is shown in Video 4, available at ). The actin-based MSK is visible on image sections 81–110. Individual actin filaments, forming a network as well as bundles, can be identified. Given the high density of the actin filament meshwork, which is much smaller than the optical resolution, conventional fluorescence microscopy will be unable to visualize the MSK meshwork and can visualize only the bundles of actin filaments. The filaments of the MSK that are directly associated with the cytoplasmic surface of the plasma membrane and may be involved in partitioning the plasma membrane were systematically determined. Out of the stack of 121 image slices taken every 0.85 nm from the cytoplasmic surface (∼100-nm thick altogether), 16 consecutive image slices from the membrane surface (∼13.6-nm thick altogether) were used for this analysis (). In (four images on the right) and (the second to fourth images), the boxed areas in the left-most images were expanded, and the sections of every 1.7 nm (superposition of two 0.85-nm–thick slices; 330 × 330 nm) are displayed between 0 and 11.9 nm. Using these sections, the filaments that are closely associated with the cytoplasmic surface of the cell membrane were determined. Because the thickness (width in the image) of the actin filament after platinum shadowing is between 9 and 11 nm (consistent with ) and the thickness of the platinum replica is ≤2 nm (consistent with and ), the height of the actin filament that is associated with the membrane will be 7–9 nm (because the height is given by the actin thickness and one replica thickness, whereas the width in the image is determined by the actin thickness plus two replica thicknesses), with 8 nm being a reasonable estimate. In the series of electron tomography sections shown in (A and B), the existence of three major classes of filaments with regard to the distance from the membrane surface can be discerned (for details of this analysis, see Materials and methods). The first class of filaments is distinct in computer-reconstructed sections close to the cytoplasmic surface of the plasma membrane, even in the first ∼0–1.7-nm section (because the contrast is reversed in these micrographs, they look more lucent or white), but fade out of the reconstructions 8–10 nm away from the membrane surface (for details, see Materials and methods). These filaments are drawn in green in . We interpret that these filaments are in true contact with the plasma membrane (the gap between the filament and the inner membrane surface is <0.85 nm) because they can be seen clearly even in the first 0.85-nm section. These filaments are likely to be the significant ones for generating membrane corrals. The third class of filaments is not apparent in sections closest to the plasma membrane but becomes clear some distance away from it (>2–4 nm) and also does not fade out until ∼14 nm. We interpret these as being filaments that definitely do not contact the plasma membrane directly and, thus, should not contribute to forming corrals. The second and third classes of filaments are drawn in red in . Therefore, we considered that only the first class of filaments (those drawn in green in ) forms the MSK fences and pickets, and the area surrounded by these filaments is colored green in the 0–6.8-nm section shown in . Note that areas are excluded from this analysis in which bundles of actin filaments are present (e.g., the structure crossing diagonally from the bottom left to the top right in ), actin filaments are too crowded to be individually discerned, an actin filament is terminated in the middle of a domain (domains that contain a loose end of an actin filament), or CCPs, caveolae, and the smooth surface membrane invaginations are present (the white regions in ). Similar determination of the MSK meshwork was performed for FRSK cells. Representative meshes of the MSK are shown in (for an FRSK cell, colored to aid in visualization). We performed such analyses for 10 representative stacks of image sections (1,290 × 1,290-nm plane) each for NRK cells and FRSK cells (eight different cell membrane sheets for each cell type) and identified 76 and 1,300 areas bounded by the MSK meshwork, respectively (excluding the regions occupied by stress fibers and other membrane undercoat structures such as CCPs and caveolae; about the same total membrane areas were examined for each cell type, and, thus, the difference in the number of identified areas represents the difference in the area size between these two cell lines). The 2D area size for each domain was measured by Amira software. The distributions of the square root of the area size (the side length, assuming a square shape for the area) for NRK (, pink open bars) and FRSK (blue open bars) cells are shown in . The median values of the area and its square root are 3.9 × 10 nm and 200 nm, respectively, for NRK cells and 2.7 × 10 nm and 52 nm, respectively, for FRSK cells. The size distributions of the compartments for the diffusion of membrane molecules were obtained for an unsaturated phospholipid, -α-dioleoylphosphatidylethanolamine, by and for NRK and FRSK cells, respectively. The distributions of the side lengths for NRK (, pink closed bars) and FRSK (blue closed bars) cells are shown in the histograms in . The median values of the compartment area and the side length are 4.3 × 10 nm and 230 nm, respectively, for NRK cells and 2.1 × 10 nm and 41 nm, respectively, for FRSK cells (). We performed quantitative analyses of the undercoat structure of the cytoplasmic surface of the plasma membrane using electron tomography for samples prepared by a rapid-freeze, deep-etch, platinum replication technique. One of the most important limitations of this technique is that the cell has to be placed in a hypotonic medium at 4°C for 5–15 min to remove the upper cell membrane. However, with this method, large membrane fragments that were covered by the dense MSK meshwork could be obtained, which was important for the purpose of the present research. We obtained the results by specifically addressing the following three questions. Does the dense meshwork of the MSK exist everywhere on the cytoplasmic surface of the cell membrane, and, if so, how is it linked to the bulk cytoskeleton? If so, what is its relationship with other structures of the plasma membrane, such as CCPs and caveolae? How is the distribution of the MSK mesh size right on the cytoplasmic surface of the plasma membrane? The final point is important because this part of the MSK might form the corrals of the plasma membrane for the diffusion of membrane molecules. Therefore, it is interesting to compare the distribution of the mesh size of the MSK directly attached to the cytoplasmic surface of the plasma membrane, as determined by an EM method, with that of the compartment size for the diffusion of membrane molecules. NRK (median size = 230 nm) and FRSK (41 nm) cell lines were selected for such a comparison because their compartment sizes are very different (; ). This will be an interesting test for the MSK fence and MSK-anchored transmembrane protein picket models and became possible by obtaining the 3D reconstructed images of the MSK structure on the cytoplasmic surface of the plasma membrane. Rabbit anti-actin IgG was obtained from Biomedical Technologies, and colloidal gold probes (5-nm diameter) coated with secondary antibodies (produced in goat) were purchased from GE Healthcare. NRK and FRSK cells were maintained in HAM-F12 or DME mediums, respectively, supplemented with 10% FBS (Sigma-Aldrich) under a 5% CO atmosphere at 37°C. The cells used for the experiments were grown in 35-mm plastic dishes to ∼70% confluency, usually 2 d after inoculation. The cells were washed three times with ice-cold Pipes buffer (10 mM Pipes, 100 mM KCl, 5 mM MgCl, and 3 mM EGTA, pH 6.8, which mimics the environment in the cytoplasm somewhat but is slightly hypotonic) and were exposed for 15–30 s to an ice-cold 70% Pipes buffer (the Pipes buffer diluted by a factor of 1.43 with water, making this solution more hypotonic; ). After the buffer on the coverslip was drained, the remaining excess water was removed by filter paper. To expose the cytoplasmic surface of the upper cell membrane, the upper cell membrane was removed from the rest of the cell after it was adhered to a coverslip placed on top of the cell layer (; ). 5 × 5-mm coverslips (Matsunami) coated with positively charged Alcian blue 8GX (Wako; Alcian blue–coated coverslips were prepared by first immersing them in 1% Alcian blue in distilled water at room temperature for 10 min, washing them with distilled water, and drying them in the air) were placed on top of the cells (upper surface facing the medium rather than the coverslip) and incubated at 4°C for 5–15 min. During this period, good contact between the cell surface and the coverslip was developed. Then, the coverslips were gently floated off from the cell using the surface tension of the buffer by slowly adding ice-cold Pipes buffer containing 1% PFA/0.25% glutaraldehyde into the space between the culture dish and the coverslip. When the coverslip floated off, the cells were cleaved, and the upper cell membrane came off with the coverslip. Then, the cells were further fixed by incubation in fresh ice-cold 1% PFA/0.25% glutaraldehyde in Pipes buffer for 10 min. After fixation, the coverslips were washed three times, for 10 min each time, with PBS (8.10 mM NaHPO, 1.47 mM KHPO, 137 mM NaCl, and 2.68 mM KCl, pH 7.4). Each coverslip was placed on the plunger tip of the rapid-freezing device (Eiko; ) with the cytoplasmic surface of the membrane facing down. The specimen was slammed down (free fall) onto a polished pure copper block, which was precooled by direct immersion in liquid helium. The frozen coverslip was placed in liquid nitrogen and was transferred into the freeze-etching shadowing chamber (FR7000-S; Hitachi). The excess ice covering the cytoplasmic surface of the membrane was shaved off with a prechilled glass knife using a microtome placed in the chamber at −140°C or below. The cytoplasmic surface was then etched for ∼10 min after the specimen temperature was raised to −90°C. The etched specimen surfaces were then rotary shadowed with platinum at an angle of 22.5° from the surface and with carbon from the top. The molecules as well as the gold probes localized on the cytoplasmic surface of the cell membrane were immobilized to the deposited platinum (; ). Collodion was applied immediately after the platinum-carbon replicas were removed from the cold chamber to fortify them. The platinum-carbon replica was removed from the glass coverslip in 1% hydrofluoric acid in distilled water. After the replicas were successfully removed from the glass surface and mounted on the grid, the collodion coat was dissolved away in n-pentyl acetate. In this protocol, the sodium hypochlorite solution, which is generally used to remove the replicas from the coverslip and also to clear the membrane and the undercoat structure of the replicas, was replaced with 1% hydrofluoric acid to keep the cell membrane, the undercoat structure, and the immunogold probes that had been attached to these structures on the platinum replicas (1% hydrofluoric acid is likely to only dissolve the glass, leaving the cell membrane molecules bound to the platinum replica; ; ). An additional advantage of using 1% hydrofluoric acid is that the platinum replicas break less often, probably because it does not remove the membrane components and, thus, leaves the replicas rather intact. In addition, to keep as many colloidal gold particles and membrane molecules attached to the platinum replicas as possible, we included Photo-Flo 200 (Kodak), a detergent used to prevent water-drop stains on photographic film in all of the solutions used here (advice given by J. Heuser). After the replicas were washed with distilled water, they were mounted on 100–200 mesh copper grids (Ted Pella) coated with polyvinyl formvar (Nisshin EM) and observed at magnifications of ∼10,000–70,000 with a transmission electron microscope (1200EX; JEOL). The following methodological precautions and improvements were made to reproducibly produce large cell membranes and replicas without excessive fragmentation. An Alcian blue coat rather than poly--lysine coat was used (; ). Before overlaying the coverslips, excess water was removed from the specimen, leaving just enough buffer to cover the cell. To cleave off the upper membrane attached to the overlaid coverslip, the coverslip was floated off very gently by adding cleavage medium (using the surface tension of the buffer to float the coverslip). If this was not performed gently enough, the membrane was fragmented. The frozen sample was shaved with a glass knife, with the angle between the knife and the coverglass adjusted to a shallow angle (<6°) so that most of the excess water and the cytoplasm were removed and the cytoplasmic surface of the cell membrane could be exposed after light etching. Because replicas with too many variations in height tend to break when they are removed from the coverslip and placed on the water surface, removal of the excess cytoplasm helps to avoid replica breakage. Collodion was applied immediately after the replicas were removed from the cold chamber (before the replicas were removed from the coverslip on the water surface) to fortify the replica (a technique learned from T. Baba and S. Ohno). This step also helped to prevent replica breakage when the replicas were removed from the coverslip. After the large replicas were removed from the glass surface, the collodion coat was dissolved away in n-pentyl acetate. A solution of 1% hydrofluoric acid was used to slightly dissolve the glass surface to facilitate the removal of replicas from the coverslip. A detergent, Photo-Flo 200 (Kodak), was included in all of the solutions that contacted platinum replicas. For 3D reconstruction, the replica was imaged at tilt angles of every 1.0° in the range between ±48 and 70° (total of 97–141 images) for a single field by a transmission electron microscope (Tecnai Sphera F20; FEI) equipped with a CCD camera (1,024 × 1,024 pixels). The pixel size at the specimen was 0.85 nm. The image acquisition was fully automated as previously described (). The 100–121 image sections of every 0.85–1.34 nm were obtained by a calculation based on the set of 97–141 tilt images using an IMOD software package running on Linux (University of Colorado; ) . Corrections for the tilt of the specimen and the long wavelength undulations of the membrane were also achieved with IMOD software. 3D rendering (displaying 3D images in different ways) was performed using the Amira DEV software package (Mercury Computer Systems) operating on a Linux system. In the series of electron tomography sections shown in (A and B), the existence of three major classes of filaments with regard to the distance from the membrane surface was found in the following way. The first class is the filaments that are highly electron dense in the first ∼0–1.7-nm section (because the contrast is reversed in these micrographs, they look more lucent or white) and are continuously seen in the image sections up to the ∼6.8–8.5-nm section, which then dim rapidly in the ∼8.5–10.2- and ∼10.2–11.9-nm sections. To quantitatively evaluate such signal intensity changes within individual filaments, we selected points that are clearly seen in the image of the first ∼0–1.7-nm section every 100–250 nm on each filament, measured the signal intensity on each spot (five pixels), and looked for the section where the signal intensity on the spot decreases by >25% from that for the adjacent section closer to the membrane (the signal intensity tends to drop very rapidly around the threshold sections). If the 25% decrease in the signal intensity occurred between the sections of ∼6.8–8.5 and ∼8.5–10.2 nm or between the sections of ∼8.5–10.2 and ∼10.2–11.9 nm, these filaments were categorized into the first class (i.e., those closely associated with the cytoplasmic surface of the cell membrane). These filaments are drawn in green in (different regions within a single filament might become dim in either of these two sections). The third class is the filaments that exhibit dim signals in the first ∼0–1.7-nm section and show higher electron densities in farther sections, at least up to the section of ∼10.2–11.9 nm, before fading out in the ∼11.9–13.6- and ∼13.6–15.3-nm sections (the latter two sections are not depicted). These filaments were again assumed not to contribute to forming membrane corrals. The second and third classes of filaments are drawn in red in . There were regions that were not amenable to such analysis. They were the areas where bundles of actin filaments were present (e.g., the structure crossing diagonally from the bottom left to the top right in ), actin filaments were too crowded to be individually discerned, an actin filament was terminated in the middle of a domain (domains that contain a loose end of an actin filament), or CCPs, caveolae, and the smooth surface membrane invaginations were present. They were excluded from this analysis (, white regions). Video 1 shows a series of 131 tilt images of the undercoat structure on the cytoplasmic surface. Video 2 presents a 3D reconstructed image of the undercoat structure on the cytoplasmic surface of the plasma membrane, which is shown by rotating the reconstructed image. Video 3 shows a series of 97 tilt images of the MSK in an NRK cell, and Video 4 shows a series of 121 sliced images of every 0.85 nm of the MSK of an NRK cell calculated from the data shown in Video 3. Online supplemental material is available at .

Cells generate two major types of actin-based protrusive organelles, lamellipodia and filopodia. Sheetlike lamellipodia contain a branched actin network () and are thought to be the major engine for cell locomotion. Spikelike filopodia are thought to be the cell's sensory and guiding organelles, which function to explore the local environment and form cell–substratum or cell–cell interactions (). Filopodia contain actin filaments that are organized into parallel bundles (; ; ). The proximal part of the bundle is usually embedded in the lamellipodial network, whereas the distal part of the bundle may or may not protrude beyond the leading edge. Nonprotruding filopodia are also called microspikes (or ribs), but, as we documented, microspikes and protruding filopodia are mechanistically related (). Therefore, throughout this paper, we do not distinguish protruding and nonprotruding structures and collectively call them filopodia. Based on our recent kinetic and structural investigation, we proposed the convergent elongation mechanism for filopodial initiation by reorganization of the lamellipodial dendritic network (). The first step in the process is the association of processively elongating actin filaments at their barbed ends that leads to the formation of the so-called Λ-precursors. Subsequently, these self-segregated filaments are bundled to make mature filopodia. Formation of filopodia-like bundles by a similar mechanism was reconstituted in vitro using cytoplasmic extracts or pure proteins (). Cross-linking of actin filaments is proposed to be a critical step in filopodia protrusion because individual long actin filaments lack the stiffness required to efficiently push the membrane (; ). The leading candidate for filament bundling in filopodia is fascin, which is a 55-kD monomeric protein that cross-links actin filaments in vitro into unipolar and tightly packed bundles (; ). In various cells, fascin localizes to filopodial bundles (; ; ; ; ; ) and is highly expressed in specialized cells that are particularly rich in filopodia, such as neurons and mature dendritic cells, as well as in many transformed cells (; ). Formation of filopodia-like bundles in vitro was also dependent on fascin (). However, additional actin cross-linking proteins, including α-actinin, espin, fimbrin, and villin, are involved in the formation of certain parallel actin bundles, such as those in microvilli, bristles, or stereocilia (; ). A recent in vitro study () showed that fimbrin, α-actinin, and filamin, as well as fascin, can support transient motility via actin bundling. Therefore, it is still uncertain whether fascin is sufficient for filopodia formation or multiple bundlers share the role. During the dynamic process of the filopodial life cycle, filament bundling should be coordinated with actin polymerization. Biochemical experiments predict very tight bundling of actin filaments by fascin. The actin-bundling activity of fascin (; ) and fascin localization in cells () are regulated by phosphorylation of serine 39 within the N-terminal actin-binding domain, suggesting that phosphorylation–dephosphorylation cycles of fascin may be coupled with filopodial dynamics. This hypothesis needs to be experimentally tested, as the dynamics of fascin in filopodia has not yet been closely investigated. In this work, we investigated the function and dynamics of fascin during filopodia formation by a combination of RNAi and the expression of phosphomimetic mutants. Our results demonstrate that fascin plays critical roles in the initiation and protrusion of filopodia by providing them the necessary stiffness, and that fascin-mediated cross-linking of actin filaments in filopodial bundles is unexpectedly highly dynamic. We evaluated the presence of known actin cross-linkers in filopodia of B16F1 mouse melanoma cells. Expression of YFP-fascin () and immunostaining with fascin antibody (not depicted) demonstrated that fascin localized throughout the length of all filopodia. Espin, villin, fascin 3, and L-fimbrin were not expressed in these cells, as revealed by microarray analysis (see Materials and methods). These results are consistent with previous reports on tissue-specific expression (), whereas α-actinin and T-fimbrin (the ubiquitous isoform) were expressed. Coexpression of YFP-fascin and CFP–α-actinin showed different distribution of these two cross-linkers (). As expected, α-actinin was found in stress fibers and focal contacts (; Fig. S1 A, available at ), in actin “spots” (), probably corresponding to invadopodia or podosomes (), and weakly in lamellipodia and, occasionally, in the internal, but not the protruding, parts of filopodial bundles (). Fimbrin and espin are actin cross-linkers in microvilli of epithelial cells and in hair cell stereocilia (, ; ; ; ). In B16F1 cells, expressed GFP–T-fimbrin localized to lamellipodia and filopodia, whereas its localization to stress fibers was very weak (). Similar localization of the endogenous protein was detected by immunostaining (Fig. S1 B). Ectopically expressed GFP-espin was targeted to many, although not all, filopodia, but predominantly to their proximal parts (), whereas mCherry-fascin always localized throughout the entire length of filopodia (Fig. S1 C). Thus, molecular marker analysis showed that endogenously expressed actin cross-linkers, other than fascin, did not exhibit specific filopodial targeting, but also localized to other actin structures, lamellipodia, and/or stress fibers. These results suggest that fascin is a major specific bundling protein in filopodia of B16F1 cells, although the contribution of other cross-linkers is not necessarily excluded. The roles of fascin in filopodia formation were investigated by RNAi using pG-SUPER plasmid, which coexpresses small hairpin RNA (shRNA) and GFP simultaneously (). The following three target sequences were used (): Tc, which is common to mouse and human fascin; mouse-specific Tm; and human-specific Th, as a control. Fascin silencing in B16F1 cells was assayed by immunoblotting and immunostaining. For immunoblotting, GFP-expressing cells were collected by FACS 1 d after transfection and cultured for an additional 4 d. Blots () were analyzed by densitometry (two independent experiments), and normalized to a loading control and to the percentage of GFP-positive cells (∼75% on day four). The average reduction in fascin was 90% for the Tc and 85% for the Tm shRNAs. The control Th shRNA did not decrease fascin. Immunostaining showed that expression of Tm or Tc, but not of Th, shRNAs significantly decreased levels of fascin in mouse B16F1 cells (). Conversely, Th, but not Tm, was effective in fascin knockdown in human HeLa cells (Fig. S2, available at ). Phalloidin staining of fascin-depleted cells showed a 4–5-fold decrease in the number of filopodial bundles, whereas lamellipodia were apparently unaffected (). Remaining filopodia did not contain detectable amounts of fascin as determined by immunostaining (not depicted) and were often wavy and running parallel to the leading edge (, right), as if they were buckled because of the compromised stiffness. EM analysis (see the following paragraph) confirmed that these lateral actin-rich structures were bundles of long filaments, but not ruffles. By platinum replica EM, the remaining filopodia of fascin-depleted cells did not contain tightly packed straight actin filament bundles as in control cells (). Instead, they consisted of rather loosely arranged actin filaments that were wavy and ran along the cell edge, which is consistent with light microscopic data. The internal parts of these bundles were remarkably long (9.5 ± 3.9 μm versus 3.0 ± 1.7 μm in control cells; P < 0.001; = 6 cells, 25 filopodia; = 8 cells, 49 filopodia). In contrast, the structural organization of the lamellipodial network looked indistinguishable from normal B16F1 cells. These results establish that fascin is required for filopodia formation in B16F1 cells and suggest that it participates in filopodial protrusion by cross-linking actin filament bundles and, thus, providing them the necessary stiffness. The specificity of fascin knockdown phenotype was confirmed by rescue experiments. YFP-tagged human fascin refractory to RNAi (WT*-fascin) restored filopodia formation in cells expressing Tc-shRNA together with CFP marker (). Almost all cells (90%; = 80) expressing both CFP-Tc and YFP-WT*-fascin had numerous filopodia (5.5 ± 2.5 per 20 μm of the perimeter) containing fascin along their length, which is similar to control cells (6.8 ± 2.6). Rescue also established that fascin tagged with derivatives of GFP is functional. Furthermore, we tested whether different actin filament cross-linkers, i.e., α-actinin or fimbrin, could rescue filopodia formation in the absence of fascin. Cells cotransfected with CFP-Tc and YFP–α-actinin showed α-actinin localizing to focal adhesions and stress fibers, but not to filopodia; phase-contrast images of such cells did not show restoration of filopodia (). In similar conditions, GFP–T-fimbrin was able to partially rescue filopodia formation in ∼50% of cells ( = 31), which showed on average 2.4 ± 0.6 filopodia per 20 μm of the cell perimeter, whereas other cells did not show any signs of rescue. This variability might depend on the level of GFP–T-fimbrin overexpression. Thus, other actin filament cross-linkers cannot fully substitute for fascin function in vivo. Phosphorylation of fascin at serine 39 is important for its actin bundling activity in vitro (; ) and proper localization in vivo (). To examine the roles of serine39 phosphorylation in filopodia formation, we produced point mutants mimicking the active dephosphorylated (S39A) or inactive phosphorylated (S39E) states of fascin. In vitro bundling assays with F-actin and bacterially expressed and purified wild-type (WT), S39A, or S39E fascins confirmed the actin-bundling activity of WT and S39A fascins, but not of S39E fascin (Fig. S3, available at ), which is thus similar to the previously characterized S39D mutant (). WT and mutant fascin proteins were also assayed for their ability to associate with filopodial bundles in detergent-extracted and buffer-incubated cytoskeletons. Long incubation of lysed cells in a stabilizing buffer results in complete dissociation of endogenous fascin, whereas the actin cytoskeleton remains almost intact (). When applied to fascin-depleted cytoskeletons, WT and S39A, but not S39E, fascin associated with filopodia along their length (unpublished data). Expression of GFP-tagged S39A or S39E fascin mutants in B16F1 cells had opposite effects on filopodia formation. The S39A mutant induced long, overabundant filopodia extending from the cell edges, as well as from the dorsal surface (). Quantification of lateral filopodia revealed an ∼1.6-fold increase in filopodia frequency compared with control cells (). The actual degree of filopodia stimulation was even greater because dorsal filopodia were not scored. Expression of the S39A mutant also led to elongation of the protruding parts of filopodial bundles, whereas their internal parts remained of the same length (). In contrast, expression of the S39E mutant reduced the number of filopodia by ∼2.5-fold and had no effect on the length of remaining filopodia (). The dominant-negative effect of S39E was also observed in 3T3 cells, where S39E expression significantly reduced the number of filopodia from 4.3 ± 2.9 to 1.9 ± 1.9 ( = 31; P < 0.0002). 3T3 cells expressing S39A mutant had a similar frequency of filopodia as control cells (3.9 ± 5.1; = 30). The structural organization of filopodia in B16F1 cells expressing the fascin mutants was analyzed by EM. Similar to normal ones (), filopodia induced by the S39A mutant were straight, orthogonal to the edge, and composed of tightly bundled actin filaments. However, these bundles were typically thinner than in control cells (), correlating with the lower fluorescence intensity of phalloidin staining in S39A-induced filopodia (). In S39E-transfected cells, filopodial bundles had normal thickness, but were often loosely bundled, especially farther away from the tip (, right). In contrast to siRNA-treated cells, the length of the internal parts of filopodial bundles (2.9 ± 1.1 μm; = 7 cells, 33 filopodia) was not significantly different from control cells (3.0 ± 1.7 μm; = 8 cells, 49 filopodia), which is consistent with light microscopic measurements. However, the surrounding lamellipodia frequently contained multiple series of long actin filaments that converged to a common tip at the leading edge (, bottom right). Such structures were similar to Λ-precursors, which are the intermediates during filopodia initiation (), except that they were about fourfold larger than normal Λ-precursors (0.48 ± 0.66 μm versus 0.10 ± 0.17 μm in control cells, P < 0.001; = 28 and = 56 in 5 cells each). These results suggest that the S39E fascin mutant acted as a dominant negative by inhibiting filopodia formation at the stage of initiation of bundling. In live B16F1 cells, WT and active S39A fascins essentially colocalized with actin in filopodia (), except for the proximal regions of filopodial bundles, where the fascin/actin fluorescence ratio was slightly lower, as previously reported (). In contrast, S39E fascin was present only in the distal (∼65%) regions of filopodial bundles visualized by phalloidin staining. Approximately 38% of filopodia also displayed prominent enrichment of S39E fascin at filopodial tips (). The dynamics of fascins were analyzed by three different approaches. FRAP of GFP-fascin was used to analyze fascin turnover in filopodia. = 9 ± 6 s ( = 41; ). Furthermore, 96 ± 37% of fluorescence was recovered, indicating that the majority of fascin population in filopodia was dynamic. In contrast, FRAP of GFP-actin in B16F1 cells resulted in no detectable recovery (). Instead, bleached zones displayed retrograde flow, as shown previously (). Next, we examined whether fascin dynamics depends on Ser39 phosphorylation. FRAP of GFP-S39A fascin (Fig. S4 A, available at ) was similar to that of GFP-WT fascin both for the half-life time ( = 6 ± 4 s) and the final recovery level (101 ± 31%; =39), indicating that active dephosphorylated fascin also undergoes rapid association/dissociation cycles. The similar fast dynamics of WT and S39A fascins were also observed in mouse neuroblastoma Neuro2A cell line (unpublished data). We tried FRAP of GFP-S39E fascin in B16F1 cells, but a small number of filopodia remained in S39E-expressing cells, and their lateral movement, rather than persistent extension, precluded obtaining enough data for meaningful curve fitting. As an alternative approach to evaluate the dynamics of fascins, we monitored the dissociation of GFP-fascins from filopodia after cell lysis during incubation in the stabilization buffer. Cell lysis almost immediately removed cytoplasmic fluorescence of all GFP-fascins, suggesting that this pool represented soluble and weakly bound proteins. Filopodia-associated WT and S39A fascins dissociated slowly, with half-life times of 62 ± 40 min and 25 ± 6 min, respectively (). Faster dissociation of S39A may be related to lower thickness of S39A-induced filopodia, where internal fascin molecules can reach the surface more rapidly. In contrast, S39E fascin was lost within 1 min upon cell lysis. Importantly, the brighter tip fluorescence seemed to dissociate more quickly than the dimmer shaft fluorescence, such that the fluorescence profile of filopodia at intermediate stages of extraction lost the characteristic peak at the tip (). These results show that S39E fascin is very weakly bound to actin bundles and that its enrichment at the tips may be independent of actin. Next, the dynamics of mutant fascins, on the background of the silenced endogenous fascin, were analyzed in living cells during filopodia initiation. Refractory to silencing YFP-fascins (S39A* and S39E*) were coexpressed with CFP-Tc plasmid (). In these conditions, the active YFP-S39A* fascin enhanced filopodia formation (7.1 ± 1.4 filopodia per 20 μm of cell perimeter; = 18), as in normal cells (). Coexpression of the inactive YFP-S39E* and Tc shRNA did not exhibit additive inhibition of filopodia formation (1.0 ± 0.9; = 33) as compared with YFP-S39E alone (). Time-lapse imaging showed that at the initial stages of filopodia formation (), WT, S39E, and S39A fascins similarly appeared as dots at the leading edges, suggesting that recruitment of fascin to the tips is not critically dependent on the phosphorylation state of fascin. In contrast to WT and S39A that are continuously incorporated into the shaft of the elongating filopodium, S39E fascin seemed to dissociate from the filopodia soon after it appeared at the tip. Consistent with fast dissociation of S39E mutant from filopodia in lysed cells, the lifetime of S39E in filopodia in live cells was very short, estimated as a few tens of seconds, whereas WT and S39A filopodia lived for minutes. Thus, incorporation of fascin into filopodial shafts is likely dependent on Ser39 dephosphorylation. #text The pEGFP-fascin construct was provided by J. Adams (Cleveland Clinic Foundation, Cleveland, OH). A site-directed mutagenesis kit (QuikChangeII; Stratagene) was used to create point mutations. YFP/CFP/mCherry-fascin constructs (WT and mutants) were obtained by transferring the fascin cDNA from pEGFP-fascin to pEYFP/pECFP-C1 (CLONTECH Laboratories, Inc.) and pmCherry-C1 with BsrGI–BamHI sites. pmCherry-C1 was constructed by replacing EGFP cDNA in pEGFP-C1 with a PCR-amplified mCherry sequence. The original mCherry construct was provided by R. Tsien (University of California, San Diego, CA). An expression construct of EGFP–α-actinin was obtained from C. Otey (University of North Carolina, Chapel Hill, NC). The α-actinin–coding sequence was transferred to HindIII gaps of pEYFP/pECFP-N1 to prepare pEYFP/pECFP–α-actinin. pEGFP–T-fimbrin and pEGFP-espin2B were provided by J. Bartles (Northwestern University Medical School, Chicago, IL). pG-SUPER was previously described elsewhere (). pG-SUPER-Fascin-Th, -Tm, and -Tc were constructed as previously described (). The selected target sequences were as follows: for Th, nt 741–759 of human fascin1 (NM_003088); for Tm, nt 741–759 of mouse fascin1 (NM_007984); and for Tc, nt 393–411 of human/mouse fascin1 (conserved). The selected sequences did not have significant homology to any other known genes in the mouse database, as determined using the Basic Local Alignment Search Tool (National Center for Biotechnology Information). Th and Tm had two base mismatches, in that Th served as a negative control of Tm (). pC-SUPER-Fascin-Tm and -Tc were constructed by replacing EGFP cDNA of pG-SUPER with ECFP cDNA from pECFP-N1 (CLONTECH Laboratories, Inc.) using SacII–NotI. The rescue construct, pEYFP-fascin* contained three silent point mutations in two codons, S133S (AGC–TCC) and A137A (GCC–GCT), and was refractory to Tc-siRNA. Light microscopy was performed using a microscope (Eclipse TE2000; Nikon) equipped with a Plan 100×, 1.3 NA, objective (Nikon) and a back-illuminated cooled charge-coupled device camera (model CH250; Roper Scientific). The MetaMorph Imaging software (Universal Imaging Corp.) was used for image acquisition and analysis. For live cell imaging, cells were kept in a CO-independent culture medium (L-15; Invitrogen) at 37°C. For actin staining of FACS-purified cells, the procedures are described elsewhere (). For actin staining of cells expressing GFP fusion proteins, cells were fixed with 4% formaldehyde for 20 min, extracted with 1% Triton X-100 for 2 min, and stained with 33 nM Texas red–X phalloidin (Invitrogen) for 10–30 min. For T-fimbrin staining, cells were fixed with 4% formaldehyde, extracted with 1% Triton X-100, and reacted to rabbit anti-fimbrin antibodies (a gift from M. Arpain, Institut Curie, Paris, France). For simultaneous fascin and actin staining, the formaldehyde-fixed samples were stained for F-actin with Texas red–X phalloidin and the first set of images was taken. The specimens were treated with methanol at –20°C for 5 min, incubated by mouse monoclonal anti-fascin antibody (clone 55K-2; DakoCytomation) for 30 min, and subsequently reacted with Cy5-conjugated anti–mouse IgG (Jackson ImmunoResearch Laboratories). The second set of images was aligned with the first set of actin staining by locators on etched grid coverslips (Bellco Glass, Inc.). Secondary antibodies were obtained from Jackson ImmunoResearch Laboratories and Invitrogen. Transfectants of pG-SUPER-fascin and pEGFP-fascin were collected by FACS. The procedures of platinum replica EM were previously described (, ). Quantification of the projected area of Λ-precursors was performed on scanned negatives by measuring the base and the height of triangles approximating the shape of Λ-precursors. Cells expressing EGFP-fascin were incubated on a microscope stage with the extraction solution (1% Triton X-100, 4% polyethylene glycol [40,000 kD], 100 mM PIPES, pH 6.9, 1 mM MgCl, and 1 mM EGTA) supplemented with 2 μM Texas red–X phalloidin. GFP images were taken immediately before and after addition of the extraction solution (within 15–30 s), and then at 5-min intervals. 10–20 cells were analyzed for each group (WT, S39A, or S39E). The total GFP intensity of each filopodium was measured at every time point. After subtraction of the background level, the relative fluorescence intensities were calculated by normalization to the initial intensity in the first image after adding the extraction solution. ), where k is the dissociation rate constant, is time, and I is the initial fluorescence immediately after extraction. was calculated as = 2/k. Transfectants of pG-SUPER-fascin and pEGFP-fascin were collected by FACS. Spread cells containing prominent lamella with convex-shaped leading edges were chosen from phalloidin-stained samples. The contour lengths of the lamellipodia were measured by the MetaMorph Imaging software. The number of filopodial bundles touching or crossing the edge was counted. Only bundles having fluorescence intensity of at least 1.2 times above the background were considered. For each filopodium, the lengths of the internal and protruding parts were also measured. Two researchers performed quantification independently. Statistical analysis was done using SigmaPlot software (Systat Software, Inc.). 50–150 cells were quantified for each sample. Analysis of variance, the Holm-Sidak method, was used as a significance test, with P ≤ 0.001 and an overall significance level of 0.05. For the rescue experiments, the number of filopodia analyzed was from cells expressing YFP-fascin. 12 h after transfection with pG-SUPER-Tm, -Th, or -Tc, GFP-positive cells were collected by FACS and cultured for an additional 3–4 d in the culture medium. The detailed procedures of immunoblotting were previously described elsewhere (). The protein samples (40 μg per lane) were assayed with mouse anti-fascin (clone 55K-2). The blots were analyzed with Image 1.6 software (National Institutes of Health). The linearity of the signals was confirmed by dilution series of purified fascin. FRAP experiments were performed on a confocal microscope (LSM510; Carl Zeiss MicroImaging, Inc.) with a 110×, 1.3 NA, PlanApochromat oil objective (Carl Zeiss MicroImaging, Inc.). Cells were maintained at 37°C with a heated stage. GFP-tagged proteins in the midregions of protruding filopodia were bleached with a rectangular region of area ranging from 6 to 11 μm for ∼1 s using the 488-nm laser line at 100% laser power (25 mW). Thereafter, fluorescence recovery within the bleached region was monitored every 0.5–1 s over a period of 30–60 s. For quantification, MetaMorph software was used. The fluorescence recovery was analyzed as follows: the average intensity over the bleached zone at each time of imaging was measured. To calculate the loss of fluorescence attributed to photofading during image acquisition, the fluorescence intensity of a nonbleached protruding filopodium was determined over time. ). ), at each time point in the experiment. ) was used to curve fit the corrected recovery intensities, where I() = intensity at a given time (), I = intensity at final time, I = intensity at photobleaching event, and k = fluorescence recovery rate constant. Half recovery time was calculated as 2/k. Final recovery was calculated by dividing I by the average intensity over the flanking regions next to the bleached zone. Examples of fluorescence recovery curves are presented in Fig. S4 B, with normalization to the flanking regions. RNA was purified from B16F1 cells using RNeasy kit (QIAGEN) and analyzed to Affymetrix MOE430A chips in triplicate. The microarray data can be viewed in the National Center for Biotechnology Information's Gene Expression Omnibus () under the accession no. . Fig. S1 shows the localization of actin cross-linking proteins in B16F1 melanoma cells. Fig. S2 shows fascin depletion by shRNA in HeLa cells. Fig. S3 shows actin-bundling activity of fascin WT and mutant proteins. Fig. S4 shows FRAP analysis of GFP-fascin. Online supplemental material is available at .

Cell migration is a key component of development, wound healing, and immunological responses (). It also plays essential roles in pathological processes, e.g., invasion of primary tumor cells and subsequent metastasis to distant sites (). Elucidation of the molecular mechanisms involved in the initiation and control of cell migration is therefore essential to understanding these processes (). Migrating cells extend protrusions in the forward direction, form new attachments to the extracellular matrix, and release attachments at the rear to allow the cell mass to pull forward (). The newly formed attachments at the cell anterior lead to adhesion-induced activation of Rac1 that causes localized actin polymerization to create broad protrusive structures (lamellipodia; ). When detached cells adhere, adhesion-induced Rac activation leads to unpolarized lamellipodial extension, resulting in cell spreading (). Thus, cell spreading and lamellipodial extension are closely related Rac-mediated events. The Ras family GTPase Related-Ras (R-Ras) can regulate adhesion-mediated Rac activation and cell migration (; Wozniak et al., 2005). R-Ras shares 55% sequence homology with related paralogues in the Ras family of small GTPases, has an almost identical effector-binding region to H-, N-, and K-Ras (; ), and couples to common Ras effectors, including Raf1, RalGDS, RapL/NORE1, and PI3-kinase (). However, R-Ras has distinct cellular functions from other Ras paralogues. In addition to its distinct effects on integrin activation, R-Ras inhibits cell proliferation in endothelial and smooth muscle cells () and promotes cell adhesion (; ), cell spreading, haptotactic migration (; ), and neurite outgrowth (; ). R-Ras mediates these effects on cells through coupling to downstream effectors; however, to date, no bona fide R-Ras effectors have been described to account for its functions distinct from H-Ras and Rap1. A hindrance to the identification of effectors for small GTPases is the absence of isoprenylation or correct subcellular targeting of Ras GTPases in commonly used methods, such as yeast two-hybrid screens (). The subcellular localization of these proteins is essential for correct targeting to specific effectors and for resulting biological functions (). To facilitate the hunt for isotype-specific effectors of Ras GTPases, we have used tandem affinity purification (TAP) tags (; ) and high-throughput mass spectroscopy to isolate and characterize proteins that interact with posttranslationally modified Ras GTPases in murine fibroblasts. The results of these studies permitted us to develop and analyze a comparative proteomic database of Ras-interacting proteins that is publicly available (). Using this database, we have identified RLIP76 (RalBP1; ; ) as a novel R-Ras effector. We show that RLIP76 directly binds R-Ras in a GTP-dependent manner, but does not interact with the closely related Ras isotypes H-Ras or Rap1A, confirming that RLIP76 is an authentic effector with relative R-Ras specificity. Furthermore, RLIP76 mediates the effect of R-Ras on cell spreading and migration by functioning as a key link in a cascade of small GTPases in which RLIP76 binding to R-Ras capacitates adhesion-mediated activation of Arf6 GTPase. Activated Arf6 GTPase is then required for adhesion-induced Rac activation and the resulting lamellipodia and cell migration. To identify potential effectors of R-Ras, we exploited a comparative proteomic effort () to isolate and characterize Ras-binding proteins. In brief, this database used a modified TAP scheme originally described for yeast proteomics by for a mammalian system (). To do this, we fused the TAP tag-coding region to the N terminus of human Ras cDNAs containing point mutations to render them constitutively active (R-Ras[G38V]) and dominant negative (R-Ras[T43N]); in the case of R-Ras, we used two effector loop mutants, R-Ras(G38VD64A) and R-Ras(G38VD64E), built on the activated background. To assess specificity amongst Ras family members, similar constructs were prepared for activated variants of H-Ras(G12V) or Rap1A(G12V). The Ras constructs were expressed in murine 3T3 cells and purified, and associated proteins were identified by mass spectroscopy (). This unbiased proteomic screen identified RLIP76 (RalBP1) as a protein interacting with activated R-Ras(G38V), but not with dominant-negative R-Ras(T43N). Furthermore, RLIP76 was not identified amongst the proteins co-isolating with activated variants of H-Ras(G12V) or Rap1A(G12V) (), suggesting RLIP76 might be involved in R-Ras–specific signaling. To confirm the specificity of RLIP76 interactions with R-Ras, we coexpressed authentic human RLIP76 cDNA with R-Ras variants in murine 3T3 cells. RLIP76 coprecipitated with activated R-Ras(G38V), but not R-Ras(T43N; ), demonstrating the GTP dependence of the interaction. Furthermore, activated variants of H-Ras(G12V) and Rap1A(G12V) did not interact with RLIP76, establishing the specificity of the interaction amongst Ras family members and confirming a previous report of the failure of RLIP76 to bind H-Ras in a yeast two-hybrid experiment (; ). In contrast to H-Ras, activated R-Ras promotes integrin activation (; ) and cell spreading (; ). Mutations in the switch 1 effector-binding region of R-Ras can selectively disrupt its ability to interact with known effectors and can alter its ability to increase integrin activation (). We examined the effect of such switch 1 mutants of R-Ras on integrin-mediated cell spreading. In confirmation of previous work (), transfection of NIH 3T3 mouse fibroblasts with constitutively active R-Ras(G38V), but not R-Ras(T43N; dominant negative), stimulated cell spreading (). The effector loop mutant R-Ras(G38VD64A), which fails to bind or activate PI3-kinase and RalGDS but is competent to activate integrins (), promoted cell spreading, whereas R-Ras(G38VD64E) failed to do so. Thus, the increase in spreading caused by R-Ras does not require direct activation of PI3-kinase or Ral. In TAP-Ras pulldown experiments we isolated RLIP76 with R-Ras variants that support cell spreading, i.e., R-Ras(G38V) and R-Ras(G38VD64A), but we did not detect RLIP76 in proteins isolated with R-Ras(T43N) or R-Ras(G38VD64E; unpublished data). To directly assess whether RLIP76 was a candidate effector for this R-Ras function, we examined its ability to coprecipitate with each of these variants. RLIP76 coprecipitated with R-Ras(G38V) and the G38VD64A mutant, but not with the dominant-negative T43N variant or the G38VD64E double mutant (). Thus, RLIP76 interacts with R-Ras variants that support cell spreading and integrin activation and is a candidate to mediate these activities of R-Ras. The specificity of the interactions of RLIP76 with the R-Ras variants that promote spreading and integrin activation suggested that RLIP76 may play a role in these processes. We used double-stranded siRNA () to inhibit RLIP76 expression in mouse 3T3 cells. This siRNA reduced the expression of endogenous RLIP76 by >90% in mouse 3T3 cells and CHO cells and, as previously reported (), also suppressed the expression of human RLIP76 ( and unpublished data). To assess a role for RLIP76 in cell spreading, 3T3 cells were transfected with the RLIP76 siRNA, plated onto fibronectin-coated coverslips, and evaluated for spreading. The cells transfected with the RLIP76 siRNA showed markedly reduced cell spreading, and this effect could not be reversed by cotransfection of R-Ras(G38V; ). In contrast, reconstitution of RLIP76 expression with a cDNA encoding three silent mismatches (RLIP76m) in the siRNA target sequence, restored RLIP76 expression and cell spreading (). R-Ras can regulate integrins' affinity (Zhang et. al. 1996), but siRNA-mediated knockdown of RLIP76 had no effect on the activation state of integrins in CHO cells (; unpublished data). Furthermore, R-Ras can reverse the suppressive effect of H-Ras on integrin activation (); however, RLIP76 siRNA had no effect on this activity of R-Ras (unpublished data). Thus, RLIP76 is required for cell spreading and for the increased spreading mediated by activated R-Ras, but did not modulate integrin affinity in CHO cells. The requirement of RLIP76 expression for spreading suggested that RLIP76 may be downstream of signals from R-Ras in cell spreading. As a test of this hypothesis, we expressed dominant-negative R-Ras(T43N) and found that it inhibited cell spreading ( and ) and the defect in cell spreading induced by R-Ras(T43N) was reversed by overexpression of RLIP76 (). Thus, RLIP76 is required for R-Ras–mediated cell spreading, and RLIP76 overexpression can bypass a requirement for R-Ras activity in cell spreading. These relationships, in combination with the R-Ras effector loop specificities of the physical interaction of R-Ras with RLIP76, demonstrate that RLIP76 is downstream of R-Ras in promoting cell spreading. The association of RLIP76 with activated R-Ras and its role in R-Ras–mediated spreading suggested the possibility that these proteins may interact directly and that RLIP76 might be an authentic R-Ras effector. In prokaryotic expression systems, we were unable to obtain a sufficient quantity of purified full-length RLIP76 protein to analyze its interaction with R-Ras because of low expression levels (unpublished data), and therefore used coimmunoprecipitation studies to initially map the RLIP76-binding domain for R-Ras. The RLIP76 polypeptide consists of an N-terminal coiled coil region, a Rho GTPase-activating protein (RhoGAP) domain, a short connecting segment, a Ral-binding domain (RBD), and a C-terminal region (; ; ; ). We constructed N-terminal RLIP76 lacking the RBD and C-terminal region (ΔRBD) or that were further truncated at the C-terminal half of the GAP domain (ΔGAPn; ). In coimmunoprecipitation experiments, the full-length RLIP76 and the RBD-deleted protein (ΔRBD) coprecipitated with activated R-Ras, whereas the ΔGAPn truncated protein did not (). These data suggested that the RBD, responsible for direct interaction of RLIP76 with Ral GTPase, (), is not required for RLIP76 interaction with R-Ras. To test whether RLIP76 can interact directly with R-Ras, we generated N-terminal poly-histidine–tagged bacterial expression constructs encoding RLIP76ΔRBD or -ΔGAPn; both were well expressed. An N-terminal glutathione--transferase fusion to R-Ras(wt) (GST–R-Ras) protein was coupled to GSH beads and either loaded with 10 μM GTP to generate activated R-Ras or maintained in 10 mM EDTA to produce nucleotide-free R-Ras. Western blotting showed that the ΔRBD fragment protein bound directly to R-Ras in a GTP-dependent manner. In contrast, the ΔGAPn RLIP76 fragment did not detectably interact with R-Ras in either state (). Thus, RLIP76 binds directly to R-Ras, in a GTP-dependent manner, and the RLIP76 RhoGAP domain, but not to the RBD, is required for this interaction. We next evaluated the abilities of the truncated RLIP76 mutants to reconstitute spreading in RLIP76-depleted cells. The RLIP76 mutants were constructed on the siRNA mismatch RLIP76 mutant background and cotransfected with RLIP76 siRNA. RLIP76mΔRBD, which binds R-Ras, rescued the siRNA-induced spreading defect to a similar extent to the full-length construct (). The ΔGAPn truncation mutant, which did not bind R-Ras (), did not reverse the effect of RLIP76 siRNA. Thus, R-Ras binds the RhoGAP, but not the RBD of RLIP76 in a GTP-dependent manner, and the RhoGAP domain is required for RLIP76-dependent cell spreading. Cell spreading can be induced by the small GTPase Rac1 (). The RhoGAP domain of RLIP76 has Rac GAP activity in vitro (; ; ). The RLIP76 RhoGAP domain bears strong sequence similarity to the RhoGAP domains of BCR-GAP and n-chimaerin, which are both GAPs for Rac (; ). Despite sequence divergence among GAP proteins, their mechanism of action is highly conserved, and depends on an “arginine finger,” which is a conserved arginine residue in a loop segment that stabilizes the GTP moiety on the substrate, increasing the substrate's intrinsic GTPase activity (). RLIP76, bearing a mutation in the conserved Arg required for GAP function (RLIP76m[R232D]), still supported cell spreading in RLIP76-depleted cells, indicating that GAP activity of RLIP76 is not required for its effects on cell spreading (). Activated R-Ras stimulates cell spreading and migration by augmenting the activation of the small GTPase Rac1 (), although this effect appears to be cell type specific (Wozniak et al., 2005). In murine 3T3 cells, as reported by , transfection of R-Ras(G38V) resulted in an approximately threefold increase in adhesion-induced activation of Rac1, as measured by its binding to the p21-binding domain of PAK1 (; ; ). Cotransfection of RLIP76 siRNA completely blocked this effect of R-Ras(G38V); this decrease could be reversed by reexpression of RLIP76m (). Thus, RLIP76 expression is required for R-Ras–induced Rac activation. As noted above, the capacity of dominant-negative (T43N) R-Ras to inhibit spreading and Rac activation suggests that endogenous R-Ras mediates these processes. To determine whether RLIP76 is required for adhesion-dependent activation of Rac, we assessed the effect of siRNA-mediated RLIP76 knockdown on adhesion-induced Rac activation in the absence of exogenous R-Ras. In control siRNA-transfected cells, Rac activation was low in cells maintained in suspension, and increased dramatically after five minutes of adhesion to fibronectin (). This increase in Rac activation was followed by a diminution over the next 10 min. In contrast, cells with reduced RLIP76 expression failed to increase Rac activation in response to cell adhesion. This defect in Rac activation was reversed by expression of mismatched RLIP76 (). Defective Rac activation in RLIP76-deficient cells was also partially restored by reconstitution with RLIP76ΔRBD (), which also restored spreading (). The spreading defect induced by RLIP76 knockdown was reversed by expression of an activated variant of Rac (Rac1A[G12V]; ). Furthermore, overexpression of RLIP76 failed to complement the spreading defect induced by dominant-negative Rac(T17N). Thus, Rac1 activation is downstream of RLIP76. The requirement for RLIP76 is specific to adhesion-induced Rac activation because RLIP76 depletion had little effect on Rac activation stimulated by EGF (). Thus, RLIP76 mediates cell spreading by regulating adhesion-induced activation of Rac. Rac1 activation is associated with its localization to the plasma membrane and concentration in nascent lamellipodia (; ). Depletion of RLIP76 led to diminished lamellipodial extensions and loss of Rac localization at the peripheral membrane. This phenotype was rescued by reexpression of RLIP76m (). Furthermore, depletion of RLIP76 blocked lamellipodia and Rac membrane localization even in the presence of activated R-Ras(G38V) (). In contrast, RLIP76 overexpression rescued the loss of lamellipodia and peripheral Rac localization in R-Ras(T43N)–transfected cells (). Thus, RLIP76 regulates the localization and the activation of Rac GTPase. The requirement for RLIP76 in adhesion-induced Rac activation suggested that RLIP76 is important for cell migration. To test this idea, we wounded monolayers of GFP-transfected murine 3T3 cells and measured the migration of GFP-positive cells into the wound. Cells cotransfected with RLIP76 siRNA exhibited markedly impaired migration relative to control vector-transfected cells or cells cotransfected with the rescue construct RLIP76m in combination with the siRNA (). Thus, RLIP76 is required for directional cell migration into a wound. Arf6 GTPase, which is a class III ADP-ribosylation factor, regulates endosomal trafficking of Rac1 (; ) and mediates adhesion-dependent DOCK180/Elmo-induced Rac activation (). Therefore, we asked if RLIP76 effects on Rac localization and activation are mediated by Arf6. Cell adhesion activates Arf6 and we found that siRNA-mediated knockdown of RLIP76 blocked adhesion-induced Arf6 activation (). We had previously observed () that RLIP76 overexpression can bypass a requirement for R-Ras activity in cell spreading, suggesting that RLIP76 might render adhesion-induced Arf6 activation independent of R-Ras. Dominant-negative R-Ras(T43N) inhibited adhesion-induced activation of Arf6 (); however, ectopic RLIP76 restored Arf6 activation in R-Ras(T43N)–expressing cells (). We next evaluated the contribution of Arf-GEFs to RLIP76-dependent cell spreading. Transfection of ARF nucleotide-binding site opener (ARNO), which is a Sec7 domain–containing GEF for Arf6 (; ; ), overcame the effect of siRNA knockdown of RLIP76 on cell spreading (), and this rescue could be fully reversed by expression of a dominant-negative Arf6(T27N) and partially reversed by Arf1(T31N) (). Moreover, transfection of either dominant-negative Arf1 or Arf6 alone inhibited cell spreading, indicating that sequestration of Arf-GEFs was sufficient to block cell spreading. Furthermore, Arf6(T27N) blocked lamellipodia formation and peripheral localization of Rac1 (). These two results suggested that RLIP76 might interact with an Arf-GEF; indeed, we found that RLIP76 and ARNO physically associate in vivo (). Thus, RLIP76 links R-Ras to adhesion-induced Arf6 activation with consequent enhancement of the activation of Rac1 GTPase. Therefore, RLIP76 is a critical R-Ras effector in a cascade of GTPases leading to cell spreading and migration (). R-Ras has pleiotropic cellular effects, enhancing cell spreading, migration, and neurite outgrowth, and inhibiting cell proliferation; frequently, R-Ras effects differ from those of closely related paralogues such as H-Ras and Rap1A (; ; ; ). We used a recently developed experimental database of Ras GTPase–interacting proteins to identify RLIP76 (RalBP1) as an effector protein for R-Ras, but not for the closely related paralogues H-Ras and Rap1. We find that RLIP76 binds directly to R-Ras in a GTP-dependent manner and that the R-Ras–binding site is distinct from that of Ral. Furthermore, RLIP76 is required for adhesion-mediated Rac activation and cell spreading and migration, and for the stimulation of these processes by R-Ras. Arf6 GTPase is a mediator of adhesion-induced Rac activation, and RLIP76 is needed for adhesion-induced activation of Arf6. Reconstitution of Arf6 activity by transfection with an Arf6–guanine nucleotide exchange factor (GEF), ARNO restored Rac activation and cell spreading in RLIP76-depleted cells. Thus, we identify a new R-Ras effector, RLIP76, which is a critical link in a cascade of three GTPases that regulates cell spreading and migration. The proteomic strategy used in this study identified RLIP76 as a candidate R-Ras effector, and subsequent studies established that it is an authentic effector that mediates the activities of R-Ras in cell spreading. In particular, RLIP76 binding to R-Ras was direct and specific for the GTP-bound form of R-Ras. Moreover, we previously identified a switch 1 domain mutation (D64A) that perturbs R-Ras interaction with effectors such as RalGDS, Raf1, NORE1, and PI3-kinase (). This mutation did not block interaction with RLIP76, nor did it block the effect of R-Ras on cell spreading, thus, implicating RLIP76 as a previously proposed (; ) novel R-Ras effector. In contrast, the R-Ras(D64E) mutant was deficient in binding to RLIP76 and did not promote spreading, suggesting that the R-Ras–RLIP76 interaction involves the effector-binding region of R-Ras. Collectively, these data indicate that RLIP76 is an authentic R-Ras effector that mediates R-Ras stimulation of cell spreading; suppression of cell spreading by siRNA-induced RLIP76 depletion confirmed this conclusion. RLIP76 binds R-Ras, but not the closely related Ras proteins H-Ras or Rap-1. This distinguishes RLIP76 from effectors such as Raf-1, RalGDS, PI3-kinase, and NORE1 that bind to multiple Ras family members (; ; ). Recently, showed that PLC interacts with activated R-Ras(G38V) better than wild-type R-Ras, although a direct interaction was not demonstrated. However, PLC can also bind to other Ras isotypes through its Ras-association domain (). It seems likely that our success in identifying the RLIP76 interaction may be attributable to the use of prenylated R-Ras in mammalian cells. Most previous studies have used yeast two-hybrid approaches in which the Ras protein is neither posttranslationally modified nor appropriately localized (). The posttranslational modification and localization of Ras proteins has a critical effect on their interactions and resulting biological functions (); indeed, this principle has been experimentally demonstrated for R-Ras (; ; ). RLIP76 binds the small GTPase, RalA (; ; ; ; ); however, RalA is not involved in R-Ras–RLIP76–mediated cell spreading. An activated R-Ras effector loop mutant that does not bind RalGDS (G38VD64A; ) still promoted cell spreading. In addition, the RBD of RLIP76 was not required for direct binding to R-Ras. This finding is consistent with earlier structure–function studies in which the RLIP76 RBD did not bind to Ras (). Indeed, the RBD is predicted to form a coiled coil, unlike the ubiquitin fold common to RBDs, suggesting that RLIP76 interacts with R-Ras through a structural mechanism that is distinct from the canonical interactions of Ras proteins with effectors (). The RLIP76 RBD was also not required for Rac activation or cell spreading. Thus, RLIP76 is a bona fide R-Ras–specific effector that mediates R-Ras–stimulated cell spreading. RLIP76 promotes cell spreading by mediating adhesion-dependent activation of Rac GTPase. Specifically, adhesion-dependent spreading and Rac activation required RLIP76. Furthermore, activated Rac1(G12V) rescued the spreading defect in RLIP76-depleted cells, and RLIP76 overexpression did not restore spreading in cells expressing dominant-negative Rac1(T17N); hence, RLIP76 is upstream of adhesion-mediated Rac activation. The RhoGAP domain of RLIP76 was important in its ability to mediate adhesion-dependent Rac activation. This domain, which is similar to the GAP domains of Bcr-GAP and n-chimaerin (; ), has been reported to act as a Rac and Cdc42 GAP in vitro (; ; ). Nevertheless, our studies show that in vivo, RLIP76 is required for adhesion-induced activation of Rac, and this function is insensitive to a mutation of the arginine finger of the RhoGAP domain that is predicted to disrupt GAP activity (). Furthermore, forced targeting of RLIP76 to the plasma membrane causes extensive membrane ruffling and protrusive activity, which are Rac-dependent processes (), which is consistent with a positive role in Rac activation. Thus, these data show that RLIP76 is upstream of adhesion-dependent Rac activation that leads to cell spreading. RLIP76 mediates adhesion-dependent Rac activation and cell spreading by regulating Arf6 GTPase. Cell adhesion activated Arf6, which is a small GTPase known to participate in adhesion-mediated Rac activation () and cell spreading. Furthermore, enforced activation of Arf GTPases by an Arf-GEF (ARNO) rescued spreading in RLIP76-depleted cells, establishing that Arf6 activation can bypass the requirement for RLIP76. ARNO activates both Arf1 and Arf6; however, only Arf6 can act at the plasma membrane to promote Rac activation (; ), implicating Arf6 as the major target of ARNO in cell spreading (). However, ARNO is an exchange factor for both Arf1 and Arf6, and we found that dominant-negative Arf6(T27N) and Arf1(T31N) both reduced the ability of ARNO to rescue spreading in RLIP76-depleted cells. Previous studies implicated Arf6 in adhesion-dependent Rac activation (; ), possibly because Arf6 activity is necessary for delivery of Rac to the plasma membrane (; ; ) and Arf6 is implicated in ARNO-mediated recruitment of the Rac-GEF, DOCK180/Elmo (). Furthermore, we found that RLIP76 interacts with ARNO in cells, suggesting a physical as well as a functional connection between RLIP76 and Arf-GEF activity. RLIP76 has been shown to mediate glutathione conjugate transport activity and to be localized to the cell surface (). Our results describe an additional signaling function for RLIP76 in the cytosol. Thus, Arf6 acts between RLIP76 and Rac in adhesion-mediated cell spreading; moreover, the capacity of Arf6 to regulate Rac localization and activity provides a plausible mechanism for its role in RLIP76-dependent cell spreading. In sum, our data establish that RLIP76 is a new R-Ras effector that connects R-Ras to the activation of Arf6 and, consequently, to Rac1 leading to adhesion-dependent cell spreading and migration (). Furthermore, the multiple functions of RLIP76 may account for the pleiotropic effects of R-Ras. For example, the RLIP76-dependent activation of Rac GTPase may contribute to the ability of R-Ras to promote cell migration and neurite outgrowth, which are processes that are known to depend on Rac (; ; ). Also, RLIP76 mediates the adhesion-induced activation of Arf6 GTPase, which is a regulator of vesicle trafficking (); hence, regulation of Arf6 may contribute to the ability of RLIP76 to control endocytosis (). Indeed, RLIP76 physically associates with the μ2 chain of AP-2, which is an adaptor whose membrane recruitment is controlled by Arf6, suggesting that these two proteins can coordinately regulate clathrin-mediated endocytosis (; ). Endocytosis of growth factor receptors can limit their ability to stimulate cell proliferation (); therefore, the interaction of RLIP76 with R-Ras may explain the capacity of R-Ras to limit the proliferation of endothelial and smooth muscle cells (). R-Ras can have complex and differing effects on apoptosis, cell adhesion, and migration, depending on cellular context (; ; ; ; ; ; ; ). The capacity of R-Ras to interact with RLIP76, and thereby regulate Arf6, links R-Ras to the regulation of vesicle trafficking and helps to explain its wide-ranging effects on cell growth, apoptosis, cell adhesion, and cell migration. Monoclonal anti-Rac1 antibody (23A8) was obtained from Millipore. Goat anti–human RalBP1 (RLIP76), rabbit anti–R-Ras, goat anti-MLCK, mouse monoclonal anti-Arf6, and rabbit H-15 anti–His tag antibodies were purchased from Santa Cruz Biotechnology, Inc. Monoclonal anti-HA antibody was obtained from Covance. Restriction endonucleases were obtained from New England Biolabs. NIH 3T3 cells (American Type Culture Collection) were maintained in DME (Cellgro) supplemented with 10% FBS, 2 mM -glutamine, 50 U/ml penicillin, 50 μg/ml streptomycin sulfate, and 1% nonessential amino acids (Sigma-Aldrich) at 37°C in 5% CO. Cells were transfected with plasmids using Lipofectamine reagent (Invitrogen) following the manufacturer's instructions. Cells were analyzed 48 h after transfection. For cotransfection with siRNA after plasmid transfection, the medium was changed to complete DME without antibiotics after plasmid transfection and the cells were cultured for 24 h. Cells were subsequently transfected with siRNA as described below and cultured for an additional 24 h before analysis. For siRNA-mediated RLIP76 knockdown, complementary single-strand RNAs (GUAGAGAGGACCAUGAUGdTdT and ACAUCAUGGUCCUCUCUACdTdT) targeting human and mouse RLIP76 were purchased from Invitrogen and dissolved in TE (10 mM Tris-Cl, pH 7.5, 0.1 mM EDTA). Sequence-scrambled complementary single-strand RNAs were also obtained (GAAGAAGAUCGUCAGUGGdTdT and CCACUGACGAUCUUCUUCdTdT). Double-stranded RNA was generated by annealing the RNAs in annealing buffer (100 mM KOAc, 30 mM Hepes-KOH, pH 7.4, and 2 mM MgOAc). Formation of double-stranded RNA was confirmed by polyacrylamide gel electrophoresis. For siRNA-mediated knockdown of RLIP76, 3T3 cells were maintained overnight in complete DME with 10% FBS, but without antibiotics. Cells were transfected with Lipofectamine 2000 in Opti-MEM (Invitrogen) according to the manufacturer's instructions. Cells were analyzed 24 h after transfection. pEGFP-C1 was obtained from CLONTECH Laboratories, Inc. pEGFP-C1-Rac1 WT was a gift from M. del Pozo (Fundación Centro Nacional de Investigaciones Cardiovasculares, Madrid, Spain). N-terminal TAP-tag fusions of human Ras isotype variants were made by subcloning human cDNAs of R-Ras, H-Ras, and Rap1A containing the indicated mutations () into the N-terminal TAP vector (a gift from B. Séraphin, European Molecular Biology Laboratory, Heidelberg, Germany) using BamH1–Xba1 sites. J. Han provided the nTAP-H-Ras and –Rap1A plasmids. The human RLIP76 cDNA was obtained from the American Type Culture Collection. An oligodeoxyribonucleotide encoding a 5′ influenza HA tag joined to six bases from the 5′ end and another oligodeoxyribonucleotide complementary to the 3′ end of RLIP76 (5′-RLIP76: 5′-GGAGATATCGGCGTCATGTACCCATACGATGTTCCAGATTACGCTCTCGAGATGACTTGCTTCCTGCCCCCCACCAGC-3′ and 3′-RLIP76: 5′-CGCAACCTTGCTCAGATGGACGTCTCCTTCCTATCCCTGCTGGG-3′) were obtained from GenBase. The oligodeoxyribonucleotides were used in a PCR reaction to amplify HA-tagged RLIP76 from the RLIP76 cDNA. The PCR product was sequenced and cloned into pcDNA3.1(−) (Invitrogen) using EcoRV and HindIII sites. HA-tagged RLIP76 truncation and deletion plasmids were generated by PCR from the HA-RLIP761/pcDNA3.1(−) template using the following primer sets: ΔRBD (truncated at amino acid 392): 5′-RLIP76 and 5′-GGCTGATGGATCCTTGATGCCCGCCTAGGTCTCTGGCAGCGTGGG-3′; and ΔGAP-n (truncated at amino acid 292): 5′-RLIP76 and 5′-CGCAAGCTTTTACTGCTTCAGCAAACTGGA-3′. HA-RLIP76/3.1(−) mismatch construct was produced by QuikChange mutagenesis (Stratagene) according to the manufacturer's instructions, using the primers: 5′-GGCTGATGCAGTCGAAGAACTATGATGTATGATGGCATTCGGCTGCCAGCC-3′ and 5′-CCTTAAGGAAACCGACTACGTCAGCTTTCTTGATACTACATACTACCGTAAGCC-3′. HA-RLIP76-RD/3.1(−) Arg232Asp mutant was also produced by QuikChange mutagenesis, using the primers: 5′-GGCATGAAGTGTGAAGGCATCTACGACGTATCAGGAATTAAATCAAAG-3′ and 5′-CCTTGATTTAATTCCTGATACGTCGTAGATGCCTTCACACTTCATGCC-3′. 6-His-tagged RLIP76 constructs were made by Xba1–HindIII digestion of the appropriate constructs in pcDNA3.1(−) and ligation into Xba1–HindIII–digested pRSET(A) vector (Invitrogen). The GST–R-Ras(wt) plasmid was as previously described (). HA-tagged Arf expression constructs were a gift from J. Casanova (University of Virginia, Charlottesville, VA). TAP of TAP–R-Ras protein complexes was performed as described at the National Cell Migration Consortium website (). Solution-based digestion: The TAP-purified protein samples were treated with 10 μl of 200 mM CaCl and 20 μl of 100 mM ammonium bicarbonate. Reduction of the disulfide bonds in the proteins was accomplished by treatment of the samples with 1 μl DTT (20 mM final) for 1 h at 52°C, followed by carboxyamidation by treatment with 1 μl iodoacetamide (40 mM final) for 1 h at RT in the dark. Proteolytic digestion of the reduced and alkylated sample was conducted by addition of 1 μl of modified trypsin (0.5 μg; Promega) and incubating the mixture at 37°C for 9 h. The digestion was quenched by acidification with 2 μl of glacial acetic acid and frozen at −35°C until further analysis. Analysis by mass spectrometry: For the identification of interacting proteins, the digested samples were loaded onto a C18 column. Peptides were analyzed by nanoflow reverse-phase high-performance liquid chromatography microelectrospray tandem mass spectrometry (RP-HPLC/μESI/MS/MS) interfaced with a Finnigan LCQ ion trap mass spectrometer (Thermo Electron Corp.). Peptides were gradient-eluted using a linear gradient of 0–60% B in 120 min (A, 0.1 M acetic acid in NANOpure water; B, 70% acetonitrile in 0.1 M acetic acid). The LTQ spectrometer was operated in a data-dependent top 10 MS/MS mode. The data was then searched against a human, rat, and mouse GenBank protein database compiled by National Center for Biotechnology Information (), using the SEQUEST search algorithm (version 27; ). Peptide sequence assignments were verified by manual interpretation of MS/MS spectra. To produce 6-His-tagged RLIP76 proteins and GST–R-Ras(wt), 1 mM IPTG was added to a suspension containing BL21/pLys(s)-DE3 (Stratagene) bacteria (OD = 0.5) bearing the appropriate plasmid, and the bacteria were incubated with continuous shaking at 37°C for 4 h. Bacteria were lysed by sonication in lysis buffer (GST–R-Ras: 20 mM Tris-Cl, pH 7.0, 150 mM NaCl, 1 mM MgCl, 1% Triton X-100, 1 mM DTT, 5 mg/ml lysozyme, and protease inhibitors; His-tagged proteins: 20 mM Tris-Cl, pH 7.9, 5 mM imidazole, 500 mM NaCl, 0.1% NP-40, 5 mg/ml lysozyme, and protease inhibitors) and centrifuged at 39,000 for 20 min, and the soluble proteins were purified on Ni-coupled Sepharose (His-tagged proteins; Novagen) or GSH-Sepharose (R-Ras; Roche), according to the manufacturer's instructions. GST-PBD protein was generated and purified as in del Pozo et al.. GST-GGA3 was generated and purified as in . Glass coverslips were incubated with 5 μg/ml plasma fibronectin in 0.1 M NaHCO at 4°C overnight; after washing, the coverslips were incubated for 30 min with 1% BSA/PBS that had been heat inactivated by incubation at 80°C for 30 min to block unreacted sites. Cells were detached with 0.1% trypsin and kept in suspension for 1 h at RT in DME containing 0.2% BSA, and then plated on the coated coverslips for 45 min at 37°C. Nonadherent cells were removed by washing two times with PBS, and the adherent cells were fixed in 3.7% formaldehyde/PBS for 20 min at RT. Fixed cells were washed and permeabilized with 0.1% Triton X-100/PBS for 5 min, washed with PBS, and incubated with 5 μg/ml rhodamine-phalloidin (Invitrogen) in PBS at 37°C for 1 h. Coverslips were washed and mounted on slides with ProLong Gold antifade reagent (Invitrogen) and imaged on an epifluorescence microscope (DMLS; Leica) fitted with a SPOT charge-coupled device camera (Diagnostic Instruments). The area of GFP-positive cells was measured using ImageJ software (National Institutes of Health). Cells were transfected and plated on fibronectin-coated dishes as described in the previous section. 48 h after transfection, cell monolayers were scratched with a pipette tip. Cells were maintained in complete medium at 37°C in 6% CO, and images were taken at 30 min intervals for 4 h. Cell centroid movements of GFP-positive cells were tracked in successive images using ImageJ software. Adhesion-dependent Rac activation was measured following the method of del Pozo et al.. Cells were transfected as indicated, and cultured in MEM/0.2% FBS overnight. Cells were detached and maintained in suspension for 1 h at RT. The cells were plated on prewarmed, fibronectin-coated dishes, incubated at 37°C with 6% CO for the indicated times, rinsed briefly with ice-cold PBS, and scraped into Rac assay lysis buffer (50 mM Tris-Cl, pH 7.0, 0.5% NP-40, 500 mM NaCl, 1 mM MgCl, 1 mM EGTA, 100 mM NaVO, and protease inhibitors) containing 5 μg/ sample GST-PBD protein. Cell lysates were cleared of insoluble material by centrifugation and incubated with GSH–Sepharose for 30 min. The Sepharose beads were pelleted, washed three times with ice-cold lysis buffer (without GST-PBD), and resuspended in SDS gel-loading buffer. Bead-bound proteins were extracted by boiling, and the protein samples were separated by SDS-PAGE. Proteins were transferred to nitrocellulose membrane, and Rac was detected by Western blotting with 23A8 anti-Rac antibody (Millipore). Cells were detached and kept in suspension for 1 h in serum-free DME containing 0.2% BSA. Cells were then plated on dishes coated with 5 μg/ml fibronectin. After 5 min, cells were lysed and GTP-bound Arf6 was assayed by its binding to a GST-fusion protein containing the VHS domain to the GAT region of an Arf effector, GGA3 (Golgi-localized, γ-ear-containing Arf-binding protein 3; GST-GGA3; ), as previously described ().

Integrins are heterodimeric cell surface receptors, consisting of an α and a β subunit, which play an important role in cell migration and tissue integrity by mediating cell attachment to the surrounding ECM and to other cells (). Integrins can adopt high- and low-affinity conformations, and ligand binding to integrins is preceded by intracellular changes, resulting in increased integrin affinity (inside-out signaling; ). Tight regulation of integrin affinity is crucial for the physiological function of integrins. During inflammation, for example, leukocytes can only extravasate into the affected tissue after a chemokine-induced increase in integrin affinity (). Similarly, blood clotting is restricted to wounds by efficient control of the affinity of platelet integrin (). Although the molecular mechanisms controlling integrin-mediated adhesion by inside-out signaling are not well understood, much is known about the structural changes that occur during integrin activation (). Crystal structure analysis suggests a bent, hooklike conformation of the extracellular domain for the inactive state and an extended conformation in the active, high-affinity state (). Recent three-dimensional EM structural analysis, however, demonstrates that the bent conformation of αvβ3 is also able to bind efficiently to ligands, at least in the presence of Mn, indicating that Mn-induced activation takes place through small local conformational changes and not by straightening the extracellular domain (). Although all these studies were done with β3 integrins, the conclusions conceivably can be extended to β1 integrin because of the high extent of structural similarity. Many lines of experimental evidence indicate that binding of talin to the cytoplasmic domain of β1 integrin is a final common step in integrin activation (; ). Talin is a rodlike molecule with a globular head domain, which links integrins to the actin cytoskeleton. It binds to the membrane-proximal NPXY motif of β1, β2, and β3 integrin via a phosphotyrosine binding–like region in the FERM domain of the talin head. This binding is suggested to disrupt a salt bridge between the α and β subunits, which is believed to keep the integrin in an inactive state (). In addition, specific van der Waals interactions between the transmembrane regions of the α and β subunits are supposedly altered after talin binding, leading to conformational changes that are propagated across the plasma membrane, resulting in integrin activation (). Talin binding to integrin is promoted by proteolytic cleavage of talin and binding of phosphoinositol phosphates and may be regulated by phosphorylation of talin (). It was proposed that phosphorylation of the NPXY motif in the integrin β subunit interferes with integrin activation by talin in two different ways. First, tyrosine phosphorylation of the NPXY motif might reduce the affinity for talin to such an extent that talin can no longer bind the β subunit. Second, NPXY phosphorylation might increase the affinity for other phosphotyrosine binding domain–containing proteins, to competitively inhibit the interaction with talin (). Indeed, tyrosine-phosphorylated β3 integrin preferably interacts with Shc and not with talin (). Mutational studies support a negative regulation function of tyrosine phosphorylation on integrin activation. Fibroblastoid cells expressing a β1 integrin in which the tyrosines in both NPXY motifs were substituted with nonphosphorylatable phenylalanines (YY783/795FF) were able to more effectively bind and assemble fibronectin (FN), similar to β1 mutants in which formation of the salt bridge to the α subunit is prohibited (D759A; ). In addition, YY783/795FF mutants showed partially altered outside-in signaling characterized by defective FAK activation but normal p130Cas phosphorylation (). Transformation with v-src results in tyrosine phosphorylation of β1 integrin, correlating with decreased β1 integrin–mediated adhesion (). V-src did not induce decreased adhesion in the nonphosphorylatable YY783/795FF β1 mutant, suggesting that tyrosine phosphorylation of integrins inhibits integrin activation and contributes to cell transformation by v-src. In this study, we investigated the mechanism of integrin activation under physiological conditions and generated mice in which we replaced endogenous β1 integrin by mutant forms of β1, lacking either the aspartic acid essential for the salt bridge with the α subunit (D759A) or the tyrosine residues of the two cytoplasmic NPXY motifs (YY783/795FF), thus preventing inactivation by tyrosine phosphorylation. Surprisingly, both mutations did not cause any obvious phenotype, indicating that neither the salt bridge nor the phosphorylation of the intracellular tyrosines of β1 integrin is important for the regulation of β1 integrin activation. Substitution of the tyrosines with alanines (YY783/795AA), however, completely abolished β1 integrin function, confirming the crucial role of the NPXY motifs for integrin activation. To study the role of amino acid residues and protein motifs important for integrin activation during mouse development as well as in adult tissues, we established several mouse strains with point mutations in the cytoplasmic domain of the β1A integrin subunit (). First, we tested the relevance of the membrane-proximal salt bridge between α and β1 subunits by exchanging the aspartic acid residue 759 to alanine (D759A [D/A]), which destroys the salt bridge. Second, we addressed the potential role of tyrosine phosphorylation of the two conserved NPXY motifs within the β1 cytoplasmic domain, the proximal of which was shown to recruit talin. To this end, we established mice in which one or both tyrosines were exchanged to the nonphosphorylatable phenylalanine (Y783F; Y795F; YY783/795FF [YY/FF]). A nonconservative change of the NPXY motifs was achieved by replacing the tyrosines by alanine (YY783/795AA [YY/AA]). All mutations were introduced by homologous recombination into embryonic stem (ES) cells () and confirmed by sequence analysis (not depicted). Positive ES cell clones of each mutation were injected into blastocysts to establish the mutant knock-in (KI) mouse strains (KI; ). The loxP-flanked neomycin gene was then removed by intercrossing the mutant mouse strains with a deleter-Cre strain, resulting in mice carrying an intronic loxP site and defined point mutations in the following exon of the β1 integrin gene (KI; ). To exclude adverse effects of the intronic loxP sites, we established mice carrying only a loxP site in front of exon 15 or 16 but no mutation in the exons (E15 and E16). Homozygous E15 and E16 animals were born at the expected Mendelian ratio, were fertile, had a normal life span, and did not display any overt phenotype (Fig. S1, A and B, available at ; and not depicted). FACS analysis revealed normal expression and activation of β1 integrin (Fig. S4). Because impaired β1 integrin function severely affects interfollicular epidermis and hair follicle cycling (), we chose to more closely investigate the skin of these mice for defects. E15 and E16 showed normal skin development, and no hyperthickening of the epidermis, blister formation, or dermal fibrosis could be observed. Similarly, the hair follicle frequency, length, and morphology were unaffected in E15 and E16 mice (Fig. S1, C and D). Finally, RT-PCR analysis of the cytoplasmic domain of the β1 integrin mRNA from lox/lox wild-type and lox/lox mutant keratinocytes revealed a normal level of β1 integrin expression and faithful splicing of the cytoplasmic exons to generate the β1A mRNA (). RNA samples from striated muscle were used to amplify the muscle-specific exon D β1 integrin isoform (β1D). Neither the introduction of the loxP site upstream of exon 15 or 16 nor the point mutations in exon 15 or exon 16 led to isoform switching. Altogether, these findings indicate that the presence of the single loxP in the intron preceding exon 15 or 16 of the β1 integrin gene did not affect integrin expression and function. Homozygous D/A mutant mice lacking the salt bridge connecting the β1 and α subunits were born at the Mendelian ratio (not depicted), as determined by genotyping of the heterozygous D/A intercrosses (). The mutant mice were fertile and had a normal life span. Furthermore, D/A mutant mice had normal growth rates (Fig. S2, A and B, available at ) and did not display any obvious impairments (). Detailed histological analysis of the mutant skin at postnatal day (P) 1, P7, P14, P21, and 3 mo revealed normal interfollicular epidermis, normal epidermal–dermal junction, and normal hair follicle formation and progression through the hair cycle ( and not depicted). As in control skin, immunostaining of the D/A mutant skin showed β1 integrin and α6 integrin restricted to basal keratinocytes and laminin 5 (LN5) deposition at the dermal–epidermal junction (; and not depicted). Furthermore, proliferation as determined by BrdU incorporation was normal in the basal keratinocytes expressing the D/A mutant integrin, with 11.1 ± 0.07% BrdU-positive cells in control and 11.2 ± 2.26% in the D/A mutant mice in 14-d-old mice and 3.5 ± 0.14 and 3.7 ± 0.42% in 3-mo-old mice, respectively. Similarly, the proliferation of hair bulb cells was unaltered (unpublished data). To investigate the role of tyrosine phosphorylation of the β1 subunit for embryonic and postnatal development, we intercrossed heterozygous Y783F and Y795F mice, respectively. Surprisingly, both mutant mouse strains were viable and fertile (Fig. S1 E). The lack of an obvious phenotype in the single-tyrosine mutant mice prompted us to investigate mice in which both tyrosine residues were changed to phenylalanine (YY/FF). Astonishingly, the homozygous YY/FF mutant mice () were also viable and fertile and, moreover, did not show any obvious phenotype. Their weight and size was comparable to the age-matched littermate controls at all time points analyzed (Fig. S2, C and D). Furthermore, histological analysis of back skin at P1, P7, P14, P21, and 6 mo did not reveal any abnormalities with respect to the morphology of the interfollicular epidermis and the hair follicles ( and not depicted). Regardless of their age, YY/FF mutant mice displayed neither skin blisters nor signs of dermal fibrosis. Immunostaining of skin sections from 14-d- and 6-mo-old homozygous YY/FF mutant mice revealed normal expression of the β1 and α6 integrin subunits ( and not depicted) and normal deposition and assembly of LN5 at the epidermal–dermal junction ( and not depicted). Finally, the number of proliferating cells in the basal layer of the epidermis was not changed in the mutant skin, with 8.4 ± 0.14% BrdU-positive cells in control and 10 ± 0.21% in YY/YY mutant mice at 14 d of age and 3.55 ± 0.63 and 2.85 ± 0.78%, respectively, at 6 mo of age. Furthermore, the proliferation of hair bulb cells was similar in control and YY/FF mutant hair follicles (unpublished data). Both the D/A and YY/FF mutant cells attached to a mixture of collagen I (Col1) and FN-coated surface, spread, and grew in a manner indistinguishable from control cells (unpublished data). In vitro–expanded D/A and YY/FF keratinocytes were assayed for their ability to adhere to increasing concentrations of LN5, Col1, and FN. Dose-dependent adhesion of D/A mutant keratinocytes () and YY/FF mutant keratinocytes () was comparable to their respective control cells on all matrix substrates analyzed (). Similarly, the D/A as well as the YY/FF mutant keratinocytes spread to a similar extent on a Col1/FN mixture as control cells ( and not depicted). To investigate focal contact organization and F-actin distribution in mutant keratinocytes, cells were cultured for 2 d on a Col1/FN mixture and subsequently stained with antibodies against β1 integrin, talin, FAK, pY397FAK, vinculin and paxillin, and fluorescently labeled phalloidin. No obvious differences were observed in the size or number of focal adhesions (FAs) and in F-actin organization (; and Fig. S3, available at ). Similarly, the β1 integrin carrying either the D/A or YY/FF mutation localized normally to FAs ( and Fig. S3). Furthermore, total FAK, pY397FAK, and talin were found in FAs of D/A and YY/FF mutant keratinocytes ( and not depicted). To quantitatively assess the extent of talin recruitment into FA of the YY/FF mutant keratinocytes, we simultaneously stained mutant keratinocytes with anti-talin and anti-FAK antibodies and subsequently determined the ratio of the mean fluorescence intensities. The talin/FAK ratios were almost identical in FAs of control and YY/FF mutant cells (), suggesting that the YY/FF mutation does not result in increased talin recruitment to integrin adhesion sites. To test migration of mutant keratinocytes in vitro, we performed transwell migration and scratch assays. In transwell assays, control or D/A or YY/FF mutant keratinocytes were plated in transwell motility chambers and allowed to migrate toward medium lacking or containing EGF. In neither the absence nor the presence of EGF was the extent of migration significantly different between control and D/A or control and YY/FF mutant keratinocytes (). In scratch assays, we wounded a keratinocyte monolayer and monitored wound closure over a period of 6 h. The results demonstrate that the YY/FF mutant keratinocytes closed the wound with a speed that was not significantly different from control cells (). Hyperactivation of β1 integrin could lead to increased activation of downstream signaling molecules such as FAK or extracellular signal–regulated kinase (Erk). To assess the activation status of these integrin effectors, we analyzed wounding-induced phosphorylation of FAK and Erk in YY/FF keratinocytes. No significant difference was found between control and mutant cells either before or after wounding, suggesting that NPXY tyrosine phosphorylation is not crucial for inducing the activity of these effector molecules (). Previous studies have suggested that both D/A and YY/FF mutations will result in constitutive activation of β1 integrin (, ). Because the results of our cellular assays were inconsistent with constitutively active β1 integrin, we examined integrin activation more closely on different primary cells by measuring the binding of the 9EG7 antibody, which recognizes an epitope that is exposed only on activated β1 integrins. Freshly isolated keratinocytes from adult D/A mutant mice expressed β1, α2, α6, and β4 integrins at similar levels as control keratinocytes (). Recognition of the 9EG7 epitope of the β1 integrin identified a similarly small fraction of activated β1 integrins on D/A homozygous mutant and control keratinocytes (). Primary keratinocytes of YY/FF mutant mice had a slightly reduced expression of β1 integrin with a similar reduction of the 9EG7 epitope. Expression levels of α2 integrin were also slightly decreased, whereas α6 and β4 integrin were unchanged compared with control cells (). The 9EG7 epitope is not easily detected on epidermal keratinocytes (). In addition, the trypsin treatment during the preparation of the keratinocytes may abrogate the 9EG7 epitope. Therefore, we also analyzed the β1 integrin activation by FACS on granulocytes, macrophages, B-lymphocytes, and erythroblasts freshly isolated from the bone marrow of control and mutant mice (Fig. S4). Expression of β1 integrin was comparable between control and mutant cells, with the exception of the granulocytes and macrophages in YY/FF mice, which showed a reduction of surface β1 integrin by ∼50 and 30%, respectively (Fig. S4). We could not detect increased basal 9EG7 exposure on D/A and YY/FF mutant cells (Fig. S4). Interestingly, treatment of cells with 5 mM Mn markedly increased the 9EG7 epitope exposure to a similar extent on control (E15 and E16) and D/A or YY/FF mutant cells (Fig. S4). The assessment of ligand binding is an alternative method for evaluating integrin activation. To this end, we expressed and fluorescently labeled the central cell binding domain of FN (FNIII7-10), which is bound by the α5β1 integrin expressed on cultured primary keratinocytes. In the presence of 5 mM Mn control, D/A () and YY/FF () mutant keratinocytes bound the Alexa 647–labeled FNIII7-10 in a dose-dependent manner and to a comparable extent (). As expected, the exposure to 2 mM EDTA or Tris-buffered saline alone revealed only very weak or no ligand binding on all cell types analyzed (). Although heterozygous YY/AA mice were normal, homozygous YY/AA mutant mice died during embryonic development (unpublished data). To generate mice that express the YY/AA mutation only in keratinocytes, we intercrossed heterozygous YY/AA mutant animals with floxed β1 integrin mice expressing the Cre recombinase under the keratin 5 (K5) promoter () to obtain β1(YY/AA/fl) K5-Cre (YY/AA) mice. Upon Cre-mediated deletion of the floxed β1 allele, only the YY/AA mutant allele is expressed in keratinocytes. The YY/AA mice were born without obvious abnormalities, but by 2 wk, skin pigmentation and hair coat development was markedly impaired (). At 5 wk of age, almost the entire hair coat was lost and wounds appeared in mechanically stressed regions (unpublished data). This phenotype closely resembled the abnormalities observed in mice with a keratinocyte-specific deletion of the β1 integrin gene (). Histological analysis of skin sections further corroborated the similarities between the YY/AA and the β1 integrin–null mutation. The skin of 14-d-old YY/AA mutant mice displayed hyperthickened epidermis with aberrantly shaped cells, blister formation at the epidermal–dermal junction, and severe hair follicle abnormalities (). To confirm that only the YY/AA-mutated β1 integrin was expressed in skin, we performed LacZ staining, the expression of which indicates the deletion of the floxed β1 integrin allele (). The LacZ staining was clearly detectable in the interfollicular epidermis and the outer root sheath of hair follicles derived from YY/AA mutant mice (). Immunostaining revealed that basal and suprabasal keratinocytes of mutant mice express β1 integrin (, arrowheads indicate suprabasal expression), although the staining was partially discontinuous (not depicted). Similarly, the staining of the α6 integrin was discontinuous. Interestingly, β1 integrin is expressed in regions of skin blistering, suggesting that the mutant integrin is nonfunctional (, arrows). In 14-d-old mutant mice, LN5 was diffusely deposited at the epidermal–dermal junction in a manner similar to the β1-deficient epidermis (). To quantitatively assess the integrin levels on keratinocytes, we performed FACS analysis with freshly isolated keratinocytes from 6-d-old control and YY/AA mutant mice. The levels of β1, β4, and α6 integrin were reduced by ∼50% and the levels of α2 integrin by ∼25% (). 9EG7 staining revealed almost no signal on YY/AA keratinocytes, in contrast to controls, suggesting that YY/AA integrins are present in an inactive conformation on mutant keratinocytes (). The YY/AA mutant keratinocytes do not display the 9EG7 epitope and show a 50% reduction of the β1 integrin expression on the cell surface. To confirm that the skin phenotypes in mutant mice arose from the loss of talin–β1 integrin tail interaction and defective integrin activation and not from the reduced integrin expression, we established a mouse strain carrying a hypomorphic, wild-type β1 integrin allele (). The mouse strain was obtained by fusing a 141-bp cDNA fragment encoding the cytoplasmic domain of the β1 integrin gene in frame into exon 15, just behind the transmembrane span encoding sequence (Fig. S5, available at ). Homozygous mice carrying the allele died during development (unpublished data). To express the mutation exclusively in keratinocytes, we intercrossed mice with floxed β1 integrin mice expressing the Cre recombinase under the K5 promoter (). Although freshly isolated keratinocytes expressed only ∼25% of the normal β1 integrin level on their surface (), they developed a milder phenotype when compared with the YY/AA, which express 50% of the mutant β1 integrin (). At 2 wk of age, the hair coat of the mice was slightly reduced and only a few of the hair follicles were misshapen and resembled the YY/AA or K5-β1 phenotype (). Similarly, the degree of epidermal hyperthickening, aberrant shape of basal keratinocytes, and extent of blistering was much less prominent in the mice when compared with the YY/AA or K5-β1 mutant mice ( and not depicted). LN5 deposition at the dermal–epidermal junction was broadened in skin, although to a much lesser extent when compared with YY/AA and K5-β1 mutant skin (). Altogether, these data suggest that the phenotypic alterations observed in YY/AA mutant mice are likely caused by an impairment of the talin–β1 integrin tail interaction, rather than a reduced β1 integrin expression. To analyze adhesion and cytoskeleton of YY/AA keratinocytes, freshly isolated mutant keratinocytes were seeded on Col1/FN-coated plastic dishes. The majority of control cells appeared well spread 24 h after plating and were confluent 72 h later. In sharp contrast, only a few mutant cells were attached 24 h after seeding. They appeared round and failed to spread and proliferate (). Immunostaining of the few adherent cells (cultured for 2 d on Col1 and FN) for β1 integrin, paxillin, and vinculin revealed a complete absence of FAs and stress fibers and the formation of F-actin clumps and/or fine F-actin filaments extending radially from the perinuclear region to the plasma membrane. Almost the entire immune signal of β1 integrins and the FA components was found within the cytoplasm (). Altogether, these data suggest that the YY/AA mutation in the cytoplasmic domain of β1 integrins renders the integrins inactive. A key feature of integrins is their ability to switch between an inactive and an active conformation. It is believed that talin binding to the cytoplasmic domain of the β subunit disrupts the α/β tail interaction, leading to tail separation and changes in the conformation of the extracellular domain, consistent with integrin activation (; ). In the present study, we report the phenotypic analysis of mouse strains, which harbor point mutations in either the talin binding motif or the membrane-proximal salt bridge of the β1 integrin subunit. It has been shown that the substitution of the tyrosine residue with alanine in the proximal NPXY motif of β tails is sufficient to inhibit binding of talin and lock integrins in an inactive conformation (). To examine the consequences of this finding during the development of a multicellular organism, we generated mice in which we substituted the tyrosine residues in the proximal and distal NPXY motifs of the β1 integrin subunit with alanine. We obtained compelling evidence that functional NPXY motifs are required for β1 integrin function both in vivo and in vitro. First, mice with a homozygous YY783/795AA (YY/AA) mutation die during development. Second, expression of the YY/AA mutation specifically in keratinocytes mirrored the β1 integrin–null skin phenotype () with epidermal hyperplasia, skin blistering, abnormal distribution of basement membrane proteins, and a dermal fibrosis. Finally, primary keratinocytes from YY/AA mutant skin did not express detectable activated (9EG7-positive) β1 integrins and exhibited severely compromised integrin functions characterized by profound adhesion defects, impaired spreading, FA assembly, and F-actin distribution. The YY/AA mutant keratinocytes expressed only ∼50% of the normal β1 integrin level on their cell surface. Reduced surface expression of β1 integrin in the absence of talin binding is in agreement with a previous report showing that talin controls β1 integrin exit from an early compartment of the secretory pathway (). To exclude the possibility that the reduced integrin levels rather than the abolished integrin activation is responsible for the YY/AA phenotype, we generated a hypomorphic β1 integrin mouse strain expressing only ∼25% of wild-type β1 integrins on keratinocytes. The hypomorphic mouse strain developed a much milder skin phenotype than the defects observed in the YY/AA mutant mice, which clearly suggests that loss of integrin activation rather than the reduced expression is responsible for the severe skin and hair abnormalities in YY/AA mutants. Because integrins can quickly switch between an active and an inactive conformation, the binding of talin to the β integrin tails must be tightly and efficiently regulated. Several studies reached the conclusion that this regulation may be achieved by reversibly phosphorylating the tyrosine residues of the β tail. In this model, tyrosine phosphorylation of the β tail NPXY motif by Src family kinases inhibits talin binding and displaces β1 integrins from FA sites (; ; ). Inhibiting phosphorylation of the NPXY motif by substituting the tyrosine with a nonphosphorylatable phenylalanine resulted in enhanced FN binding and assembly (), which was suggested to be due to enhanced talin binding to the NPXF motif. We tested the role of tyrosine-783 and -795 phosphorylation in vivo by substituting either each tyrosine individually (Y783F and Y795F) or both together (YY783/795FF; YY/FF) with phenylalanine. Surprisingly, the mice with a single tyrosine mutation displayed no obvious defects. Moreover, mice with mutations in both tyrosine residues were also born without phenotypic changes. When we searched the skin of YY/FF for subtle defects, we found normal keratinocyte layers in the epidermis, normal keratinocyte proliferation, regular distribution of basement membrane proteins, absence of skin blisters, and normal hair follicle formation and cycling. Moreover, adhesion, spreading, FA formation and distribution, F-actin organization, and migration were unaffected in freshly isolated YY/FF mutant keratinocytes. The absence of abnormalities in YY/FF mice strongly indicates that the mutant β1 integrin cytoplasmic tail is still able to recruit talin and activate the integrin. Furthermore, the in vivo findings also indicate that the YY/FF mutation still allows switching β1 integrins between active and inactive conformations. Several additional experiments confirm this notion. First, we found the surface levels of β1 integrins detected with an activation-associated epitope increased neither on keratinocytes nor on hematopoietic cells. Second, we could induce integrin activation with Mn on keratinocytes and different hematopoietic cells to a similar extent as on the corresponding control cells. Finally, migration-induced integrin activation triggered comparable FAK and Erk phosphorylation (outside-in signaling) in YY/FF mutant and control keratinocytes. These observations allow some important conclusions: first, β1 integrin cytoplasmic tyrosine phosphorylation is dispensable to reverse talin-mediated integrin activation under physiological conditions. It is possible that NPXF bound talin is released from the mutant β1 integrin cytoplasmic domain by other β1 tail binding proteins, such as integrin cytoplasmic domain–associated protein (ICAP-1; ) or filamin (). Second, β1 integrin cytoplasmic tyrosine phosphorylation plays no obvious role in integrin outside-in signaling, although an increasing number of proteins are capable of binding the NPXY motifs, including myosin X and Kindlin-1 (; ). However, it also possible that the integrin NPXY phosphorylation plays an important role during stress situations such as wound healing or inflammation, where cells move at high speed to restore tissue integrity. Such experiments are currently being performed in our laboratory. The disruption of the membrane-proximal salt bridge between the α/β tails and their separation were suggested to be important steps toward integrin activation. Surprisingly, however, the destruction of this salt bridge by the substitution of the aspartic acid 759 with an alanine (D759A; D/A) in the β1 membrane-proximal region produced no obvious phenotype. Furthermore, freshly isolated keratinocytes or hematopoietic cells from the bone marrow of control and D/A mutant mice expressed low but similar levels of activated β1 integrins on their surface. Incubation of D/A mutant primary bone marrow cells with Mn induced integrin activation (as assessed with the 9EG7 antibody) to an extent similar to corresponding control cells. Moreover, adhesion, spreading, and migration of D/A mutant keratinocytes through transwell chambers was unaffected by the disruption of the α/β1 integrin salt bridge. Together, these data indicate that the genetic disruption of the membrane-proximal salt bridge between the α and the β1 integrin subunits neither locks β1 integrins in a more activated state nor increases their sensitivity to activation-inducing agents such as Mn, permitting normal integrin function in vivo and in vitro. Although our findings indicate that the salt bridge between the α and the β cytoplasmic domain plays no rate-limiting role for development and postnatal life of mice, previous biochemical and cell biological studies have demonstrated that the salt bridge plays an important function in integrin activation (), integrin clustering (), and cell migration (). Interestingly, structural analysis of the αIIβ and β3 tails by nuclear magnetic resonance spectroscopy also produced conflicting results. Although one group reported an interaction (, ), albeit a very weak one, two other reports failed to detect α/β tail interactions (; ). It is possible that the weak, handshake-like interaction between the α/β tail exists but plays a less important role as long as other α/β interactions are present, for example, in the membrane-spanning helical regions (; ). This can be rigorously tested by introducing specific mutations into the transmembrane domain of β integrin subunits. Mouse genomic clones used to generate the targeting constructs were described previously (). The loxP-flanked neomycin cassette was inserted into either the NcoI site upstream of exon 15 () or into the PstI site upstream of exon 16 (). Both constructs were used to produce mice (E15 and E16). Point mutations were introduced in exon 15 and 16 using the AlteredSites II in vitro mutagenesis system (Promega). The D/A mutation was subsequently transferred into the E15 construct, and the F and A mutants were introduced into the E16 construct. The targeting constructs were electroporated into R1 ES cells. ES cell clones that underwent homologous recombination were isolated, identified by Southern blot using BamHI-digested genomic DNA and a 1-kb 3′ external probe, and injected into blastocysts to generate germline chimeras. Upon germline transmission, mutant mice were intercrossed with deleter-Cre transgenic mice () to obtain heterozygous E15, E16, D/A, Y783F, Y795F, YY/FF, and YY/AA mutant mice, which were subsequently interbred to obtain homozygous mutant mice. For all experiments performed with these mouse lines, either heterozygous or wild-type littermates served as control animals. Because YY/AA mutants were embryonic lethal, they were intercrossed with mice carrying a floxed β1 integrin gene () and mice carrying K5 promoter-driven Cre recombinase transgene (). Littermates carrying the floxed and the YY/AA mutated β1 integrin allele without the Cre recombinase transgene served as controls for experiments performed with the YY/AA mutant. Genotyping was performed by Southern blot or PCR using DNA isolated from tail biopsies. All animal studies were approved by the Regierung von Oberbayern. For RT-PCR, 50 ng of total RNA were reverse-transcribed using the iScript synthesis kit (Bio-Rad Laboratories), according to the manufacturer's protocol. The single-strand cDNA was used as a template for a PCR reaction, using a forward primer hybridizing to exon 14 (5′-AGGACATTGATGACTGCTGG-3′) and a reverse primer hybridizing to exon 16 (5′-CCAAAACTACCCTACTGTGAC-3′) of the β1 integrin gene. Hematoxylin-eosin (HE) and immunofluorescence staining were performed as described previously (). Because freshly isolated keratinocytes adhere and spread slowly on ECM substrata, all immunostainings on primary cells were performed after 48 h. BrdU staining was performed according to the manufacturer's instructions (Roche Diagnostics). β-Galactosidase activity was determined as previously described (). Images were collected by confocal microscopy (DMIRE2; Leica) using Leica Confocal Software version 2.5 Build 1227 with 40× or 63× oil objectives, by fluorescence microscopy (DMRA2; Leica) using SimplePCI software (version 5.1.0.0110; GTI Microsystems) with 63× oil objectives, or by bright field microscopy (Axiovert; Carl Zeiss MicroImaging, Inc.), using IM50 software (Leica) with 10× or 40× objectives. All images were collected at RT. Immunofluorescence staining intensities from talin and FAK confocal images (detected by Cy3- and Alexa 488–coupled secondary antibodies, respectively) were quantified using the Leica Confocal Software. In brief, confocal stacks with highest talin fluorescence intensities were selected, and single FAs were selected for quantification. The mean fluorescence amplitude of talin and FAK of each FA was used to perform the ratiometric analysis shown in . Images were processed with Photoshop (Adobe). Primary keratinocytes were isolated from P6 or adult mice according to and grown to confluence on a mixture of Col1 (Cohesion) and 10 μg/ml FN (Invitrogen) coated plastic in keratinocyte growth medium containing 8% chelated FCS (Invitrogen) and 45 μM Ca. To test keratinocyte migration, transwell chambers (Falcon) with 8-μm-diameter pores were coated on the lower surfaces with a mixture of Col1 and 10 μg/ml FN. In duplicate assays, 4 × 10 primary keratinocytes were suspended in 45 μM calcium-containing Eagle's minimum essential medium (Sigma-Aldrich) supplemented with 1× -glutamine (Invitrogen) and seeded onto the transwell chambers. The chambers were placed in keratinocyte growth medium with or without 50 ng/ml EGF (Sigma-Aldrich) and incubated for 14 h at 37°C. Afterward, cells in the upper surface of the membrane were removed and the cells on the lower surface fixed and stained with 20% methanol/0.1% crystal violet. The relative number of migrated cells (compared with the number of all attached cells) was determined from five randomly chosen microscopic fields of ∼1.2 mm per duplicate experiment. Cell adhesion of primary keratinocytes to LN5 (provided by M. Aumailley, University of Cologne, Cologne, Germany), FN, and Col1 was assayed as described previously (). All experiments have been performed in duplicate. To assess the spreading kinetics of primary keratinocytes, cells were plated on mixture of Col1 and 10 μg/ml FN–coated plastic and monitored at 37°C in 5% CO by phase-contrast live cell imaging microscopy using a microscope (Axiovert 200M ; Carl Zeiss MicroImaging, Inc.) coupled to a camera (Visitron Systems). Images were collected with 10× objective. The cell area of 50 randomly selected cells per time point and genotype was quantified using MetaMorph 6.0 software (Molecular Devices). Wound healing assays and the measurement of the activity of signaling molecules upon wounding was performed as described previously (). An expression plasmid (obtained from H.P. Erickson, Duke University Medical Center, Durham, NC) encoding a His-tagged version of FN type III repeats 7–10 (FnIII7-10) containing the central cell binding domain () was used to express FnIII7-10 in . Recombinant FnIII7-10 polypeptides were purified by TALON Metal Affinity chromatography according to the manufacturer's instructions (BD Biosciences). The final preparation was dialyzed in PBS. To label FnIII7-10 polypeptides with a fluorescent probe, the purified fragment (2 mg/ml) was first adjusted to pH 8.4 and subsequently incubated with 1 mg of Alexa Fluor 647 carboxylic acid and succidimidyl ester (Invitrogen) for 1 h at RT. The uncoupled probe was removed by dialysis in two changes of PBS at 4°C. To assess ligand binding properties of primary keratinocytes, confluent cells were harvested and incubated (3 × 10 cells per sample) with Alexa 647–coupled FNIII7-10 fragment in Tris-buffered saline in the presence or absence of 2 mM EDTA or 5 mM MnCl and, finally, subjected to flow cytometry analysis. Freshly isolated keratinocytes (5 × 10 cells per sample) were stained with antibodies against α6 integrin, α2 integrin, β1 integrin (Ha2/5 epitope; recognizes all β1 integrin subunits), β1 integrin (9EG7 epitope; recognizes an activation-associated epitope and hence only activated β1 integrins), or β4 integrin and subjected to FACS analysis (). All antibodies were either FITC labeled or unlabeled and used in a 1:200 dilution, except 9EG7, which was used in a 1:100 dilution. Unlabeled antibodies were visualized with FITC α-rat IgG (1:200). All antibodies were obtained from BD Biosciences. Dead cells were excluded from FACS analysis by the addition of 2.5 μg/ml propidium iodide before FACS analysis (FACSCalibur CellQuest Pro Software; Becton Dickinson). Freshly prepared single-cell bone marrow suspensions (10 cells per sample; ) were stained with the β1 integrin–specific antibodies Ha2/5 or 9EG7 in Tris-buffered saline, containing 3% BSA with or without 5 mM MnCl (). In addition, cells were counterstained with antibodies against CD45R/B220 (B cell marker), Ly6G/Gr-1 (granulocyte marker), CD11b/Mac-1 (macrophage marker), or Ly-76/Ter-119 (erythroblast/erythrocyte marker), which were either PE-labeled or biotinylated. Biotinylated antibodies were detected with Streptavidin-Cy5 (Jackson ImmunoResearch Laboratories). Fig. S1 shows the development of the lox/lox mice and mice carrying a single Y/F mutation. Fig. S2 shows the weight and size of aging D/A and YY/FF mutant mice. Fig. S3 shows the FA and stress fiber formation in YY/FF mutant keratinocytes. Fig. S4 shows the β1 integrin expression and activation on E15, D/A, E16, and YY/FF bone marrow cells. Fig. S5 shows the targeting strategy used to obtain the mice. Online supplemental material is available at .

My father's a scientist, so it's an “apple doesn't fall far from the tree” kind of situation. He's at Harvard Medical School as a chemist who works in biology. I sort of grew up seeing science as a happy way to live a life. I enjoyed the way my dad seemed to approach the world, as puzzles and wanting to find out the truth, that sort of stuff. It seemed like an appealing way to have a career. Yes, actually my first experience in a lab was during high school. I did a summer program called the Research Science Institute, and worked in a lab in Georgetown (Washington, DC). Then, the following summer I worked in my dad's lab. And when I went to university (at Harvard) I did an undergraduate thesis in Tom Maniatis's lab, working on the NFκB transcription factor. I worked in Tom's lab full-time in the summer holidays and whenever I could during the year. Yeah, there's never been a huge question about it. Weidong Wang, in the lab, was just beginning to clone subunits of mammalian ATP-dependent chromatin remodeling complexes, and one of the intriguing surprises was that actin-related proteins turned out to be a fairly common component of these complexes. So I was interested in what an actin-related protein would be doing in a nuclear complex; actin wasn't supposed to do anything in the nucleus, according to canon. The reason I got into genomic approaches to chromatin is that I thought, “Well, if there was some direct role for actin polymerization in chromatin structure, it wasn't likely to be shifting a single nucleosome two base pairs to the left.” Presumably what would be happening would be large-scale changes in which nucleosomes are positioned or in some other aspect of chromatin structure. I assumed, being at Stanford while all the microarray stuff was going on with Pat Brown and Ron Davis, that there would also be genomic ways to look at chromatin structure. I was surprised to find that there weren't at the time. It's like faculty with training wheels. They're very similar to Whitehead Fellows, although there are differences in detail. But essentially it was a five-year position where we had PI rights and we had funding for ourselves plus two other people. In other words, they would fund you to run a lab of three people. One of the great things about the program was that there were approximately ten fellows at a time, and they were drawn from a wide range of disciplines. I was the boring molecular biologist, but there were mathematicians and physicists and evolutionary biologists. I collaborated with a number of the other fellows and learned a lot even from the ones I didn't work with. Most extensively, we did a lot of our early chromatin work with Steve Altschuler and Lani Wu, who are mathematicians. Going through the process of learning to talk to someone who speaks such a different scientific language was illuminating, and I think it really forced me to reevaluate some of the hidden assumptions that are built into how I would usually talk about biology. It was very stimulating working with people from different disciplines. It was a wonderful opportunity, a great position. When you look at a single gene, you wonder whether that gene is unusual in its behavior or whether that's how all genes behave. A good example of the power of genomic approaches is Audrey Gasch's paper on environmental stress responses. There had been this transcriptional literature for years and years describing how gene A gets turned on when you put hydrogen peroxide on a cell, and gene B gets turned on by heat-shocking cells. This was moving forward piecemeal. Then, Audrey ran yeast through tons of different stresses and found that there was a core group of 800 or so genes that changed expression under virtually all stress situations. It would have required an almost infinite number of single-gene studies to realize that these particular genes behave so coherently, and that this group doesn't just respond to one type of stress, but to all types. Another example from our own field is histone H3 lysine-4 methylation. If you look genome wide, you find that the more a gene is transcribed, the more trimethylation there is at its 5′ end. So you see this modification all over the place. But when you look at the whole genome's response to not having that mark—if you knock out the gene responsible for lysine-4 methylation—the gene expression defects are not that widespread. In other words, this methylation is happening over all active genes, but only a fraction of genes care about it. Without a genome-wide approach you wouldn't easily have picked up on that. So, the genome-wide results help to frame the next set of questions about how histone modifications work in the cell. I feel like yeast is going to end up looking a lot like a subset of metazoan chromatin. In other words, a lot of the things we think we understand about yeast, like the role of K-36 methylation or where K-4 methylation happens, so far turn out to be the same in higher eukaryotes. The difference is that higher eukaryotes also have these other systems piled on top. So they have additional modifications. They have expansions of the various histone families to form a wide variety of subunits. And so in general I think the basic lessons learned from yeast will apply to metazoans, but then there will be additional layers of complexity. I really like my colleagues here. There's a strong feeling of excitement about the direction of the place. There's people who work on chromatin, people who work on RNA, people who are interested in transcription and nuclear structure—all things within a stone's throw of what I spend a lot of time thinking about. It's a great fit. We have an idea for how to look at chromatin's secondary structure, in other words the 30-nm fiber. It's going to be technically challenging, and I don't know if it'll have the kind of signal-to-noise one would need to get anything useful out of it. But we're interested in those questions. Right now I'm thinking a lot about histone movement, replacement, sliding, etc. Most interesting is the question of what happens to nucleosomes during genomic replication, since the details of their behavior at the replication fork will really help constrain thinking about inheritance of chromatin states. These questions about histone movement are related to our recent work measuring nucleosome exchange rate during G1 arrest, and there is a bunch of interesting mechanistic follow-up to do there as well. One of the things we find is that nucleosomes are exchanging in and out of the promoters of most genes, but the highest exchange rates are found at genes that are regulated by ATP-dependent remodelers. It's possible, then, that either dissociation or reassociation or some part of the turnover process involves ATP-dependent remodelers. So we'd also like to look into that. We're also planning on extending the histone turnover work to doing studies in the mouse. We'd like to make mice where we can ask what is the turnover rate of a given histone variant in the whole animal and in different tissues. Yes, I'm having a blast right now. I'd like to be having a blast doing science in 20 years, 50 years. I really enjoy what we're able to think about and look at, and I hope that I manage to structure my life in such a way that I can continue to have fun doing science and thinking about problems well down the road.

There is increasing evidence for a spatial organization of transcription (; ). Pol II molecules form clusters within cells (), and nascent transcripts accumulate there, defining these clusters as transcription sites (; ). These transcription sites can transcribe different genes from distant parts of the same chromosome or potentially even different chromosomes (). According to the transcription factory model (), transcription sites contain immobilized pol II molecules that spool the chromatin template in and out of the site. To date, however, evidence for movement of the chromatin template through a transcription site is largely theoretical. It has been argued that because transcripts appear within a restricted volume defining the transcription site, the polymerase cannot move very far, and so it is more likely that the chromatin template moves (). This scenario also solves entanglement problems of the transcript and template (). If the chromatin template is reeled in and out of a transcription site, this site should be adjacent to or surrounded by decondensed, transcribed chromatin. In fact, transcription sites are surrounded by chromatin (), but, because most structures in the nucleus are found within chromatin, it has not been clear whether the chromatin seen around any one transcription site is associated with loci being transcribed by that site. To investigate the spatial organization of chromatin at a transcription site, we have taken advantage of a mouse cell line harboring a tandem array. The array is composed of 200 directly repeated copies of a 9-kb element composed of the mouse mammary tumor virus (MMTV) promoter followed by reporter gene sequences (). Transcription from the array can be induced above basal levels by a hormone-stimulated GFP-tagged glucocorticoid receptor (GR) that also enables visualization of the array in live or fixed cells (). Hormone induces a transcriptional response at the array comparable with that at single-copy MMTV promoters, including the recruitment of cofactors (), specific nucleosome remodeling (), and adaptation to prolonged hormone treatment (). In addition, higher order chromatin structures at the array are indistinguishable from the structures observed in transcriptionally active domains of natural chromosomes (). Therefore, the array exhibits several features that are characteristic of normal transcription. Because of its size, however, the array is readily detected by light microscopy. Thus, it provides a useful model system for examining in a single cell the spatial distribution of molecules associated with a transcriptionally active locus to construct a more unified picture of the nuclear organization of both these molecules and the chromatin at a transcription site. Using this approach, we report evidence for a previously undetected spatial organization at a transcription site, namely a domain of decondensed chromatin that borders or surrounds the transcription sites and appears likely to contain recently transcribed chromatin. We examined the location of transcription sites at the array by first using bromo-UTP (BrUTP) incorporation for detection of nascent transcripts. This consistently yielded a series of BrUTP puncta associated with the GFP-GR–tagged array (). As previously described, the array itself is composed of GFP-GR puncta or beads (). To ascertain whether the BrUTP puncta overlaid the GFP-GR beads, we performed 3D deconvolution for improved resolution, including corrections for residual chromatic aberration along the optical axis. We consistently found that the BrUTP puncta did not directly colocalize with the GFP-GR beads but rather interdigitated between the beads, with some overlap at the edges of these two distributions (). These observations are consistent with earlier studies suggesting that transcription occurs predominantly at or near the surface of compact chromatin domains, namely in the interchromatin or perichromatin domains (; ). Because the BrUTP incorporation procedure involves live cell permeabilization that might conceivably alter the relative distribution of transcription sites and GFP-GR beads, we used an alternate approach to address the same question. Fixed cells were probed with an antibody (H5) against the phosphorylated CTD domain of pol II to determine its association with GFP-GR at the array. This likewise yielded a punctate staining pattern for active pol II that was clearly enriched at the array (). This punctate pol II staining pattern is consistent with live cell images from a previous study that examined a GFP-tagged pol II in the array cell line (). Again, using 3D deconvolution for improved resolution, we found that the active pol II, like the BrUTP incorporation sites, did not directly colocalize with the GFP-GR beads but interdigitated between them (). Finally, we analyzed the degree of overlap between the active pol II and the BrUTP stains at the array and found a high degree of colocalization (). These observations identify the BrUTP puncta as transcription factories and are consistent with previous studies demonstrating transcription foci in various cell types (; ). The pol II factories that we detected are larger than typical pol II factories but comparable in size with pol I factories (). This similarity in size may reflect the fact that the pol I factories also associate with a tandem array (in this case, of ribosomal genes). In summary, these results establish that the transcription sites at the array are located directly adjacent to the GFP-GR beads, with some overlap at the edges between these two distributions. It has been proposed that loops of highly decondensed chromatin extrude from transcription sites (). If so, at the resolution afforded by light microscopy, each transcription site at the array should be associated with a domain of decondensed chromatin. However, our previous DNA FISH experiments suggested that array chromatin exactly coincides with the GFP-GR beads (Fig. S1, available at ; ). This GFP-GR bead chromatin could, in principle, correspond to the predicted decondensed domain, but our previous estimates suggest it is considerably more condensed than expected for transcribed chromatin (). We reasoned that if additional, more decondensed chromatin was associated with transcription sites at the array, its fragility might make it difficult to preserve by our earlier procedure of DNA FISH with denaturation at 95°C (). Thus, we performed DNA FISH at a lower denaturation temperature (70°C) and compared the results to DNA FISH with denaturation at 95°C. At 95°C, we once again detected beaded structures identical to those we had previously observed (). However, with denaturation at 70°C, we could also detect specific MMTV-labeled chromatin structures in every cell (). These structures contained some puncta that resembled the beads seen at 95°C, but the structures seen at 70°C also exhibited a haze interspersed between the puncta that was not as evident at 95°C. Furthermore, direct measurement of areas encompassed by the structures demonstrated that those detected at 70°C were significantly larger (P < 10) than those detected at 95°C (). The 70°C structures were never detected in control experiments in which the specific DNA probe was omitted, although staining of the nuclear periphery and random spots within the nucleus was still apparent (Fig. S2 a, available at ). All of the specific structures detected by these two FISH protocols contain DNA, as an RNase treatment is always included in the DNA FISH procedures, and both the 70 and 95°C structures were eliminated by pretreating cells with DNase (unpublished data). To determine whether there was any overlap between the DNA detected by the 70 and 95°C procedures, we devised a dual-temperature DNA FISH protocol that involved FISH at 70°C with a red-labeled probe followed by an additional fixation step to ensure preservation of the 70°C structure and FISH at 95°C with a green-labeled probe. This dual FISH procedure consistently enabled the preservation and detection of two distinct structures that showed virtually no overlap between the red (70°C) and green (95°C) labels in all cells (). This observation suggests that the 70 and 95°C structures are largely exclusive. Observation of many cells with the dual FISH procedure showed that the 70°C (red) structure typically surrounded the 95°C (green) structure (), although in a few cells, the 70°C (red) structure protruded largely from one side of the 95°C (green) structure (). We next investigated whether the decondensed domain arose as a result of transcription. To test this, we measured areas of the decondensed domain detected by DNA FISH at 70°C as a function of time before and after transcriptional activation by hormone induction. Before activation and consistent with the known low levels of basal transcription from the MMTV promoter (), small decondensed domains were visible in some cells, whereas in other cells, none could be detected (). In contrast, a single chromatin bead could always be detected by DNA FISH at 95°C, marking the site of the condensed array (). After activation, decondensed domains were present in every cell (), and their mean area increased substantially over time (). As a second test of the decondensed domain's association with transcription, we induced transcription by the addition of hormone but simultaneously added a transcriptional inhibitor, DRB (5,6-dichloro-1-β--ribobenzimidazole; ). This significantly inhibited (P < 10) formation of the decondensed domain (), also suggesting that formation of this domain is coupled to transcription. As another test for the possible involvement of the decondensed domain in transcription, we investigated its association with a topoisomerase. Transcription generates positive supercoils in front of a polymerase and negative supercoils behind it (). If not relieved by topoisomerase action, the resultant torsional strain may accumulate to levels that could stall transcription (). We stained the array cell line with two different antibodies against topoisomerase IIα. For each antibody, we detected a similar association pattern with the array: a region of topoisomerase IIα staining extended around and beyond the GFP-GR beads (). To determine the relationship of the topoisomerase IIα staining pattern with the decondensed domain, we performed immuno-FISH and found that the topoisomerase II stain and the decondensed domain consistently overlapped (). These results suggest that topoisomerase IIα associates with the decondensed domain and may perform some function there. To test this, we inhibited topoisomerase IIα using the drug etoposide. We found that formation of the decondensed domain was impaired () compared with controls in which cells were treated with vehicle only. As detected by RNA FISH, etoposide treatment also sharply reduced transcription from the array compared with the controls (). These results indicate that transcription sites at the array are associated with a surrounding region of topoisomerase IIα that is required both for transcription from the array and for formation of the decondensed domain around the array. Several studies in both yeast and mammals have demonstrated that histones in recently transcribed chromatin are marked with a trimethyl H3K36 modification (; ; ; ). We reasoned that if the decondensed domain contains recently transcribed chromatin extruded from the pol II factories, it should show increased levels of trimethyl H3K36. To determine whether this mark was associated with transcribed chromatin from the MMTV array, we performed chromatin immunoprecipitation (ChIP) using an antibody specific for trimethyl H3K36 and compared the levels of this mark within the MMTV promoter to the downstream ras reporter gene. Consistent with previous studies of other genes (; ; ; ), we found that compared with the MMTV promoter, the reporter gene sequence exhibited a substantial enrichment for the trimethyl H3K36 mark. This differential effect was enhanced upon the hormone induction of transcription but was still detected to a lesser degree without hormone (), which is consistent with basal transcription from the MMTV promoter () and with our unpublished observations of RNA FISH accumulation at the array in the absence of hormone. With this evidence for trimethyl H3K36 enhancement in the transcribed reporter gene sequence, we proceeded to examine the distribution of this mark at the MMTV array by confocal microscopy. Immunofluorescence with the same trimethyl H3K36 antibody used for ChIP revealed a staining pattern that surrounded the GFP-GR beads (). To follow up this observation, we also performed immunofluorescence with an antibody against the N terminus of the human huntingtin-interacting protein B (HYPB), which possesses H3K36 histone methyltransferase activity () and is an orthologue of the Set2 methyltransferase responsible for the H3K36 trimethylation mark in yeast (). This HYPB antibody also exhibited a staining pattern that surrounded the array (), suggesting that the trimethyl H3K36 mark itself as well as an enzyme potentially responsible for it were associated with the decondensed domain. In contrast, strikingly different staining patterns were observed in confocal images of antibodies directed against histone marks typically associated with active promoters and 5′ regions (; ; ; ; ). Of the three antibodies tested (generically acetylated H4, trimethyl H3K4, and acetyl H3K9), all stained the condensed chromatin domain, yielding substantial colocalization with the GFP-GR beads, but showed little or no stain of the decondensed domain (). To distinguish between these possibilities, we performed DNA FISH at both 70 and 95°C with a probe for the MMTV promoter sequence. These FISH experiments revealed that this sequence was present in both the condensed (95°C FISH) and decondensed (70°C FISH) domains (). Thus, promoter sequences do not appear to be preferentially retained within the condensed domain. Consequently, the enhancement of active promoter marks and GFP-GR staining in the condensed domain most likely reflects the increased chromatin concentration there. We then repeated these probe-specific FISH experiments, but with a probe for the ras reporter gene sequence. Here, as for the promoter sequences, we could also detect reporter sequences in both the condensed and decondensed domains (). Thus, despite both the increased chromatin concentration and the presence of reporter gene sequences in the condensed domain, staining for the trimethyl H3K36 mark is not enhanced in the condensed GFP-GR beads but rather only in the decondensed domain. Because the trimethyl H3K36 mark labels recently transcribed chromatin, this result argues that the reporter gene sequences in the condensed domain have not yet been transcribed and that as these sequences are transcribed, they appear in the decondensed domain. Structural analysis of transcriptionally active chromatin is challenging as a result of difficulties in identifying, preserving, and resolving the structures at such sites. We have overcome some of these limitations in this study by developing a new protocol for DNA FISH and applying it to a tandem gene array that is easily visualized by light microscopy. With these tools, we have now identified three different structures at the array that provide new insights into how transcription may occur there (). First, as we previously described (), we find a series of adjacent puncta or beads of relatively condensed chromatin that can be identified by either conventional DNA FISH or in live or fixed cells by the accumulation of GFP-GR. Second, directly adjacent to this condensed domain, we find transcription sites identified by either BrUTP incorporation or by an antibody against the active form of pol II. Third, we find that these transcription sites are surrounded by and contained within a larger domain that is composed of more decondensed chromatin from the array. As explained below, our results suggest that this decondensed domain arises from the extrusion of transcribed sequences from an immobilized polymerase, providing new support to the pol II factory model of transcription. Some hints for chromatin-surrounding transcription sites had previously come from electron microscopy sections of HeLa cells in which transcription sites were detected by biotinylated RNA, and the presence of chromatin surrounding them was inferred by a uranyl-EDTA regressive staining technique (). Because this procedure detects all transcription sites and all chromatin, some amount of interpretation was required to imagine where the associated chromatin might begin and end for each transcription site or even whether the chromatin adjacent to a transcription site was composed of DNA associated with that site. Our new evidence for a decondensed chromatin domain surrounding transcription sites at the array is more direct and substantial. Using a specially developed, gentler DNA FISH protocol, we were able to detect decondensed, array-specific chromatin extending to a clear boundary around only the transcription sites associated with the array. This demonstrates that a specific set of transcription sites is surrounded by a decondensed chromatin domain composed of sequences from the loci being transcribed. It seems likely that we and others have missed such decondensed domains before by DNA FISH because they are difficult to preserve, are normally composed of a variety of different DNA sequences dependent on the genes being transcribed at the transcription site (), and are likely to be much smaller for a transcription site associated with single-copy genes of moderate transcriptional activity. We made several observations linking the array's decondensed domain with transcription. DRB treatment, which blocks transcriptional elongation (), hinders formation of the decondensed domain, suggesting that transcriptional elongation is required for the formation of the decondensed domain. We also found that topoisomerase IIα associates with the decondensed domain and so is poised to remove supercoils that would arise on either side of a transcribing polymerase (). Inhibition of topoisomerase II function by a brief (45 min) drug treatment impaired formation of the decondensed domain and dramatically reduced transcription. Both effects could arise if the drug treatment blocked the elongation of pol II either as a result of accumulated torsional strain or immobilized topoisomerase complexes, although effects of topoisomerase inhibition on promoter activation are also possible (). More direct molecular evidence for the role of the decondensed domain in elongation comes from the presence within the decondensed domain of a marker, trimethyl H3K36, which is characteristically found at multiple sites along transcribed genes (; ; ; ). Indeed, we found by ChIP that the trimethyl H3K36 mark is enhanced in the transcribed reporter sequences of the array compared with the promoter sequence. This indicates that when used in immunofluorescence, the same trimethyl H3K36 antibody should reveal the location and distribution of transcribed sequences at the array. This antibody stained a region that surrounded and extended well beyond the GFP-GR beads, suggesting that chromatin within the decondensed domain was recently transcribed. However, the BrUTP incorporation experiments demonstrate that transcription occurs only at the transcription sites directly adjacent to the GFP-GR beads (). Thus, it appears that transcribed sequences from the array do not remain at the transcription sites but instead are extruded into the surroundings, giving rise to the decondensed domain (). Together, these results suggest a model for transcription site formation at the MMTV array (). Promoter regions within the condensed domain are bound by GFP-GR, resulting in its visibility within live cells as the GFP-GR beads. Some of these GFP-GR–bound promoters then associate with pol II transcription factories. This leads to production within the pol II factory of nascent transcripts from the downstream reporter gene accompanied by deposition of the trimethyl H3K36 mark at multiple sites along the reporter gene. The transcribed sequences are extruded from the pol II factory, producing the decondensed domain and an enrichment of the trimethyl H3K36 mark in this region. The tandem nature of the gene array favors iteration of this process at consecutive promoters, thereby leading to a large decondensed domain visible by light microscopy. The MMTV array cell line was grown as previously described (). For microscopy experiments, cells were grown on #1.5 coverslips. To induce GR-mediated transcription from the MMTV array, 100 nM dexamethasone was added to cells for 0.5–1.5 h. The protocol followed that in , with the following modifications. The permeabilization buffer contained 25 μg/ml instead of 5 μg/ml digitonin, 1 mM PMSF instead of 0.5 mM PMSF, and 100 nM dexamethasone. The transcription buffer contained 10 mM MgCl instead of 5 mM, and the transcription reaction was run for 15 min at room temperature. Cells were fixed in either 3.5% PFA in PBS for 20 min followed by 0.5% Triton X-100 in PBS for 10 min or in 0.5% formaldehyde in PEM buffer (100 mM Pipes, 5 mM EGTA, 2 mM MgCl, pH 6.8, and 0.2% Triton X-100) for 5 min. The former fix tended to give more intense staining patterns for markers associated with the GFP-GR beads, whereas the latter fix tended to give more intense staining patterns for markers associated with the decondensed domain, although the pattern of staining itself was not dependent on the fixation protocol. Before antibody incubation, cells were washed three times for 10 min each in PBS. l l s w e r e f i x e d f o r 3 0 m i n b y a d d i n g a n e q u a l v o l u m e o f 7 . 0 % P F A i n P B S t o t h e D M E c u l t u r e m e d i a . I m p r o v e d s t a i n i n g w a s o f t e n a c h i e v e d w h e n t h i s f i x w a s p r e c e d e d b y a 5 - m i n p r e f i x i n 0 . 5 % f o r m a l d e h y d e i n P E M b u f f e r . C e l l s w e r e t h e n w a s h e d t h r e e t i m e s w i t h P B S f o r 1 0 m i n e a c h , p e r m e a b i l i z e d f o r 1 0 m i n w i t h 0 . 5 % T r i t o n X - 1 0 0 i n P B S , a n d w a s h e d w i t h P B S a g a i n . T h e n , c e l l s w e r e i n c u b a t e d i n 5 0 μ g / m l R N a s e f o r 3 0 – 6 0 m i n a n d w a s h e d t h r e e t i m e s i n P B S f o r 1 0 m i n e a c h . D N A w a s d e n a t u r e d b y i n c u b a t i o n a t 7 0 ° C f o r 1 0 m i n i n 7 0 % f o r m a m i d e i n 2 × S S C f o l l o w e d b y d e h y d r a t i o n f o r 2 – 5 m i n e a c h i n 7 0 , 9 0 , a n d 1 0 0 % e t h a n o l k e p t o n i c e . i s w a s i d e n t i c a l t o t h e d e c o n d e n s e d d o m a i n p r o t o c o l d e s c r i b e d i n t h e p r e v i o u s p a r a g r a p h e x c e p t t h a t c e l l s w e r e f i x e d f o r 3 0 m i n w i t h t h e 3 . 5 % P F A f i x d e s c r i b e d a b o v e , a n d d e n a t u r a t i o n w a s p e r f o r m e d f o r 5 m i n a t 9 5 ° C . #text T h e f i r s t s t a g e o f t h i s p r o t o c o l f o l l o w e d t h a t f o r t h e d e c o n d e n s e d d o m a i n – s p e c i f i c f i x a t i o n a n d d e n a t u r a t i o n p r o c e d u r e , a n d t h e d e t e c t i o n p r o t o c o l f o r t h i s f i r s t s t a g e f o l l o w e d t h a t f o r p r o b e p r e p a r a t i o n a n d h y b r i d i z a t i o n . T h e n , c e l l s w e r e f i x e d a s e c o n d t i m e f o r 1 5 – 2 0 m i n i n 3 . 5 % P F A i n P B S , w a s h e d i n P B S , p e r m e a b i l i z e d f o r 1 0 m i n i n 0 . 5 % T r i t o n X - 1 0 0 , a n d w a s h e d w i t h P B S a g a i n . D N A w a s t h e n d e n a t u r e d f o r t h e s e c o n d t i m e a c c o r d i n g t o t h e c o n d e n s e d d o m a i n – s p e c i f i c f i x a t i o n a n d d e n a t u r a t i o n p r o t o c o l . T h e s e c o n d d e t e c t i o n s t e p f o l l o w e d t h a t f o r a f o r e m e n t i o n e d p r o b e p r e p a r a t i o n a n d h y b r i d i z a t i o n . RNA FISH was performed as previously described () except that cells were fixed for 30 min with 3.5% PFA in PBS, and the hybridized probe was detected with streptavidin AlexaFluor488 (Invitrogen). a n s c r i p t i o n w a s i n d u c e d w i t h 1 0 0 n M d e x a m e t h a s o n e , a n d D R B ( C a l b i o c h e m ) w a s a d d e d s i m u l t a n e o u s l y a t 1 0 0 μ g / m l ( f r o m a 1 - m g / m l s t o c k s o l u t i o n i n w a t e r d i s s o l v e d b y h e a t i n g ) . A f t e r a 4 5 - m i n i n c u b a t i o n , t h e c e l l s w e r e p r e p a r e d f o r d e c o n d e n s e d d o m a i n – s p e c i f i c F I S H . a n s c r i p t i o n w a s i n d u c e d w i t h 1 0 0 n M d e x a m e t h a s o n e , a n d e t o p o s i d e ( S i g m a - A l d r i c h ) w a s a d d e d s i m u l t a n e o u s l y a t 2 5 0 μ M ( f r o m a 5 0 0 - m M s t o c k s o l u t i o n i n D M S O ) . A f t e r a 4 5 - m i n i n c u b a t i o n , t h e c e l l s w e r e p r e p a r e d f o r d e c o n d e n s e d d o m a i n – s p e c i f i c D N A F I S H . T h e s a m e p r o t o c o l w a s u s e d f o r R N A F I S H m e a s u r e m e n t s . All image measurements were performed with MetaMorph software (Molecular Devices). RNA FISH intensities and mean areas of structures were determined as previously described (). Fig. S1 shows immuno-FISH at 95°C with a GR antibody. Fig. S2 shows 70°C FISH controls (no probe DNA or 95°C pretreatment). Online supplemental material is available at .

Ataxia-telangiectasia (A-T) represents a paradigm for several autosomal recessive ataxias characterized by defects in the recognition and/or repair of DNA damage (). The protein defective in A-T, A-T mutated (ATM), recognizes and is activated by DNA double-strand breaks (DSBs) to signal this damage to the cell cycle checkpoints and the DNA repair machinery (). Loss of ATM function results in hypersensitivity to ionizing radiation (IR), cell cycle checkpoint defects, genome instability, increased cancer incidence, and neurodegeneration (). A-T–like disorder (A-TLD), as a result of hypomorphic mutations in the gene, most closely resembles A-T in its clinical phenotype (). Mre11 functions in a complex with Rad50 and Nbs1 (defective in Nijmegen breakage syndrome) to localize to sites of DNA DSB. This complex acts upstream of ATM in sensing DSB and ensures efficient activation of ATM (; ; ). Once activated, ATM phosphorylates a series of substrates, including Nbs1, which acts as an adaptor molecule for control of the intra-S and G2/M cell cycle checkpoints (). A third syndrome, ataxia oculomotor apraxia (AOA) type 1, also overlaps in its clinical phenotype with A-T (; ). Mutations in the gene are responsible for this neurological disorder (; ). Recent evidence shows that the protein mutated in this syndrome, aprataxin, plays a role in the repair of DNA single-strand breaks (SSBs; ; ; ), possibly by resolving abortive DNA ligation intermediates (). A distinct form of AOA linked to chromosome 9q34, AOA2 also has an overlapping clinical phenotype with the three disorders described in the previous paragraph (; ; ). This syndrome is characterized by cerebellar atrophy, oculomotor apraxia, peripheral neuropathy, and elevated serum α-fetoprotein in some cases (; ). The gene defective in AOA2, , was recently identified (). Mutations in are also associated with an autosomal dominant, juvenile onset form of amyotrophic lateral sclerosis (). Senataxin, the predicted protein encoded by is 2,677 amino acids in length and contains a seven-motif domain at its C terminus, typical of the superfamily I of DNA/RNA helicases (). Senataxin has extensive homology to the Sen1p proteins that possess helicase activity and are required for the processing of diverse RNA species that include transfer RNA, ribosomal RNA, small nuclear RNA, and small nucleolar RNA (). Sen1p proteins are also related to other DNA/RNA helicases, Upf1, involved in nonsense-mediated decay () and IGHMBP2, defective in a form of spinal muscular atrophy (). Use of global and candidate-specific two-hybrid screens identified Rpo21p, a subunit of RNA polymerase II and Rnt1p, an endoribonuclease required for RNA maturation, as a Sen1p-interacting protein (), providing further support for a role in RNA processing. Recently, showed that a single amino acid mutation that compromises Sen1 function in altered the genome-wide distribution of RNA polymerase II, providing evidence for a role in transcription regulation. Interestingly, Sen1p was also shown to interact with Rad2p, a DNase required for nucleotide excision repair after DNA damage (). These observations on yeast orthologues, together with an overlapping phenotype with other autosomal recessive ataxias with oculomotor apraxia, which are characterized by defective DNA repair, led us to investigate whether senataxin might also play a role in the DNA damage response. We show here that senataxin is primarily a nuclear protein and that AOA2 cells have increased sensitivity to HO, camptothecin (CPT), and mitomycin C (MMC), but a normal response to IR, compared with controls. To determine whether senataxin was responsible for this cellular phenotype, we cloned full-length cDNA and demonstrated complementation of agent sensitivity in stably transfected AOA2 cell lines. The increased sensitivity to HO was associated with a defect in DNA DSB repair, but there was no defect in repair of SSBs or in DSBs produced by IR exposure. These results add further substance to the hypothesis that a defective DNA damage response contributes to the neurodegenerative phenotype in a subgroup of the autosomal recessive ataxias. Cell lines were established from two patients with AOA2, fibroblasts (SETX-1RM) and a lymphoblastoid cell line (SETX-2RM). Identification of the mutations in was performed by PCR followed by DNA sequencing. The results in Fig. S1 A (available at ) show that for SETX-1RM, a homozygous deletion of 1 kb occurred at the cDNA level as a result of a large deletion in genomic DNA, which resulted in the deletion of exons 14–21 of . Mutation in was also homozygous for SETX-2RM, involving exon 23 skipping because of a missense mutation at the splice site (IVS 23 + 5 G > A; Fig. S1 B). To detect the presence of senataxin protein and investigate its subcellular localization, we produced two polyclonal antibodies against the C-terminal (Ab-1/Ab-3) and another against the N-terminal (Ab-2) regions of the protein (Fig. S2 A, available at ). Specificity of the antibodies was determined by dot blot analysis (Fig. S2, B–E). The two C-terminal antibodies reacted only with a GST fusion protein corresponding to this C-terminal region of the molecule (senataxin GST-1), whereas the N-terminal antibody specifically recognized this region of the molecule (senataxin GST-2). Preimmune serum failed to detect senataxin GST. The results in show that Ab-l detected the presence of a single prominent band at ∼300 kD for two control cell lines (C2ABR and C3ABR), two AOA1 cell lines (L990 and L938), an unclassified AOA cell line (ATL2ABR) but was absent in an AOA2 cell line (SETX-2RM), as predicted from the exon-skipping mutation. The same pattern was observed with all three antibodies (unpublished data). Immunoprecipitation with Ab-1 followed by immunoblotting with the same antibody also detected a single band of the same size, which was again absent in the AOA2 cells (). Nonspecific antisera failed to immunoprecipitate senataxin. It was not possible to detect senataxin in primary fibroblasts with up to twofold increased protein loading by immunoblotting on total cell extracts (unpublished data). However, immunoprecipitation with Ab-1 followed by immunoblotting with two senataxin antibodies (Ab-1 and Ab-2) revealed the presence of senataxin in normal foreskin fibroblasts (NFFs; ). This might be explained by a lesser amount present in the fibroblasts or a reduced relative amount of nuclear protein in fibroblasts compared with lymphoblasts. A slightly lower molecular size band for senataxin was detectable in an AOA2 fibroblast (SETX-1RM), consistent with the deletion of exons 14–21, which would maintain the reading frame but would lead to the loss of 322 amino acids (). Fractionation of cell extracts followed by immunoblotting with senataxin antibodies revealed that this protein was primarily a nuclear protein (). Senataxin was present in the cytoplasm but markedly reduced compared with that in the nucleus. It is notable that the amount of senataxin varied with cell type being highest in lymphoblastoid cells. Comparison of AOA2 (SETX-1RM) fibroblasts with NFFs (controls) using immunofluorescence demonstrated reduced nuclear staining in the AOA2 cells (). To confirm that the nuclear staining was specific, we performed competition experiments with antigen (senataxin GST-1) corresponding to a C-terminal peptide (). The results in confirm that senataxin, detected by both N- and C-terminal antibodies, is predominantly a nuclear protein and appears to be excluded from the nucleolus, as shown by a staining pattern similar to that of RNA polymerase II, a nucleoplasmic protein. Further support for this is provided by the failure of senataxin to colocalize with nucleolin in the nucleolus of NFFs or HeLa cells (). This was also the case for SETX-1RM fibroblasts, which contain a truncated form of senataxin. Finally, evidence that senataxin is predominantly a nucleoplasmic protein was provided by subcellular fractionation and immunoblotting (). Under these conditions, a prominent senataxin band was detected in the nucleoplasm but not in the nucleolus or cytoplasm. Immunoblotting for RNA polymerase II and nucleolin was performed to demonstrate the integrity of the cellular fractions (). Because it has been demonstrated that cell lines from patients with other autosomal recessive ataxias, which overlap in their clinical phenotype with AOA2, are sensitive to DNA damaging agents, we determined whether this might also be the case for AOA2. The results in reveal that AOA2 lymphoblastoid cells (SETX-2RM) show a normal pattern of sensitivity to IR but have increased sensitivity to HO, CPT, and MMC, compared with controls. AOA1 cells are also sensitive to HO, as we have previously shown (). A-T cells showed normal sensitivity to these DNA damaging agents but hypersensitivity to IR. Increased sensitivity was also observed with HO, MMC, and CPT in AOA2 fibroblasts (). Although these agents give rise to either cross-links or SSBs and DSBs in DNA, there is also evidence that their toxicity is associated with redox-related pathways (). AOA1 fibroblasts were also sensitive to HO (). As an additional measure of sensitivity, we compared the levels of HO-induced chromosome aberrations. The results in reveal approximately two induced chromosome aberrations per metaphase in HO-treated control cells, whereas a further twofold increase was observed in AOA2. To demonstrate that the increased sensitivity to HO was due to loss of senataxin, we cloned full-length cDNA into the Epstein-Barr virus–based expression vector pMEP4, containing a FLAG tag sequence, and established stable cell lines as previously described (). DNA sequencing detected five polymorphic changes in and three nonconserved amino acid changes compared with the sequence in the database (available from GenBank/EMBL/DDBJ under accession no. ). This construct, pSETX1, contains a metallothionein promoter, which allows inducible expression of senataxin by CdCl. Induction of senataxin was detected by immunoprecipitation with anti-senataxin and anti-FLAG antibodies followed by immunoblotting with the respective antibodies (, right). Immunoblotting on total cell extracts confirmed that AOA2 (SETX-2RM) parental cells lack senataxin as compared with control cells (, left). Exposure of -transfected AOA2 cells to HO after induction of senataxin corrected the HO hypersensitivity in these cells (). On the other hand, AOA2 cells transfected with empty vector remained hypersensitive to HO. cDNA also corrected the HO-induced chromosome aberrations in AOA2 cells (), whereas vector alone did not change the number of aberrations. Because AOA2 cells were sensitive to HO, CPT, and MMC, all of which are capable of generating oxidative stress (), we investigated whether there might be an inherent defect in coping with oxidative damage in these cells. The presence of 8-oxo-deoxyguanosine (8-oxo-dG) was determined as a marker of reactive oxygen species–mediated DNA damage (). The results in reveal a high basal level of 8-oxo-dG in AOA2 fibroblasts (SETX-1RM), but not in controls (NFFs). Quantitation of 8-oxo-dG fluorescence intensity reveals an approximately twofold increase in AOA2 cells compared with controls (). Exposure of cells to HO caused an increase in 8-oxo-dG in both controls and AOA2, but the extent of staining remained higher in the AOA2 cells (). To determine whether this represented more general oxidative damage in AOA2 cells or a defect at the level of DNA, we also assayed for protein nitrotyrosination and lipid peroxidation. The results showed background levels for both 3-nitrotyrosine and 4HNE-Michael adducts in AOA2 cells similar to those observed in normal fibroblasts, suggesting that these cells were not undergoing generalized oxidative stress (). Given that AOA2 cells were sensitive to agents that cause oxidative stress, and because these cells had evidence of constitutive oxidative DNA damage, we investigated their DNA repair capacity. For DNA SSB repair, we used alkaline elution analysis. Using this assay, we failed to observe any difference between control and AOA2 cells over a time course of 1-h repair (). These results were confirmed by determining intracellular NAD(P)H levels, which represents a reliable method to monitor imbalance in break repair in cells (). Again, in this case, there was no evidence for a defect in the extent or duration of DNA single-strand breakage (). HO is also capable of inducing DSBs in DNA (). To detect DSBs arising in DNA as a result of oxidative damage and eliminate those that occur as a consequence of DNA replication fork movement across a site of damage, cells were grown to confluency. DNA DSBs were detected by measuring H2AX phosphorylation visualized as foci at sites of DNA damage, a quantitative assessment for the appearance and repair of DNA DSBs (). Approximately equal numbers of γH2AX foci were detected in control and AOA2 cells between 10 and 50 min after treatment with HO (Fig. S3, available at ), and in both cases, these foci coincided with MDC1 foci, further supporting that they were sites of DNA DSB (). However, the extent of disappearance of these foci appeared to be slower in AOA2 cells at 8 h after treatment than in controls (). Quantitation of these results showed that there was a significant difference between AOA2 and controls at 4, 6, and 8 h after treatment. By 8 h after treatment, 10% of the foci remained in control nuclei compared with 39% in AOA2 cells, indicating a defect in repair of DSBs (). To determine whether this was a general defect in repair of DSBs, cells were exposed to IR, and the rate of disappearance of γH2AX foci was measured (). No difference in the rate of DSB repair after IR was observed (). The reduced capacity of AOA2 cells to repair the DSB induced by HO does not appear to be explained by a trivial reason, such as widespread oxidative damage to proteins after incubation with HO in the form of nitrotyrosinated proteins, as there was no evidence of gross damage (). To investigate further whether senataxin played a direct role in the DNA damage response, we performed transient transfections of AOA2 cells with -GFP cDNA and assayed for correction of the DNA DSB defect after HO treatment. Use of the -GFP construct allowed us to differentiate between transfected and untransfected cells, providing an internal control for scoring of γH2AX foci. The results in Fig. S4 A reveal that the number of γH2AX foci are comparable in unlabeled and GFP-labeled cells. However, by 8 h after treatment, the number of γH2AX foci in GFP-labeled cells is significantly lower than that of γH2AX foci in nontransfected cells (Fig. S4 B). Similar experiments in transfected control fibroblasts revealed that the number of γH2AX foci was the same in unlabeled and GFP-labeled cells at 30 min after treatment with HO, and both cell types efficiently repaired the DNA DSBs by 8 h (Fig. S4, C and D). Foci were counted in at least 50 GFP-transfected and unlabeled cells and quantitated to reveal that -cDNA corrected the DSB defect in AOA2 cells. These data appear in , indicating that full-length cDNA corrects the DNA DSB defect in AOA2 cells. Antibodies directed against both extremities of senataxin detected a protein of ∼300 kD, the predicted size from the reported open reading frame (). This protein was not detected in AOA2 lymphoblastoid cells with a mutation in predicted to give rise to a prematurely truncated protein, but was detected as a lower molecular size form in fibroblasts (SETX-1RM) where an in-frame deletion was observed. We also provided firm evidence that senataxin is a nuclear protein with only minimal amounts in the cytoplasm. This protein is diffusely distributed throughout the nucleus with no evidence of localization to any subnuclear compartments. This is in contrast to a recent report demonstrating that senataxin was diffusely present in the cytoplasm and in the nucleolus in cultured cells (). These data suggested that senataxin was expressed strongly in the cytoplasm in primate deep cerebellar nucleus but dull and diffuse in the nucleus. Using three different antibodies directed against both termini of senataxin, we showed diffuse nuclear labeling by immunofluoresence, which is supported by immunoblotting, revealing predominantly nuclear localization of the protein. We did not detect any senataxin in the nucleolus using either immunofluoresence or subcellular fractionation. It is of interest that , using an N-terminal FLAG-tagged construct, demonstrated only strong immunoreactivity throughout the nucleoplasm with an anti-FLAG antibody, in contrast to the results obtained with their anti-senataxin antibody. The authors interpreted this to mean that FLAG-senataxin is mislocalized. On the contrary, our data suggest that this is the correct localization. Similar to that observed in A-T, A-TLD, and AOA1 cells, AOA2 cells are also characterized by sensitivity to DNA damaging agents. A-T cells are hypersensitive to IR and radiomimetic agents that give rise to DSBs in DNA (; ). A-TLD cells are also sensitive to these agents but not to the same extent as A-T (). The pattern of sensitivity described for AOA1 cells overlaps with that reported here for AOA2 (; ). Both cell types show increased sensitivity to HO, CPT, and MMC, but are not sensitive to IR. All three agents to which AOA2 cells are sensitive can give rise to DNA SSBs, but we found no evidence for a defect in SSB repair in AOA2. On the other hand, there is some evidence that AOA1 cells have a defect in SSB repair (; ), which is compatible with a role for aprataxin in resolving abortive DNA ligation intermediates (). Thus, the basis for the sensitivity to these agents appears to differ in AOA1 and AOA2 cells. The toxicity of these agents (MMC, CPT, and HO) is also associated with redox-related pathways (). A common observation in neurodegenerative disorders associated with DNA repair/signaling defects is an increase of spontaneous oxidative damages (; ). Investigation of oxidative stress markers in AOA2 cells revealed higher basal levels of oxidative DNA damage (8-oxo-dG) compared with controls, suggesting a reduced capacity to repair this type of DNA lesion (). This is further supported by a failure to observe more general oxidative damage in these cells. At a higher concentration, HO-induced DSBs are also detected in DNA, as determined by the formation of γH2AX foci (). They described the appearance of γH2AX foci in HO-treated cells after 2 h, which gradually decreased over 24 h. Coincidence was demonstrated between γH2AX and 53BP1 foci, providing additional evidence that these were DNA DSBs. In this study, we revealed a coincidence between γH2AX and MDC1 foci, a γH2AX-interacting protein that bridges the binding of the DNA damage response machinery to sites of DNA breaks (). In the present study, these foci occurred to the same extent in both control and AOA2 cells in response to HO treatment. However, the rate of loss of γH2AX foci was significantly reduced in AOA2 cells compared with controls from 4–8 h after treatment. The presence of high constitutive levels of 8-oxo-dG in AOA2 cells did not affect the number of breaks introduced into DNA by HO, but it is possible that these lesions might interfere with the rate of repair of the DSBs. An alternative but less likely explanation for the persistence of γH2AX foci in AOA2 cells is that these foci are indicative of aberrant chromatin structure by inappropriate rejoining of DNA breaks. provided this as an explanation to account for the persistence of γH2AX foci at a time when all the breaks were repaired on mitotic chromosomes. The defect in repair to DSBs was not a general one because AOA2 cells were as efficient as controls in repairing breaks induced by IR. These results suggest that the nature of the breaks is different after HO and IR exposure. DNA damage caused by oxidation is complex and is deposited along the DNA molecule as single alterations or in clusters termed multiple DNA damage sites (). This damage is primarily oxidized base damage and, when present at close proximity on opposite strands, can give rise to abortive base excision repair that leads to the formation of DNA DSBs (). What distinguishes a DSB resulting from IR or HO treatment? There is no easy answer to this, but it is evident that a DNA DSB arising from IR is efficient in attracting the Mre11 complex, activating ATM and the components of DNA repair (; ). On the other hand, HO is less efficient in activating ATM (; unpublished data). Furthermore, PDGFβ receptor transactivation acts as an upstream mediator of ATM kinase stimulation in response to HO and, under these conditions, p53 is phosphorylated only on Ser15 (). These data suggest that the mechanism of activation of ATM by IR and HO are different and provide support for a difference in the nature of the DNA DSB generated in each case. Further support for a difference in the nature of the damage inflicted by HO and IR is demonstrated by a lack of correlation in sensitivity to these agents in different cell lines (; ; ) and difference in mutagenesis capacity () and in their effects on cell cycle progression (). Thus, it is not inconceivable that senataxin plays a specific role only in the processing of breaks generated by oxidative DNA damage. It appears from the shape of the DNA repair curve that the initial rate of removal of breaks is comparable in control and AOA2 and that the defect is in a slower component of repair. This could be explained by a defect in repairing a subgroup of DNA breaks arising as a consequence of oxidative damage. 8 h after treatment with HO, 10% of breaks remained unrepaired in control cells, whereas almost four times as many (39%) were still detected in AOA2 cells. The subgroup of breaks may relate to double-strand ends with damaged termini. Although it is likely that ·OH generated from HO is responsible for much of the damage, there is also evidence for metal-mediated DNA damage, which is not protected against with hydroxyl radical scavengers (). Although Artemis-dependent processing of a subgroup of DNA breaks in response to radiation damage has been reported (), it is unlikely that the involvement of senataxin is in the ATM–Artemis pathway because DSBs induced by IR are normally repaired in AOA2. The predicted protein sequence of senataxin contains a seven-helicase motif near its C terminus related to that present in the helicase superfamilies I and II (), which play essential roles in maintaining genome integrity through their involvement in DNA replication, transcription, recombination, and repair (). Two members of the RecQ helicase family, WRN and RECQL4, mutated in Werner syndrome and Rothmund-Thompson syndrome, respectively, also protect against oxidative DNA damage (; ). The increased sensitivity to agents that cause oxidative stress and the reduced DSB repair in response to HO exposure in AOA2 cells might also be explained by a defect in a helicase. However, although senataxin is an orthologue of yeast DNA/RNA helicases, no such activity has yet been demonstrated for this protein. A recent report by showed that the yeast Sen1 helicase controlled the genome-wide distribution of RNA polymerase II, and a mutation that affected the function of this helicase led to profound changes in polymerase II distribution over noncoding and protein-coding genes, suggesting that Sen1 has an important role in control of gene expression. Given the similarity between the yeast and human Sen1 proteins, they suggested that mutations in senataxin may also lead to misregulation of transcription and in turn account for the progressive neurological defect in this syndrome. If it were to emerge that senataxin plays a similar role to yeast Sen1 in mammalian cells, it is not immediately evident how the transcriptional misregulation might account for some of the cellular characteristics that are described here for AOA2. Mouse cells lacking the WRN helicase exhibit altered expression of genes responding to oxidative stress (). Furthermore, combining the defect with abrogation of the DNA repair gene poly (ADP-ribose) polymerase-1 (PARP-1) increased the extent of misregulation of gene expression. Aberrant transcription, chronic cellular stress, and apoptosis have also been suggested to contribute to the phenotype in Alzheimer's disease (). The appearance of oxidative stress in AOA2 as described here could thus be explained by misregulation of transcription. Failure to respond normally to DNA damage and repair DSBs arising as a result of oxidative DNA damage might also be explained by the down-regulation of specific genes that are important for this process. Alternatively, as senataxin is a large protein characterized by only one putative helicase domain, it is possible that it has other activities or controls other proteins through interaction. Being able to reconcile the AOA2 cellular characteristics described here with a possible defect in regulation of gene expression requires a greater understanding of the functioning of this protein. Lymphoblastoid cell lines (LCLs) from control (C3ABR and C2ABR), AOA1 (L938 and L939), Friedrich's ataxia (FRDA1), and AOA2 (SETX-2RM) patients were cultured in RPMI 1640 medium (Invitrogen) containing 10% FCS (JRH Biosciences), 2 mM -glutamine (Life Technologies), 100 U/ml penicillin (Invitrogen), and 100 U/ml streptomycin (Invitrogen), and maintained in a humidified incubator at 37°C/5% CO. Fibroblasts from an AOA2 patient (SETX-1RM), NFFs (a gift from P. Parsons, Queensland Institute of Medical Research, Queensland, Australia), and HeLa were cultured in DME medium (Invitrogen) containing 10% FCS. LCLs were used at a density of 1 × 10 cells/ml, and adherent cells were used at 75% confluency except in the case of γH2AX foci experiments, where confluent cells were used. MMC, CPT, and HO were purchased from Sigma-Aldrich. Irradiations were performed at room temperature using a Cs source (Gammacell 40 Exactor [MDS Nordion]; dose rate 1.1 Gy/min). Lymphoblastoid cell viability (triplicate wells for each drug concentration) was measured by adding 0.1 ml of 0.4% trypan blue to 0.5 ml of cell suspension. The number of viable cells was counted, and viabilities were expressed as the number of cells in drug-treated wells relative to cells in untreated wells (percentage of viable cells), as previously described (). The cells were incubated with the genotoxic agents for 1 h (MMC), 3 h (CPT), and 30 min (HO) before washing twice with PBS and suspension in culture medium. The number of viable cells was counted daily up to 4 d after treatment, and viability was calculated as described. The conditions were the same for fibroblasts, but survival was determined by colony formation. Cells were left for 2–3 wk to form colonies before staining with methylene blue and counting. Chromosomal aberrations were determined by treating cells with 2 mM HO for 30 min in media under aerobic conditions. Colcemid (final concentration 0.1 μg/ml) was added immediately after treatment, 2 h before harvesting. The cells were treated for 15 min in 0.0075 M KCl and fixed in methanol ± glacial acetic acid (3:1), and the fixed cells were spread onto glass slides, air-dried, and stained with Giemsa. 50 metaphases were analyzed for each sample. To produce senataxin antibodies, two regions of were PCR amplified and cloned into bacterial expression vectors. In brief, a region of 450 bp of human cDNA was PCR amplified from the 3′ end of senataxin using Ab-1F/Ab-1R primer pair. This fragment was subsequently cloned into NheI and NotI sites of pTYB1 plasmid (New England Biolabs, Inc.), containing a C-terminal chitin binding domain. A second region of 1.1 kb of was PCR amplified from the 5′ end of senataxin using Ab-2F/Ab-2R primer pair. The 5′ fragment was cloned into EcoRI and NotI sites of pGEX-5X-1 plasmid (GE Healthcare), containing an N-terminal GST tag. Both of these constructs were transformed into BL21 (DE3) pLysE cells, and overexpression was induced using 0.3 mM IPTG. To produce senataxin antibodies, two regions of corresponding to the N and C termini of the protein were PCR amplified, cloned into bacterial expression vectors, and transformed into (BL21 [DE3] pLysE). N- and C-terminal fusion proteins were affinity purified using chitin beads and glutathione–Sepharose resin, respectively, using the manufacturer's protocol. Sheep or rabbits were inoculated with senataxin antigens, and polyclonal antibodies were generated against the C (Ab-1/Ab-3) and N termini (Ab-2) of human senataxin using methods described previously (). Senataxin antibodies were affinity purified using a series of GST-only and GST-N-ter/GST-C-ter columns. Full-length was PCR amplified and cloned into KpnI–NotI digested pMEP4 (see the supplemental text, available at ). For complementation studies, expression of senataxin was induced or mock-induced from pMEP4-transfected cells with 5 μM CdCl. was also cloned into pEGFP-C2 by applying full-length cDNA from pSPORT1_Sfi (RZPD Deutsches Ressourcenzentrum). -GFP was used in transient transfection experiments. PCR primers were designed to amplify two fragments overlapping a unique SphI restriction site located at position 4432. Primer pairs HEF1/HER2 amplified a 5′ region from 1 to 4,375 bp and HEF3/HER3 amplified a 3′ region from 4,131 to 8,038 bp. PCR amplification was performed with 100 ng C3ABR cDNA, 1 μg of each primer, 100 μM dNTP, 1 × buffer 3, and 5 U (Expand high fidelity ; Roche). Thermal cycling was performed in a PE-9700 PCR machine with the following settings: 95°C for 2 min, 30 cycles of 95°C for 30 s, 65°C for 30 s, and 6 min at 72°C. After amplification, DNA were purified on agarose gels. PCR fragments were subsequently A-tailed and cloned into pGEM-T-Vector (Promega). The 4-kb 3′ senataxin fragment was excised with SphI and SphI (pGEM-T vector site) and cloned into the 5′ SphI–SphI (pGEM-T vector site) restriction sites. After screening for correct orientation, the resulting full-length cDNA clone was called pSETX. For expression of senataxin, the full-length senataxin cDNA was amplified from pSETX with Pfu Turbot (Stratagene) using HEF5/HER5 primer pair, and the resulting product was cloned into KpnI–NotI digested pMEP4. The resulting pMEP4- construct is referred to as pSETX1. Cells on coverslips were fixed with 100% prechilled methanol for 5 min and immersed in 100% prechilled acetone for 5 min. Coverslips were subsequently air-dried, treated with 0.05 N HCl for 5 min on ice, and washed three times with PBS. RNA was digested by incubating the coverslips in 100 μg/ml RNase in 150 mM NaCl with 15 mM sodium citrate for 1 h at 37°C. After RNA digestion, coverslips were sequentially washed in PBS, 35, 50, and 75% ethanol for 2 min each. DNA was denatured by incubating the coverslips with 0.15 N NaOH in 70% ethanol for 4 min. A series of washes was performed starting with 70% ethanol containing 4% vol/vol formaldehyde and then 50% ethanol, 35% ethanol, and finally PBS for 2 min each. Proteins were digested with 5 μg/ml proteinase K in TE, pH 7.5, for 10 min at 37°C. After several PBS washes, coverslips were incubated with anti-8-oxo-dG antibody (1:250; 4355-MC-100 [Trevigen]) in PBT20 (1× PBS/1% BSA/0.1% Tween 20) for 1 h at room temperature. After several washes with 0.1× PBS, 8-oxo-dG was detected using an Alexa Fluor 488 secondary antibody (1:500 in PBT20; Invitrogen). Nuclei were counterstained with DAPI, and slides were mounted for immunofluorescence. Cells were separated into cytoplasmic, nucleoplasmic, and nucleolar fractions as previously described (). For immunoprecipitations, cells were washed in PBS and resuspended in lysis buffer (50 mM Tris-HCl, pH 7.5, 50 mM β-glycerophosphate, 150 mM NaCl, 10% glycerol, and 1% Tween 20, supplemented with protease and phosphatase inhibitors) for 1 h at 4°C. Insoluble components were removed by centrifugation at 16,000 for 20 min at 4°C. For immunoprecipitations, 1 mg of total cell extract was precleared with 30 μl of protein G–Sepharose beads (GE Healthcare) for 3 h at 4°C. Senataxin was immunoprecipitated with 7.5 μg senataxin Ab-1 antibody overnight at 4°C. The following day, 40 μl of protein G–Sepharose beads was added for 1 h and incubated at 4°C. Immunoprecipitates were washed three times with lysis buffer and resuspended in 20 μl of sample loading buffer before separation of the proteins by SDS-PAGE. The proteins were then transferred onto a nitrocellulose membrane (Pall Life Sciences), and immunoblots were performed with the relevant antibodies. NFFs, AOA2 (SETX-1RM) fibroblasts, and HeLa were grown on glass coverslips for 48 h, washed with PBS, fixed in 2% paraformaldehyde/PBS for 10 min, and processed for immunofluorescence as previously described () using the relevant antibodies, senataxin Ab-3 (1:400) and Ab-2 (1:200), RNA polymerase II (1:400; ab5408 [Abcam]), nucleolin (1:500; M019-3 [MBL International Corporation]), 3-nitrotyrosine (1:100; 9691 [Cell Signaling Technology]), 4-HNE-Michael adducts (1:100; 393207 [Calbiochem]). 8-oxo-dG was detected according to the manufacturer's protocol. Fluorochromes conjugated to the relevant secondary antibodies were Alexa Fluor 488 and 594 (Invitrogen). Images were captured using a digital camera (AxioCam MRm; Carl Zeiss MicroImaging, Inc.) attached to a fluorescent microscope (Axioskop2 mot plus; Carl Zeiss MicroImaging, Inc.) using Plan Apochromat 1.4 oil differential interference contrast (63× magnification). Imaging medium was PBS, and acquisition was performed at ambient temperature (25°C). AxioVision LE 4.3 software (Carl Zeiss MicroImaging, Inc.) was used to capture the individual images, which were assembled using Photoshop 7.0 (Adobe). Fluorescence intensity was quantitated on the RAW images using the public domain software ImageJ version 1.34s (NIH) before their assembly in Photoshop 7.0. After assembly, contrast was enhanced on all images simultaneously in Photoshop 7.0 using the brightness and contrast tool. No further image processing (e.g., surface or volume rendering, γ adjustment) was performed. To induce SSB, cells were exposed to 20 μM HO in RPMI 1640 medium without supplements for 15 min at 37°C. To remove HO, 880 U/ml catalase was added. Cells were collected by centrifugation (1,500 U/min for 5 min) and washed with PBSCMF (140 mM NaCl, 3 mM KCl, 8 mM NaHPO, and 1 mM KHPO). The numbers of SSBs were determined by an alkaline elution assay as previously described (). The numbers of SSBs in untreated control cells were subtracted in all cases. NAD(P)H depletion after HO and MMS treatments was performed as previously described (). Cells were seeded onto coverslips, and experiments were performed on confluent fibroblasts. Cells were either irradiated or treated with HO by adding 2 mM HO into the growth medium for 10 or 30 min. Cells were washed with PBS and returned to fresh media. At the time points indicated, the cells were processed for immunofluorescence as described. Images were captured and assembled as described. Fig. S1 shows characterization of mutations in two AOA2 cell lines. Fig. S2 shows characterization of senataxin antibodies. Fig. S3 shows induction of γH2AX foci in control (NFF) and AOA2 (SETX-1RM) cells after exposure to 2 mM HO. Fig. S4 shows complementation of the DNA DSB repair defect in AOA2 cells. The supplemental text gives primers used in this study. Online supplemental material is available at .

The spindle of eukaryotic cells is a complex microtubule (MT)-based machine that segregates chromosomes in mitosis and meiosis. Shortly before or at the beginning of mitosis, spindle MTs are nucleated from tubulin subunits either at the MT-organizing center or near chromatin (). The MT-organizing center is known as the centrosome in higher eukaryotes or as the spindle pole body (SPB) in yeast. The MTs then assemble through the action of MT-associated proteins into a bipolar spindle. The mitotic spindle contains two distinct sets of MTs: the pole–kinetochore and the pole–pole MTs. The minus ends of both groups of MTs reside at the SPBs. The pole–kinetochore MTs interact at their plus end with kinetochores and move chromosomes to the spindle poles (anaphase A). The pole–pole MTs interdigitate in the middle of the spindle, thereby defining a spatially restricted region known as the central spindle or spindle midzone, and segregate chromosomes by elongating the spindle (anaphase B). The properties of MTs change dramatically as cells transit the cell cycle. Through the rise of cyclin-dependent kinase (Cdk) activity, MT turnover increases as cells enter metaphase of mitosis (; ). Increased MT dynamics helps to reorganize the MT cytoskeleton into a bipolar spindle and promotes chromosome capture by kinetochore MTs (). With anaphase onset, MTs suddenly become stabilized (; ; ). This is the combined result of decreased Cdk activity and the activation of protein phosphatases. A protein phosphatase that has been implicated in the regulation of anaphase spindle properties is the conserved Cdc14 (; ). Cdc14 is involved in the dephosphorylation of spindle-associated proteins such as the DASH component Ask1 () and directly regulates MT-binding activity of the inner centromere protein–Aurora B complex (Sli15-Ipl1 in yeast), which then, in turn, controls spindle localization of the chromosomal passenger protein Slk19 (). An additional level of complexity arises from the targeting of a subset of spindle-associated proteins such as the MT-bundling protein Ase1, the Aurora B kinase complex, kinesin motor proteins, and, in yeast, the chromosomal passenger proteins Slk19 and separase Esp1 to the spindle midzone at the beginning of anaphase, where they participate in anaphase spindle formation and stabilization (; ; ; ; ). Ase1 belongs to a functionally conserved family of MT-associated proteins () named Ase1 in fission yeast (; ), Feo in (), PRC1 in human cells (), SPD-1 in (), and MAP65 in plant cells (). Although the degree of sequence identity is low, all family members bundle MTs, display specific midzone localization, and participate in anaphase spindle stability and cytokinesis (; ; ; ; ). For example, human cells depleted of PRC1 by siRNA or yeast Δ cells assemble a bipolar spindle, but severe defects in spindle morphology and function arise upon passage into anaphase (; ; ; ). The ability of Ase1 to preferentially bundle antiparallel MT arrays makes it a key player in spindle midzone assembly (). However, the cell cycle signals that target Ase1 to the spindle midzone at the beginning of anaphase and the identity of the proteins that cooperate with Ase1 in midzone assembly are barely understood. To identify proteins involved in these processes, we performed a systematic analysis of spindle midzone components in the model organism budding yeast. We show that Ase1 acts together with the separase–Slk19 complex to establish a functional spindle midzone independently of MT-based motor proteins. Cdc14 directly regulates spindle midzone assembly through the dephosphorylation of Ase1 and influences midzone centering indirectly via the separase–Slk19 complex. Phosphorylation of Ase1 in metaphase prevents the hyperactivation of spindle extension and spindle breakage during anaphase. Failure to dephosphorylate Ase1 with anaphase onset results in the delocalization of spindle midzone proteins along the entire length of the anaphase spindle and impairs the second, slower phase of anaphase spindle extension. The spindle midzone of budding yeast contains the MT-bundling protein Ase1 (), separase Esp1 (), the Esp1 interactor Slk19 (; ), the MT plus end–binding proteins Bim1 and Bik1 (; ), the Cin8 and Kip3 kinesin-like motor proteins (; ), and the CLIP-associating protein–like molecule Stu1 (). Some of these proteins (Ase1, Bik1, Bim1, Cin8, Kip3, and Stu1) are already associated with the metaphase spindle and become focused into a discrete zone between the two spindle poles with anaphase onset. In contrast, Esp1 and Slk19 are only recruited to the spindle after the onset of anaphase. We may expect that proteins that perform a leading function in spindle midzone assembly bind earlier than others that execute a more assisting role. To gain insight into the formation of a functional spindle midzone, we compared the timing with which Ase1 associated with the spindle midzone to that of Bim1 and Slk19 association by time-lapse microscopy ( and Videos 1 and 2, available at ). We first analyzed the behavior of Ase1 and Slk19 using cells in which the chromosomal copy of was fused to GFP () and was fused to the red fluorescent tandem-dimer Tomato ([tdTomato] ). The and gene fusions were functional, as the elongated anaphase spindles of cells were stable (unpublished data and ). We confirmed that Ase1 was already bound to the short spindle of preanaphase cells (, t = 0; ). In these cells, Slk19 associated with kinetochores, which, in yeast, cluster close to the SPBs (). With anaphase onset, Ase1 accumulated at the spindle midzone between the two poles marked by the Slk19-tdTomato kinetochore signal (, min 40 s; arrow). Slk19 localized slightly later than Ase1 to the emerging spindle midzone (, min 30 s). Moreover, Slk19 left the spindle at the end of anaphase before Ase1 (, 15 min 0 s). These data were confirmed using still images of synchronized cells (Fig. S1). Thus, Ase1 binds to the assembling spindle midzone before Slk19. Analysis of cells ( fused to the red fluorescent eqFP611 from the sea anemone ; and S1; ) showed that Bim1 associated with the spindle midzone with kinetics identical to Ase1. Next, the spindle localization of Esp1 and Slk19 was determined in cells, in which the integrated was expressed from the native promoter. Because time-lapse analysis was challenging as a result of the weak Esp1-GFP signal, the localization of Esp1 and Slk19 was determined in still images (z stacks) of α-factor synchronized cells. The Esp1-GFP signal appeared with anaphase onset at both spindle poles, where it colocalized with Slk19-tdTomato (, i; arrowheads). When the spindle length was between 2 and 6 μm, Esp1 and Slk19 also colocalized at the spindle midzone (). Surprisingly, Esp1 left the spindle midzone before Slk19 (, >6 μm). Thus, Esp1 and Slk19 probably bind to the developing spindle midzone as a complex (Esp1–Slk19 complex formation has been demonstrated previously; ) but later dissociate from the midzone with distinct kinetics. The notion that Esp1 and Slk19 bind as a complex was further supported by the finding of the interdependency of the spindle midzone binding of Esp1 and Slk19 (Fig. S2, available at ). Collectively, proteins bind to and leave the spindle midzone in a defined order, indicating a cell cycle–controlled program of spindle midzone assembly and disassembly. Ase1 by itself or together with other proteins may function as a landmark for spindle midzone assembly to which other spindle midzone proteins bind either simultaneously or later than Ase1. To test this model, we analyzed the interdependency of the localization of spindle midzone proteins. Slk19 localization was analyzed in cells in the presence and absence of (). enabled the visualization of the red fluorescent spindle. Cells were synchronized with α factor in G1 phase of the cell cycle (t = 0). Around 40–60 min after release from the G1 block, a short bipolar preanaphase spindle of similar length assembled in and Δ cells (). At ∼60–80 min, and Δ cells entered anaphase, as indicated by the increase in the proportion of large-budded cells and spindle elongation (). In wild-type (WT) cells, Slk19 associated with the midzone of spindles of intermediate length (3–8 μm; , A and B; red dots). In contrast, the anaphase spindles of Δ cells were always devoid of Slk19 (). The reduced spindle length of Δ cells did not account for this defect because the anaphase spindles of Δ cells exceeded the 3-μm threshold after which Slk19 associates with spindles in WT cells ( and ). The spindle pole localization of Slk19 was not affected by the deletion of . Similar data were obtained for Esp1-GFP (Fig. S2). Thus, the Esp1–Slk19 complex requires the Ase1 protein or an Ase1-dependent structure for binding to the spindle midzone. We next tested whether the Esp1–Slk19 complex controls the spindle localization of Ase1. This analysis was performed with Δ cells carrying allowed the unequivocal localization of the SPBs. In Δ cells, as in WT cells, Ase1 bound only to a section of the anaphase spindle (). However, the Ase1-GFP zone of Δ cells was more extended than in WT cells (). In addition, in 22% of Δ cells, the Ase1-GFP signal was shifted toward one of the spindle poles (). A similar mislocalization of Ase1 was observed in conditional lethal cells (Fig. S2; ). Together, these data indicate that the Esp1–Slk19 complex is important for both the spatial restriction and the centered localization of the spindle midzone protein Ase1. Further analysis of the interdependency of spindle proteins showed that all midzone components (Bik1, Bim1, Cin8, Kip3, and Stu1) required the Ase1 landmark for recruitment to the spindle midzone and the Slk19 protein for spatial constraint to a centered spindle domain once recruited there by Ase1 (Fig. S3, available at ). In the absence of , these spindle midzone components frequently formed spatially restricted, albeit not centered, spindle domains. On the other hand, the deletion of abolished domain formation with the proteins binding uniformly along the interpolar MTs (Fig. S3). In contrast, the localization of other spindle components that evenly decorate WT anaphase spindles, such as Sli15, Ipl1, or Ndc10 (; ), was not affected by the deletion of either or (Fig. S3). Thus, only the localization of spindle midzone proteins is dependent on Ase1 and the Esp1–Slk19 complex. Finally, we asked whether the inactivation of other spindle components (, , , , , , , , , , , , and ) affected the localization of either Ase1 or the Esp1–Slk19 complex (Fig. S4, available at ). With the exception of , , and mutants that impaired Slk19 localization (; ), we found no defects in the localization of Ase1 and Esp1–Slk19 in any mutant (Fig. S4 and not depicted). This was also the case in the multiple kinesin motor mutant Δ Δ (Fig. S4; ). The latter result was surprising because in mammalian cells, PRC1 requires a kinesin motor for correct localization (). Esp1 and Slk19 are part of the Cdc14 early anaphase release network (FEAR) that activates the Cdc14 phosphatase at the beginning of anaphase (). The activated Cdc14 then regulates the spindle localization of Sli15–Ipl1, Cin8, Slk19, and Stu1 (; ). The mislocalization of Ase1 in Δ or cells that we report here could therefore arise as an indirect consequence of the failure to activate Cdc14. However, several data argue for a direct function of the Esp1–Slk19 complex in the control of spindle midzone assembly. -induced activation of () had a marginal impact on the length of the Ase1-GFP spindle domain of cells, as it only reduced it from 3.1 to 2.5 μm (, spindles of 7–8 μm), whereas in WT cells, the Ase1 spindle domain was 1.4 μm long (Fig. S2, spindles of 7–8 μm). Moreover, activation of also had a minimal impact on the mislocalization of Ase1-GFP in cells (). These data demonstrate a direct role of the Esp1–Slk19 complex in Ase1 localization. Consistently, in cells, which mislocalize spindle-associated Slk19 () but release Cdc14 as well as WT cells, Ase1 bound to spindles as in Δ cells (unpublished data). The spindle midzone localization of Cin8, Slk19, and Stu1 is severely disturbed in cells (; ). Slk19 mislocalization is caused by the failure of the Sli15–Ipl1 kinase complex to bind to spindle MTs in anaphase. The expression of , which binds constitutively to spindle MTs, is therefore partly able to suppress the localization defect of Slk19 in cells (). To further analyze the functions of Cdc14 and the Sli15–Ipl1 kinase complex in spindle midzone assembly, we determined the localization of the midzone proteins Ase1, Cin8, Stu1, Bim1, and Esp1 in cells with and without the expression of . In addition, the role of the Sli15–Ipl1 kinase complex was tested by analyzing the localization of these midzone proteins in conditional lethal and cells. In cells, Esp1 failed to concentrate between the two spindle poles. Instead, Esp1 showed a weak, relatively uniform MT-like staining (). (), suggesting that the Sli15–Ipl1 pathway regulates Esp1. Consistently, Esp1 failed to bind to spindles in cells (). Thus, Cdc14 regulates Esp1 localization through the Sli15–Ipl1 kinase complex. In cells, Ase1 was either distributed along the entire anaphase spindle (, 59%) or the Ase1 zone was shifted toward one spindle pole (, ii; and C, 32%). This result was confirmed with temperature-sensitive degron cells (; see ), in which Cdc14 was rapidly degraded upon shifting the cells to the restrictive temperature (). To obtain an understanding of whether the regulation of Ase1 was via the Sli15–Ipl1 pathway, we analyzed Ase1 spindle positioning in , and cells. was unable to suppress the Ase1 localization defect of cells (, iv–vi; and C), arguing against a major role for Sli15–Ipl1 in Ase1 regulation. In addition, the defect in Ase1 localization was much stronger in than in or cells (). This suggests that Cdc14 regulates the midzone localization of Ase1. However, contrary to Esp1 and Slk19, the Sli15–Ipl1 complex has little direct influence on Ase1 localization. Cin8-4GFP, Stu1-4GFP, and Bim1-4GFP localized to spindle MTs in cells () but failed to accumulate in the middle of the spindle, as they did in WT cells (). The expression of in cells did not restore the centered localization of Cin8-4GFP, Stu1-4GFP, and Bim1-GFP (), suggesting that the Sli15–Ipl1 complex is not directly involved in the targeting of these proteins to the midzone. The finding that in cells, Cin8-4GFP, Stu1-4GFP, and Bim1-GFP still bound in a spatially restricted manner to the anaphase spindle () reinforced this view. However, in cells, the Cin8, Stu1, and Bim1 zones were broader than in WT cells and were frequently shifted toward one of the spindle poles (). The extent of this defect was comparable with that seen in Δ cells (Fig. S3). The defective localization of Slk19 in cells () is probably responsible for the mislocalization of Cin8, Stu1, and Bim1. Furthermore, the mislocalization of all tested midzone components both in and mutants and the dependency of Ase1 spindle localization on Cdc14 raises the possibility that Ase1 is a direct target of Cdc14. Ase1 is phosphorylated by Cdk1-Clb5 at the end of S phase, dephosphorylated during anaphase, subsequently targeted for destruction by the anaphase-promoting complex ([APC] APC)/proteasome system during mitotic exit, and reaccumulates again in S phase (; ). Moreover, Ase1 is a key protein required for the correct localization of all midzone components (Fig. S3), binds to the developing midzone before the Esp1–Slk19 centering complex (), and is mislocalized in the mutant (). On this basis, we reasoned that Ase1 may be a direct target of Cdc14, allowing the coordination of spindle midzone assembly with anaphase onset. Such a model predicts an interaction between Ase1 and Cdc14. Using the yeast two-hybrid system, we identified an interaction between Ase1 and an N-terminal fragment of Cdc14 (). If Cdc14 dephosphorylates Ase1, we would expect to see an accumulation of hyperphosphorylated Ase1 in cells lacking Cdc14 activity. The phosphorylation of Ase1 was therefore analyzed in α-factor synchronized WT and cells (). The Ase1-6HA protein accumulated in both cell types around 70 min after release from the G1 block. In and cells, a fraction of Ase1-6HA became hyperphosphorylated after ∼80 min, as indicated by the accumulation of slower migrating Ase1 phosphoisoforms (, Ase1-P). In WT cells, Ase1-6HA became dephosphorylated and was then degraded with mitotic exit (Clb2 degradation and Sic1 accumulation; , 120 min). In contrast, in cells, Ase1-6HA remained in the hyperphosphorylated form (, 120–150 min). Thus, the dephosphorylation of Ase1 is dependent on Cdc14. If the proposed model is correct, the premature activation of Cdc14 should dephosphorylate Ase1 at a point in the cell cycle when Cdc14 is normally inactive. This possibility was tested in -Δ - and -Δ - cells, which lack the APC subunit Cdc26 that is only essential at 37°C (). When grown in the presence of methionine at 37°C, these cells arrested in metaphase without APC activity because of Cdc20 depletion and the absence of Cdc26. In these arrested cells, the endogenous Cdc14 was entrapped in the nucleolus. The promoter was then induced by the addition of galactose, leading to the accumulation of high levels of Cdc14 or the inactive Cdc14 in the nucleus and cytoplasm ( and not depicted). Despite this accumulation of active Cdc14, cells remained in metaphase (, top) mainly because the APC was inactive, and, therefore, Clb2 levels remained high (, lane 4). The slight accumulation of the Cdk1 inhibitor Sic1, which was induced by Cdc14 overexpression, was insufficient to promote mitotic exit, as indicated by the failure of cells to rebud or to replicate the DNA (; ). Importantly, -activated Cdc14 caused the collapse of most of the hyperphosphorylated Ase1 (Ase1-P) into the nonphosphorylated, faster migrating Ase1 band (, lane 4). This was not observed when the phosphatase-dead was expressed (, lane 6). Collectively, Cdc14 is responsible for the dephosphorylation of Ase1 in vivo. Finally, we asked whether recombinant Cdc14 was able to dephosphorylate affinity-purified Ase1 from yeast cells. A first attempt to isolate phosphorylated Ase1 from metaphase-arrested cells failed probably because Cdc14 that was released from the nucleolus during the purification procedure dephosphorylated Ase1 (unpublished data). Hyperphosphorylated Ase1-6HA (Ase1-P) was therefore obtained through immunoprecipitation from arrested cells, which lack Cdc14 activity. Enriched Ase1-P was then incubated with buffer only, recombinant Cdc14, the phosphatase-dead Cdc14, or, as a positive control, with AP (). Only Cdc14 and AP were able to dephosphorylate Ase1-P, as indicated by the mobility shift. Thus, Cdc14 directly dephosphorylates Ase1. Ase1 contains seven Cdk1 consensus sites ([ST]-P-X-[RK]; ). The serine or threonine residues within these sites were either converted to alanine () to prevent phosphorylation or to aspartic acid () to mimic constitutive phosphorylation. The Ase1 and Ase1 proteins no longer showed the mobility shift characteristic for the hyperphosphorylated Ase1 (), indicating that most Ase1 phosphorylation sites that were normally responsible for the band shift were blocked. To understand the relevance of the phosphorylation/dephosphorylation cycle of Ase1, we analyzed the spindle localization of Ase1 and Ase1 in and cells carrying the SPB marker Spc42-eqFP. In WT cells, the spindle localization of Ase1-GFP and Ase1-GFP was similar (). In contrast, Ase1-GFP mislocalized in the majority of WT cells: the Ase1-GFP zone either covered large portions of the anaphase spindle or was shifted toward one of the two spindle poles (). This mislocalization may arise because Ase1 behaves as a constitutively phosphorylated protein or because the mutations unspecifically affect the function of the protein. In the first case, Ase1 and Ase1 should show similar localization in cells in which Ase1 is hyperphosphorylated (). Indeed, in cells, Ase1-GFP and the hyperphosphorylated Ase1-GFP showed nearly identical localization patterns (, ). Thus, the dephosphorylation of Ase1 is important to assemble a focused spindle midzone. If the dephosphorylation of Ase1 is an essential step in spindle midzone assembly, Ase1 should, in part, suppress the spindle defect of cells. Indeed, in cells, the nonphosphorylated Ase1-GFP localized more to the middle of the anaphase spindle than the phosphorylated Ase1-GFP or Ase1-GFP (, purple area). The failure of Ase1 to show a WT localization in cells is probably the result of the mislocalization of Slk19 in these cells (), which, in turn, affects Ase1 localization (). Mislocalization of Ase1 in cells may affect spindle binding of other midzone proteins. This is expected because the localization of all midzone components depends on ( and S3). Analysis of Cin8-GFP, Stu1-GFP, and Slk19-GFP showed that spindle midzone localization of these proteins was severely impaired in cells (). Although Cin8 and Stu1 still bound to the spindle MTs, Slk19 was observed in all anaphase cells at the spindle poles but not along the spindle ( > 100). Collectively, these data indicate that dephosphorylation of Ase1 by Cdc14 is essential to assemble a focused spindle midzone. Analysis of anaphase B in yeast suggests that the spindle elongates in two phases (). In the first step, the completely overlapping MTs of the metaphase spindle slide apart. This relatively fast elongation is limited to only 1–2 μm. The subsequent slower elongation step of up to 8 μm requires coupling of the sliding of antiparallel MTs with the polymerization of MT plus ends to maintain an overlap zone. To understand the role of and of a focused midzone in spindle elongation in greater detail, we analyzed the kinetics of anaphase spindle extension of , Δ, , and cells by time-lapse microscopy. The initial fast phase of anaphase spindle extension was identical in all cell types (, , and Videos 3–6; available at ), indicating that Ase1 is not required for this step. Clear differences between WT and the mutants were observed for the second phase. In Δ cells, the second extension phase did not occur (). cells extended the anaphase spindle with close to double the speed of cells ( and ). In addition, in 15% of cells, the spindle broke during extension. However, this fracture was not permanent, as the spindle subsequently reformed and extension resumed (, t = 9 min; and Video 7, available at ). Such spindle breakage was not observed in WT cells. cells showed a mixed phenotype. 44% of cells (12/27) behaved in a manner that was similar to WT cells. This probably reflects the fact that Ase1 distribution is similar to that of the WT Ase1 molecule in 31% of cells (). Importantly, in 56% of cells (15/27), anaphase spindle extension stalled for up to 10 min after the initial fast phase. After this period of inactivity, the spindle suddenly extended with close to normal speed. The duration of this elongation was reduced, and spindles were shorter than in WT cells ( and ). Thus, a focused spindle midzone is required for processive extension of the anaphase spindle. has an important function in stabilizing the anaphase spindle. This requirement for becomes especially apparent when cells are arrested in anaphase (; ). In 87% of Δ cells that had been arrested for 3 h, the anaphase spindles broke, whereas <18% of cells showed broken anaphase spindles (). Consistent with the spindle breakage that we observed in the time-lapse experiments (), a large number of cells had broken anaphase spindles (73% after 3 h). This was in contrast to cells, which maintained the anaphase spindle integrity to a similar degree as WT cells (). These data suggest that phosphorylation of Ase1 before anaphase onset is important for spindle stability in anaphase. #text Gene deletions and epitope tagging of genes at their endogenous loci were performed using PCR-based methods (). cells were constructed and grown as described previously (). The strains and plasmids used in this study are listed in Table S1 (available at ). All yeast strains were derivatives of S228c with the exception of Δ (referred to as in the text and figures), which was derived from W303 and was compared with the corresponding WT, K699. Typically, cells were grown in yeast extract peptone glucose medium (YPD) at 23°C and shifted to 30°C or the restrictive temperature for 3 h before observation. For synchronization, cells were incubated with 10 μg/ml of synthetic α factor for 2.5–3 h at 23°C until >95% of cells were in G1 phase. After washing with prewarmed medium to remove α factor, cells progressed synchronously through the cell cycle. Cells were arrested in metaphase by depletion of under control of the promoter by incubating the cells for 3–4 h in yeast extract peptone raffinose (YPR) medium supplemented with 2 mM methionine and 2 mM cysteine until >95% of cells were with a large bud. The APC was inactivated by incubating arrested metaphase cells at 37°C. and were expressed from the promoter cloned into yeast integration plasmids. The promoter was induced by the addition of 2% galactose to the grow medium. was expressed from the native promoter cloned into the yeast integration plasmid pRS406. with regulatory and coding regions was cloned into the yeast integration vector pRS306. Mutations in were introduced by PCR-directed mutagenesis and confirmed by DNA sequencing. Serine or threonine residues of seven Cdk1 consensus sites ([ST]-P-X-[KR]) were mutated to alanine to avoid phosphorylation or mutated to aspartic acid to mimic phosphorylation (). The mutant resulted from the exchange of threonine 55, serine 64, serine 198, threonine 676, serine 707, serine 803, and serine 819 to alanine. In , the same set of amino acids was exchanged to aspartic acid. , , and (N-terminal fragment; amino acids 1–352) were cloned into pMM5 and pMM6 vectors. Interaction between Sli15 and Cdc14-N was used as a positive control (). Two-hybrid interactions were tested as described previously (). For in vitro dephosphorylation assay, Ase1-6HA was immunoprecipitated from a cell extract. Immunoprecipitates were incubated with buffer, maltose-binding protein–Cdc14, or maltose-binding protein–Cdc14 (both purified from ) for 1 h at 30°C as described previously (). Proteins were analyzed by immunoblotting with anti-HA antibodies. Yeast extracts were prepared using alkaline lysis and TCA precipitation (). Anti-Cdc14 (6His-Cdc14), anti-Clb2 (GST-CLB2), anti-GFP (GST-GFP), anti-Pds1 (GST-Pds1), and anti-Tub2 antibodies (GST-Tub2) were prepared in rabbits or sheep against purified recombinant proteins (). Monoclonal mouse anti-myc (9E10) and anti-HA (12CA5) antibodies were purchased from Roche, and guinea pig anti-Sic1 antibodies were a gift from G. Pereira (German Cancer Research Centre, Heidelberg, Germany). For live cell imaging ( and ), cells were adhered with concanavalin A on small glass-bottom Petri dishes (MatTek). Imaging was performed at 30°C on a microscope (DeltaVision RT; Applied Precision) equipped with GFP and TRITC filters (Chroma Technology Corp.), a plan Apo 100× NA 1.4 oil immersion objective (IX70; Olympus), softWoRx software (Applied Precision), and a camera (CoolSNAP HQ; Photometrics). Measurements were performed using softWoRx software and complete z stacks. Fig. S1 shows the spindle binding of Ase1 relative to Bim1 and Slk19 by analysis of still images of synchronized cells. Fig. S2 shows interdependency of the midzone localization of Ase1, Esp1, and Slk19. Fig. S3 shows the localization of spindle proteins in Δ and Δ cells. Fig. S4 shows Ase1 localization in spindle mutants. Table S1 lists strains and plasmids used in this study. Video 1 shows the spindle localization of Ase1 relative to Slk19 in an cell corresponding to A. Video 2 shows spindle localization of Ase1 relative to Bim1 in an cell corresponding to B. Videos 3–6 show spindle extension in , Δ, , and cells corresponding to A, respectively. Video 7 shows spindle breakage during extension in an cell corresponding to B. Online supplemental material is available at .

Chromosome segregation during mitosis depends on the action of the spindle, a protein machine that uses ensembles of kinesin and dynein motors plus microtubule (MT) dynamics to move chromatids polewards (anaphase A) and to elongate the spindle (anaphase B; ; ; ). Spindle MTs display “poleward flux,” a form of MT dynamics in which tubulin subunits within the MT polymer lattice translocate persistently poleward while their minus ends are depolymerized at the poles (). In addition, prometaphase spindle MTs use dynamic instability to search for chromosomes and then capture and align them on the spindle equator (; ). However, in many spindles, MTs suddenly become stable at the onset of anaphase (; ; ; ). In budding yeast, the suppression of MT dynamics is regulated by the cell cycle–regulated Cdc14 phosphatase and is essential for proper chromosome segregation, as loss of MT stabilization at anaphase onset leads to defects in both anaphase A and B (). In the syncytial blastoderm stage embryo, highly dynamic MTs drive remarkably rapid movements of chromosomes and spindle poles, at rates typically of ∼0.1 μm s (; ; ; ). In preanaphase B (metaphase and anaphase A) spindles, it is proposed that a kinesin-5 (KLP61F)–driven interpolar MT (ipMT) sliding filament mechanism is balanced by kinesin-13 (KLP10A)–dependent ipMT depolymerization at the poles to maintain the spindle at a steady-state length while simultaneously driving poleward flux within ipMTs. Once chromatid-to-pole motion is essentially complete, anaphase B is triggered by the suppression of kinesin-13–dependent depolymerization, which allows persistently sliding ipMTs to exert forces that drive spindle pole separation (; ; ). Here, therefore, one function of poleward flux is to constrain the length of preanaphase B spindles, and its down-regulation permits spindle elongation. Surprisingly, the ipMTs that drive anaphase spindle elongation in embryos are highly dynamic, displaying a turnover half-time of ∼5 s in FRAP experiments (). Quantitative modeling using systems of force balance and rate equations suggests that this rapid rate of MT turnover is due to the dynamic instability of ipMT plus ends and demonstrates that such dynamic ipMTs are capable of driving steady, linear pole–pole separation at ∼0.1 μm/s (). However, it is not known if the rapid dynamics is a property of MTs at all stages of mitosis in and if these spindle MTs, like those of other systems, undergo stabilization at anaphase B onset. Here, we have systematically evaluated the dynamic properties of spindle MTs throughout mitosis in embryos using FRAP of fluorescent GFP-tubulin in conjunction with time-lapse fluorescence microscopy of EB1-GFP to mark the growing MT plus ends (; ). Our studies show that before anaphase B, spindle MTs turn over rapidly and display a uniform plus-end distribution, but at anaphase B onset, in a process that requires cyclin B degradation, a stable subset of MTs develops as MT plus ends specifically redistribute into the central spindle region at the expense of MTs that depolymerize near the poles. We used quantitative modeling to investigate the dynamic parameters that could account for the full and rapid turnover of MT plus ends that are uniformly distributed throughout the preanaphase B spindle and the changes in dynamics that could produce the spatial reorganization of MTs that occurs at anaphase B onset. The results illuminate a mechanism by which a spatial change in spindle MT dynamics may redistribute MT plus ends to facilitate anaphase B spindle elongation. Using FRAP analysis of GFP-tubulin, we find that embryo mitotic spindles turn over at an extremely rapid rate (half-time of 5–10 s) and recover almost completely during preanaphase B (i.e., the metaphase–anaphase A steady state; , , Fig. S1 A; and Videos 1 and 2, available at ) in accordance with our previous analysis of the equatorial region of anaphase B spindles (). These preanaphase B spindle MTs could plausibly turn over by dynamic instability of their plus ends and/or by poleward flux (; ; ; ; ). The rate of poleward flux in these spindles (0.05 μm/s) on its own is too slow to account for the fast FRAP recovery, especially within the large bleach regions of preanaphase B spindles (see the supplemental text), so the rapid turnover is most likely due to MT dynamic instability superimposed on poleward flux (). Importantly, the turnover is independent of both the size of the bleached zone and its position along the long axis of the metaphase spindle. For example, the kinetics and extent of fluorescence recovery were very similar within several adjacent 1-μm-wide subregions of a larger, 5-μm bleach region (t = 7.4 s; percentage recovery ∼94%), indicating that FRAP was uniform throughout the bleached area ( and Video 1). The observation suggests that the recovery could result from the exchange of tubulin subunits all along the pole–pole axis of the spindle. Interestingly, if we assume that tubulin subunits exchange only at the plus and minus ends and not at internal sites within the MT polymer lattice, these results are consistent with the view that dynamic MT ends are present throughout the spindle (see section Model Result 1). Spindle MTs are stabilized at anaphase onset in yeast and vertebrate cells (; ; ; ), so to see if a similar change in MT dynamics occurs within embryo spindles, we monitored FRAP recovery before and after anaphase B onset (). These spindles characteristically remain in the preanaphase B steady state for ∼100 s before elongating during anaphase B (from ∼10–12 to ∼14–16 μm), so we used pole–pole separation as a visual cue to detect the onset of anaphase B in FRAP experiments. Our studies showed that the half-time of recovery was the same before and after anaphase B onset, but there was a notable, position- dependent difference in the percentage of recovery during anaphase B spindle elongation (, C and D; Video 3, available at ; and ). Regions proximal to the spindle equator displayed similar fluorescence recovery to preanaphase B spindles (t of 5 s and 86% recovery), but in regions proximal to the poles, the extent of recovery was substantially reduced (to ∼46% with a t of 2.8 s). These differences in recovery are consistent with a spatially regulated change in MT dynamics at anaphase B onset, which results in the evolution of two populations of MTs near the poles; a small dynamic subset of MTs that continue to turn over rapidly and recover their fluorescence, and a second, new, stable subset of MTs that do not undergo detectable turnover, accounting for the lower extent of fluorescence recovery. If the turnover primarily reflects dynamic instability of MT plus ends, the aforementioned spatial change in MT polymer dynamics at anaphase B onset should correspond to an alteration in the spatial distribution of MT plus ends. To test this idea, we monitored the dynamics of EB1 (a “plus-end tip tracker,” which localizes to growing MT plus ends) using time-lapse imaging of transgenic fly embryos expressing an EB1 fusion protein containing GFP at its C terminus (; ). Mitosis progressed normally in these embryos, suggesting that EB1-GFP expression did not create any obvious defects. The EB1-GFP formed comets that displayed antipoleward motility at 0.25 ± 0.2 μm/s (not depicted) and underwent a stage-specific relocalization; the comets were distributed uniformly throughout preanaphase B spindles, but at anaphase B onset, they redistributed into a 3–4-μm-wide band at the spindle equator (, A and B; and Video 4, available at ). Assuming that EB1 specifically marks the plus ends of growing MTs, as expected, then this redistribution must reflect changes in the distribution of growing MT plus ends, which would support our hypothesis that the spatial changes in MT turnover measured in the FRAP analysis of spindles before and after anaphase B onset reflects a change in MT plus-end distribution. We infer that MT plus ends, located throughout the half spindles, redistribute to the spindle midzone and, furthermore, by using kymography to track EB1-GFP throughout the whole spindle, we determined that this redistribution occurs abruptly at anaphase B onset when the spindle starts to elongate (; Fig. S2 B; and Video 4). Thus, at anaphase B onset, there is a change in MT dynamics that leads to the rapid redistribution of MT plus ends from throughout the half-spindles to the spindle midzone, where the overlapping plus ends of antiparallel ipMTs are found. To test the role of cyclin B degradation, which is required for cell cycle progression from metaphase through mitotic exit (; ), we injected a stable, nondegradable GST–cyclin B fusion protein into embryos expressing GFP-tubulin or GFP-histone (; ). The injected embryos displayed a gradient of phenotypes, with spindles proximal to the injection site arresting in metaphase and not exiting for at least 15 min (Fig. S2 A). Spindles further away from the injection site progressed slowly through anaphase A and partially or completely segregated their chromatids after a slight delay (). These spindles, which we term anaphase A–arrested, never entered anaphase B but instead maintained constant pole–pole spacing. The metaphase- and anaphase A–arrested spindles displayed MT turnover similar to that of wild-type preanaphase B spindles in FRAP experiments with nearly complete recoveries both proximal to the poles and at the equator ( and ). In addition, fluorescence speckle microscopy showed persistent poleward flux, suggesting that the depolymerase KLP10A at the poles remained active (; unpublished data). Finally, these arrested spindles maintained a persistent, uniform distribution of EB1-GFP (), even after anaphase A chromosome movement (Fig. S2 C). These observations strongly suggest that the redistribution of plus ends, the spatial change in MT dynamics, and the inactivation of the depolymerase at the poles, which occur at anaphase B onset in wild-type embryos, are initiated by a switch that requires cyclin B degradation and is therefore likely to be cell cycle regulated. The results so far suggest that cyclin B degradation initiates a signal transduction pathway that triggers the redistribution of MT plus ends to the overlap region at the spindle equator, raising questions about the molecular identity of the targets of the signal and the mechanism of redistribution. These issues are difficult to address experimentally because the perturbation of candidate target molecules such as EB1, KLP10A, KLP3A, and RanGTP can have multiple effects on spindle assembly, chromosome motility, and anaphase B (; ; ; ; ), which can obscure specific effects on MT plus-end dynamics at anaphase B onset. We did observe that EB1 still redistributed to the spindle interzone in the fraction of Ran- or KLP3A-inhibited spindles that underwent partial anaphase B (Fig. S2 D), but whether a reduced redistribution correlated with the decrease in spindle elongation was impossible to quantify. Another problem was that the uniform distribution of MT plus ends and the rapid, full and uniform FRAP recovery observed in preanaphase B spindles could easily be explained if dynamic MT minus ends are also uniformly distributed throughout the spindle (; ; Fig. S3 B, available at ). However, the embryo spindle assembles primarily by the centrosomal pathway, so most of its MT minus ends are likely to be proximal to the poles (, ; ). Such a biased distribution of minus ends, superimposed on a uniform distribution of MT plus ends, would intuitively predict a slower or less extensive recovery of a bleach mark near the pole versus the equator of preanaphase (as well as anaphase B) spindles, in contrast to what we observe (). In our simulations, we consider hundreds of MTs asynchronously undergoing dynamic instability at their plus ends and simultaneously sliding toward the spindle poles via forces generated by the bipolar motors at the antiparallel overlaps (). During preanaphase B, the minus ends of “virtual” MTs depolymerize at the poles with a mean rate equal to the free sliding rate of the bipolar motors at the midzone, and because the motors work near their load-free regime, the spindle length remains constant. At the onset of anaphase B, we numerically “switch off” MT depolymerization so that MT sliding is converted into spindle elongation. In the model, the dynamics of MT plus ends is determined by the four parameters of dynamic instability: the growth and shortening rates, and , and the rescue and catastrophe frequencies, and . Assuming that all spindle MTs obey the same dynamics (constant rates), our goal was to explore which regions of this four-dimensional parameter space, and which distributions of MT plus and minus ends could account for the observed rapid FRAP rates. The simulation results show that if the MT dynamic parameters are maintained within a narrow range, then the observed uniform, rapid, and complete FRAP recovery in the preanaphase B spindle can be accounted for even if MT minus ends are restricted to the spindle poles (, A and B; see the supplemental text for details). Specifically, the rescue and catastrophe frequencies should be fast enough (∼0.15 s), so the MT growth and shortening cycles are rapid; the growth and shortening rates have to be high enough (∼0.35 μm/s) so that the mean MT length is ∼2 μm; and the mean growth length must be slightly smaller than the mean shortening length during the MT growth and shortening cycle (). This model result is further supported by theoretical arguments based on expressing the mean length of MTs, <>, in terms of the characteristic lengths, = / and = /, by which the MTs grow and shrink, respectively, within one dynamic instability cycle, and thereby estimating the mean turnover rate (see the supplemental text for details). These arguments suggest that our FRAP observations can be explained if the spindle maintains ∼ ∼ 0.15 s. This predicted order of magnitude is in the same range as previous experimental estimates obtained for metaphase spindles (; ). Using these arguments, we also predict that the mean length of MTs has to be ∼2 μm, whereas ∼ ∼ 2 μm × 0.15s ∼0.35 μm/s. Indeed, when we use values within this range in our model, the virtual bleaching of an entire half-spindle ( and Videos 5 and 6, available at ), of small regions near the spindle pole, or of the equator in preanaphase B spindles, all give rise to FRAP recovery kinetics that account well for our experimental results, that is, a near complete recovery (∼90–95%) and very fast FRAP recovery rate (∼7 s). Also, under these conditions, the growing MT plus ends are uniformly distributed throughout the preanaphase B spindle (unpublished data). This result is surprising because, with the minus ends of all MTs anchored to the spindle poles, we had expected the recovery of a bleach region near the spindle pole to be incomplete as a result of the stable portions of the long MTs in the spindle. However, our simulations and theoretical estimates demonstrate that as long as the dynamic instability parameters are adjusted to maintain the mean life cycle of MTs in the order of twice the FRAP half-time, the recovery is uniform, full, and rapid. Therefore, our experimental findings on FRAP and EB1-GFP distribution are entirely consistent with the notion that the embryo spindles conform to the classic view of centrosome-directed spindle formation pathway with most minus ends anchored to spindle poles. In the alternative case, if MT minus ends are also spatially uniformly distributed in the preanaphase B spindle, the restrictions on the MT dynamic parameters that could account for the observed rapid, full, and uniform FRAP rates relax. However, in this case, the mechanical integrity of the spindle, which then depends on the small and dynamic overlaps between these short interconnected MTs, is compromised (see the supplemental text). To understand the experimentally observed spatial change in FRAP and in the distribution of growing MT plus ends (EB1-GFP) at the onset of anaphase B, we considered three different scenarios that could potentially account for the reduced extent of FRAP near the poles at anaphase B: the possibility of a spatial gradient of rescue or catastrophe () established at anaphase B onset (Fig. S3 C); a change in the spatial distribution of MT minus ends as a result of the dissociation from the poles of the minus ends of short MTs, which do not overlap with other antiparallel MTs, at anaphase B onset; and an MT-associated protein or motor-dependent differential stabilization of overlapping ipMT plus ends. We found that although the second scenario can, in principle, quantitatively explain the reduced extent of FRAP near the poles during anaphase B, the fit to the data is poor (see the supplemental text). The result of modeling scenario 3 was the induction of an “overgrowth” of overlapping ipMT plus ends into the opposing preanaphase B half-spindles followed by the depletion of growing ipMTs from the equator of anaphase B spindles. Here, the change in ipMT distribution and dynamics was totally inconsistent with the experimental observations (unpublished data). In the context of scenario 1, a rapid establishment of a gradient in rescue frequency at anaphase B onset, with decreased rescue near the poles, explains well the reduced extent of FRAP near the poles but cannot account for the maintenance of numerous growing plus ends at the spindle equator: many MTs that shorten toward the poles vanish, resulting in net loss in both long and short MTs (see the supplemental text). On the other hand, we obtain a very good fit for both the spatial change in FRAP and the distribution of the growing plus ends when we assume that a spatial gradient of catastrophe frequency is established at anaphase B onset, such that the catastrophe frequency increases threefold near the poles (; Fig. S3 C; and Video 7, available at ). This gradient in catastrophe rate, together with the rapid MT dynamics, leads to an abrupt “sorting” of MTs into short and long ones, thereby rapidly relocating the MT plus ends to the proximity of the poles and to the spindle midzone. xref #text Flies were maintained at 25°C, and 0–2-h embryos were collected as described previously (). Flies expressing GFP-tubulin were provided by A. Spradling (Carnegie Institution, Washington, DC) and those expressing GFP-histone by B. Sullivan (University of California, Santa Cruz, Santa Cruz, CA). The EB1-GFP transgenic fly was a gift from S. Rogers (University of North Carolina at Chapel Hill, Chapel Hill, NC), B. Eaton, and G. Davis (University of California, San Francisco, San Francisco, CA). The EB1-GFP transgene, a C-terminal GFP fusion to EB1 (pUASp-EB1-GFP) was expressed under the UASp promoter. Germ line expression of pUASp-GFP-EB1 transgene was driven by using the tubulin Gal4 drivers ([α]GAL4-VP16-V2H) or ([α]GAL4-VP16-V37). GST–cyclin B (provided by D. Kellogg, University of California, Santa Cruz, Santa Cruz, CA) was purified from as described previously (). During the final elution step, the protein was eluted using 10 mM glutathione in 50 mM Tris, pH 8.1, and 0.3 M KCl. The eluted protein was dialyzed into 50 mM Hepes, pH 7.6, 0.125 M KCl, and 10% glycerol and concentrated to 10 mg/ml for injection into embryos. FRAP experiments were done on a laser-scanning confocal microscope (FV1000; Olympus) with a 60× 1.40 NA objective at 23°C, and image acquisition was done using the Fluoview software (version 1.5; Olympus). The embryos expressing GFP-tubulin were dechorionated and kept in halocarbon oil to prevent dehydration and were imaged using the 488-nm line from an argon laser. A separate 405-nm laser was used to photobleach GFP-tubulin. The use of two different lasers allowed simultaneous imaging and bleaching. The spindle was bleached in rectangular or circular areas of defined width and diameter, respectively, and images were acquired every 1.1 s. The spindles were corrected for movement using MatLab (Mathworks), and the fluorescence intensities within the bleached region were measured using MetaMorph imaging software (Universal Imaging Corp.). The fluorescence intensity of the bleached region over time was normalized with the prebleached fluorescence intensity and was plotted as a function of time. The recovery half-time was obtained by fitting a single exponential curve = F + (F − F) (1 − e) () to the recovery time course curve. The images were not corrected for bleaching because it was not feasible to find an unbleached spot devoid of MT alterations within the embryos during anaphase B. Time-lapse microscopy of EB1-GFP–, GFP-tubulin–, and GFP-histone–expressing embryos was done on a microscope (Olympus) equipped with an UltraView spinning disk confocal head (PerkinElmer) with a 100× 1.35 NA objective. The embryos were prepared as outlined in the beginning of this section and kept in halocarbon oil. Images were acquired using the UltraView software (PerkinElmer) at a rate of 1.5 s/frame at 23°C and recorded using a digital camera (ORCA ER; Hamamatsu). Embryos expressing EB1-GFP and GFP-histone were injected with rhodamine tubulin (Cytoskeleton) to visualize MTs. Images were analyzed using MetaMorph. The images were processed using the “No Neighbors” Deconvolution and Low Pass Filter commands. The whole spindle kymograph was done using MatLab. Fig. S1 shows a metaphase spindle bleached at the equator and FRAP of a spindle in the presence of taxol. Fig. S2 shows a plot of spindle pole distance versus time in the presence of GST–cyclin B, kymographs of tubulin and EB1 intensity in the wild-type and GST–cyclin B–injected embryo, and EB1 distribution in the presence of anti-KLP3A antibody and RanT24N. Fig. S3 provides illustrations of possible organization of MT minus ends in the spindle and shows the predicted spatial gradients of MT dynamic parameters in the spindle, as well as a kymograph of tubulin and EB1 intensity from prometaphase through anaphase B and a kymograph of the positions of growing MT plus ends in a virtual spindle from prometaphase through anaphase B. Video 1 shows FRAP of a metaphase spindle, and Videos 2 and 3 show the simultaneous double bleaching of pole and equator of a metaphase and anaphase B spindle, respectively. Video 4 shows the time lapse of an EB1-GFP–expressing embryo injected with rhodamine tubulin. Video 5 shows a typical in silico FRAP of a virtual spindle in preanaphase B. Video 6 shows the fluorescence recovery of individual MTs in an ipMT bundle of the spindle in Video 5. Video 7 shows the fluorescence recovery of individual MTs in an ipMT bundle of a virtual spindle in anaphase B, before and after in silico bleaching. Video 8 shows the fluorescence recovery of individual MTs in an ipMT bundle of a virtual spindle in preanaphase B without poleward flux, before and after in silico bleaching.

The spindle assembly checkpoint (SAC) is critical for preventing the onset of anaphase until all chromosomes are aligned on the metaphase plate. A single misaligned kinetochore is sufficient to generate a wait anaphase signal, thereby ensuring that all sister chromatids segregate to opposite ends of the spindle and are equally distributed to the daughter cells. Failure of the SAC can lead to premature anaphase onset and aneuploidy (; ; for review see ). Such defects can have consequences for a whole organism, as mice that lack a full complement of SAC genes have more frequent DNA segregation errors and are more susceptible to tumor development (). The presence of the SAC was initially inferred from observations that cells delay in metaphase when meiotic sex chromosomes fail to pair and align or after the spindle is perturbed by either microtubule poisons or microsurgery. Molecules responsible for the SAC were later identified in yeast genetic screens and named Mad1, -2, and -3 (Mad for mitotic arrest deficient) and Bub1, -2, and -3 (Bub for budding unperturbed by benzimidazole). Subsequent work showed that these proteins together with the MPS1 kinase form distinct complexes that target to the kinetochore (for reviews see ; ; ; ). Two additional metazoan checkpoint proteins, Zw10 and Rough Deal (Rod), were later isolated as cell cycle mutants in . These two proteins, together with a third protein called Zwilch, form a complex (Rod–Zw10–Zwilch complex [RZZ]) that regulates the levels of Mad1 and Mad2 on the kinetochore (for review see ). Ultimately, the SAC pathway must lead to inhibition of the anaphase-promoting complex (APC), a multisubunit ubiquitin E3 ligase that targets multiple mitotic regulators (e.g., mitotic cyclins as well as the securin protein that inhibits the cleavage of cohesin molecules) for proteosome degradation to allow mitotic exit (). Several studies have shown that localization of the checkpoint proteins to misaligned kinetochores is essential for establishing the SAC and keeping the APC inhibited, most likely by generating a diffusible signal that inhibits the APC (; ; for review see ). The nature of the diffusible signal is still subject to debate. However, a current model suggests that the kinetochore-bound Mad1–Mad2 complex acts as a template that coverts the free, inactive Mad2 to an active form that can diffuse away from the kinetochore and bind to and sequester Cdc20, a regulatory component of the APC (for review see ). The capture of microtubules by the kinetochore and the downstream activity of two different microtubule motors are required for silencing the SAC in metazoans. One of these motors is the kinesin centromere protein (CENP) E, which may act as a tension sensor that, when stretched, inactivates the BubR1-dependent inhibition of Cdc20 (; ). The second motor is dynein, which transports Mad1, Mad2, and RZZ from the kinetochore to the spindle pole (; ). Dynein-based removal of Mad1 and Mad2 from the kinetochore may disrupt the template mechanism that generates the active Mad2 that inhibits the APC (; for review see ). After inhibition or depletion of dynein or its cofactors, metazoan cells arrest in metaphase with correctly aligned chromosomes and high levels of kinetochore-bound Mad1, Mad2, and RZZ. Resolving the mechanism of dynein recruitment to kinetochores is important for understanding how kinetochore– microtubule binding ultimately leads to inactivation of the SAC. Currently, it is thought that dynein is brought to the kinetochore by binding directly to dynactin (a multisubunit complex required for multiple dynein functions; ), which, in turn, binds to the Zw10 subunit of the RZZ complex (). Lis1, another dynein cofactor, also has been proposed to play a role in targeting dynein to kinetochores (). Dynactin, Lis1, and Zw10 are not kinetochore- specific factors, as they are involved in targeting dynein to multiple other locations in the cell (; ). It has not been clearly established whether dynactin and Lis1 are sufficient for targeting dynein to kinetochores or whether other proteins might be involved. To find new proteins that might participate in the SAC, we undertook an automated 7,200 gene mitotic index RNAi screen in S2 cells. This screen uncovered a novel gene, which we also identified in an independent screen of genes involved in S2 cell spreading and morphology. We show that this protein (termed Spindly) localizes to microtubule plus ends in interphase and to kinetochores during mitosis. Cells depleted of Spindly arrest in metaphase with high levels of Mad2 and Rod on aligned kinetochores, a defect caused by a failure to recruit dynein to the kinetochore. However, Spindly is not required for other dynein functions during interphase and mitosis. We also identify a human homologue of Spindly, which is similarly involved in recruiting dynein to kinetochores. Thus, our results have uncovered a novel conserved dynein regulator that is involved specifically in dynein's function in silencing the SAC. Using a double-stranded RNA (dsRNA) library corresponding to ∼7,200 genes (), we performed two screens using S2 cells (). The first screen measured mitotic index (the percentage of phosphohistone H3–positive cells in a population; see Materials and methods). In the second screen, the shape of S2 cells (spread on concanavalin A [Con A]–coated surfaces; ) was evaluated by visual inspection. RNAi of one novel gene, CG15415, produced strong phenotypes in both screens. CG15415 is a novel uncharacterized gene encoding a 780–amino acid protein with predicted N-terminal coiled-coil sequences and four repeats with the consensus sequence TPXKPQXKGTPVK (). In the interphase screen, many of the CG15415-depleted cells showed spiky and elongated microtubule-rich projections in contrast to the rounded shape of normal spread S2 cells (). In the mitotic index screen, the depletion of CG15415 caused an increase in mitotic index that was comparable with that observed for RNAi of the dynein heavy chain (DHC) and the APC subunit Cdc16 (). The majority of the mitotic CG15415- depleted cells were arrested in metaphase, which is also similar to DHC depletion (). This result was confirmed in live cells expressing GFP-tubulin, in which CG15415-depleted cells failed to enter anaphase within 4 h after nuclear envelope breakdown. In contrast, untreated cells initiated anaphase within 20–85 min of nuclear envelope breakdown (unpublished data). Because the depletion of CG15415 produced spindle-shaped interphase cell morphology and arrested cells with metaphase spindles, we refer to this protein as Spindly. The specificity of the Spindly phenotypes was confirmed using three nonoverlapping dsRNAs: two in the coding region and one dsRNA that targets the 3′ untranslated region (UTR; ). Using an antibody generated against Spindly's C-terminal 357 amino acids, we confirmed that the three dsRNAs effectively depleted the protein after 5 d (Fig. S1 a, available at ). As further confirmation of the specificity of the Spindly RNAi phenotype, we found that expression of a GFP-Spindly fusion protein could rescue the metaphase block after the endogenous protein was depleted with the 3′ UTR dsRNA. This result also indicates that Spindly retains its function after fusion to GFP, enabling the localization studies described in the next section. To learn more about Spindly's function, we examined the localization and dynamics of GFP-tagged Spindly. In live cells expressing low levels of GFP-Spindly, the protein was concentrated in punctae that continually moved to the periphery of the cell, which is behavior typical of microtubule plus end–binding proteins (Video 1, available at ). Fixation and staining of cells expressing low levels of GFP-Spindly with an antibody to EB1 (a well-established plus end–binding protein) confirmed this localization, although the plus end enrichment was less pronounced than that displayed by EB1 (). At higher levels of GFP-Spindly expression, the protein began to decorate along the length of the microtubule and to localize to the lamella (unpublished data). After cells entered mitosis, GFP-Spindly was no longer localized to microtubule tips but instead was found on kinetochores. In prometaphase cells, GFP-Spindly was found on most kinetochores, a localization confirmed by colocalization with anti-Cid antibodies, which recognize the homologue of CENP-A. However, in metaphase cells, the levels of GFP-Spindly were reduced considerably on the kinetochores of aligned chromosomes, and the protein was more evident on the mitotic spindle, especially at spindle poles (). During anaphase, GFP-Spindly was seen once again at high levels on kinetochores, but, after the nuclear envelope reformed in telophase, the protein was excluded from the nucleus. Time-lapse microscopy revealed that high initial levels of GFP-Spindly on misaligned chromosomes decreased as these chromosomes were pulled toward the metaphase plate ( and Videos 2 and 3, available at ). A similar distribution of endogenous Spindly in mitosis was confirmed using an affinity-purified antibody in cells expressing the homologue of the kinetochore protein Mis12 (CG18156) fused to GFP (Fig. S1). The transient targeting of Spindly to kinetochores is very similar to what has been reported for the mitotic checkpoint proteins Rod and Mad2 (; ). This dynamic kinetochore localization together with the data from our mitotic index screen led us to focus our efforts on understanding Spindly's role during mitosis. Components of the RZZ complex as well as Mad2 accumulate on kinetochores in prometaphase and are shed from metaphase kinetochores by dynein-dependent transport along kinetochore microtubules (; ). Using faster acquisition live cell imaging, we similarly observed punctae of GFP-Spindly moving processively from metaphase-aligned kinetochores toward the spindle poles ( and Video 4, available at ). Kymograph analysis revealed that GFP-Spindly moved poleward at a mean velocity of ∼12 μm/min ( and S2), which is similar to rates reported for the dynein-mediated transport of RZZ and Mad2 in (; ). However, not all GFP-Spindly particles moved uniformly; some paused or made short reversals toward the kinetochore before continuing toward the spindle pole (Video 4), which is behavior similar to that described for dynein–dynactin complexes in vitro (). To establish whether dynein is indeed the motor responsible for the poleward transport of Spindly, we examined GFP-Spindly after RNAi-mediated depletion of the cytoplasmic DHC. Under these conditions, high levels of GFP-Spindly accumulated on metaphase-aligned kinetochores ( and Video 4), which is similar to what has been described for Rod and Mad2 after the disruption of dynein (; unpublished data). Immunofluorescence localization of endogenous Spindly confirmed this result (unpublished data). We also no longer observed the poleward transport of GFP-Spindly by time-lapse microscopy. RNAi-mediated depletion of the dynein regulatory proteins Lis1 and p150 produced similar results (Video 4 and not depicted, respectively). These results indicate that kinetochore to pole movement of Spindly depends on cytoplasmic dynein and its activators, as is true of other known components of the SAC. We next sought to determine how Spindly is targeted to the kinetochore. It has been previously shown that recruitment of dynein–dynactin to the corona region of the kinetochore depends on the RZZ complex, which, in turn, links through Zwint-1 to the Ndc80 and Mis12 complexes of the kinetochore (; ; ). The depletion of any of the three RZZ polypeptides destabilizes the whole complex and prevents the recruitment of Mad2 and dynein–dynactin (, ; ). When Rod was depleted by RNAi, GFP-Spindly no longer localized to kinetochores or the spindle poles ( and Video 4). These results indicate that Spindly is a part of the corona region of the kinetochore and requires the RZZ complex (but not dynein or dynactin, as discussed above) for its kinetochore localization. Because Spindly is required for cells to complete mitosis and localizes to kinetochores in a manner similar to known SAC proteins, we decided to investigate the role of Spindly in the kinetochore localization of Rod and Mad2. In prometaphase cells, Rod and Mad2 are more abundant on misaligned than aligned chromosomes and are also observed on the spindle and spindle poles () as previously described (; ). However, after Spindly RNAi, the levels of Rod and Mad2 were comparable on misaligned and metaphase-aligned kinetochores, which is similar to the outcome of DHC RNAi (). These results indicate that both dynein and Spindly are required for the shedding of Rod and Mad2 from the kinetochore. Consistent with this interpretation, the staining of Rod and Mad2 on the spindle (likely reflecting the population of molecules undergoing transport) was severely reduced after Spindly and DHC RNAi (). The retention of Rod and Mad2 on metaphase-aligned chromosomes explains the high mitotic index and increased number of metaphases seen after Spindly or DHC depletion (). As another measure of kinetochore function, we determined the time required to align all chromosomes at the metaphase plate using a cell line expressing GFP-tagged histone H2B and mCherry-tagged α-tubulin and automated time-lapse imaging (see Materials and methods). Intriguingly, the Spindly- and DHC-depleted cells both required 50% more time to form a metaphase plate compared with untreated cells (a mean of 18.5 ± 2.3 min vs. 28.1 ± 4.9 and 28.2 ± 3.7 min [±SEM] for Spindly and dynein, respectively), which might be the result of a defect in initial kinetochore microtubule capture ( and Videos 5–8, available at ). also proposed that kinetochore-associated dynein could play an important role in making lateral attachments between chromosomes and microtubules before the final end-on attachments observed at metaphase, which could explain the delay in chromosome alignment after DHC depletion. Consistent with our results for centromere tension, cells depleted of Rod and Cdc27 took considerably longer (48 ± 16.9 min) to assemble a metaphase plate, which might reflect a requirement for the RZZ complex to incorporate multiple proteins into the outer corona of the kinetochore. In summary, these results suggest that Spindly-depleted cells do not have gross defects in kinetochores or kinetochore–microtubule interactions but rather have kinetochores that resemble those found in cells lacking dynein. The similar Spindly and dynein RNAi phenotypes of mitotic arrest, defects in Mad2 and Rod transport, and delays in forming a metaphase plate suggested that Spindly might somehow play a role in dynein function at the kinetochore. Therefore, we next examined whether Spindly affects the kinetochore localization of dynein. To more easily assay dynein localization, microtubules were depolymerized with colchicine, which causes a substantial accumulation of dynein and dynactin on kinetochores (). Spindly RNAi resulted in a profound reduction in DHC staining at kinetochores compared with untreated cells (). Interfering with dynactin function has also been reported to abolish kinetochore staining of dynein (; ; ), a finding that we repeated as well (). However, dynactin, as assayed by GFP-p150 () or with anti-p150 antibodies (Fig. S3, a and c; available at ), was still recruited to kinetochores in Spindly-depleted cells (however, Rod RNAi displaces p150 from kinetochores; ). To confirm that Spindly is required for dynein kinetochore localization and not the stability of the protein, immunoblot analysis was performed, which revealed that DHC and p150 protein levels were unaltered by Spindly RNAi (). Thus, Spindly is required for dynein but not dynactin recruitment to kinetochores. The aforementioned results clearly revealed an important role for Spindly in dynein function at the kinetochore. We next investigated whether Spindly participates in other dynein-mediated activities. In S2 cells, dynein is known to be important for spindle focusing, specifically in transporting kinetochore fibers along microtubules emanating from the centrosomes. After DHC RNAi, the centrosomes detach and move away from the minus ends of the K fibers (; ; ). However, Spindly depletion did not produce the centrosome detachment or spindle focusing defects seen in cells lacking dynein (). Additionally, after plating on Con A for 3 h, Spindly-depleted and untreated interphase cells generally cluster their endosomes (marked by GFP-Rab5) toward the cell interior, whereas endosomes in dynein- or dynactin-depleted cells tend to remain spread throughout the cell (Fig. S4, available at ; dynactin depletion data not depicted). Collectively, these experiments suggest that Spindly influences dynein function at the kinetochore but not everywhere throughout the cell. We next sought to identify Spindly homologues in other species. Standard BLAST (Basic Local Alignment and Search Tool) searches identified Spindly homologues in the insects and but not in more distant species. Multiple Em for motif elicitation was then used to identify conserved motifs present in all three insect homologues, and these motifs were used for MAST (Motif Alignment and Search Tool) searches to identify more distant homologues (; ). A conserved 32-amino acid motif found in a break between predicted coiled-coil domains in the N terminus of all three insect proteins also was found in the human protein RefSeq NP_060255 (Fig. S5 a, available at ). The overall primary sequence conservation between Spindly and human NP_060255 is low (14.3% identity), and the putative human homologue is somewhat shorter (605 vs. 780 amino acids). However, the sequences in the 32–amino acid conserved motif are 56% identical (75% similar), and the first nine amino acids of this motif are 100% identical. The predicted coiled-coil organization and charge distribution of the putative human homologue also is similar to Spindly, although the sequences of the coiled coils are not conserved. The function of the putative human homologue of Spindly had not been previously characterized. To test whether NP_060255 is a bona fide functional homologue of Spindly, we examined whether depletion of the protein by siRNA caused mitotic defects. Transfection of a siRNA pool targeted to NP_060255 reduced NP_060255 protein levels by 86% (immunoblot analysis; not depicted) and produced a twofold increase in the mitotic index of HeLa cells after 48 h (). When these mitotic cells were examined, a dramatic increase in the ratio of metaphase versus anaphase cells was apparent (), and a substantial number of these cells had misaligned chromosomes (Fig. S5 b). A similar phenotype has been reported in HeLa cells after the depletion of either CLIP-170 or dynein, which targets CLIP-170 to the kinetochore (). We next localized NP_060255 with a polyclonal antibody in HeLa cells treated with colchicine to depolymerize spindle microtubules. Similar to the protein, we observed punctae of NP_060255 that were coincident with CENP-A–stained centromeres (). This staining was eliminated by treating cells with the siRNA oligonucleotides that target NP_060255 (), confirming the localization of this protein at kinetochores. To determine whether NP_060255, like Spindly, is required to recruit dynein to the human kinetochore, we localized dynein using an antibody to its intermediate chain (dynein intermediate chain [DIC]) in colchicine-treated siRNA-transfected cells. In control siRNA-treated cells, a subset of the DIC-stained punctae colocalized with CENP-A, a marker of the centromere (). However, after siRNA against NP_060255, the colocalization of dynein with CENP-A was substantially reduced (). Similar to what was found for Spindly, the depletion of NP_060255 also decreased the stretch between paired centromeres from 1.15 to 0.98 μm (29.6 ± 4.5% decrease; P < 0.00005), a result that is in agreement with the previously reported effect of p50 microinjection (a dominant-negative inhibitor of dynactin function) on kinetochore stretch (). Collectively, our data show that the protein encoded by NP_060255 localizes to kinetochores and is required for localizing dynein to the kinetochore and for mitotic progression. Thus, we suggest that NP_060255 is a true homologue of Spindly and propose to rename NP_060255 as Hs Spindly. These results also indicate that the mechanism for localizing dynein to the kinetochore to silence the SAC is conserved between humans and flies. #text Schneider cell line (S2) cells (Invitrogen) were cultured, and dsRNA incubation was performed as previously described (; ). The 7,200 gene screens were performed with a previously described library (). After 5 d of dsRNA treatment, cells were plated in glass-bottom 96-well plates (Whatman) coated with Con A (Sigma-Aldrich). Cell shape phenotypes were manually scored and documented on a microscope (Axioplan 200M; Carl Zeiss MicroImaging, Inc.) equipped with a 40× 1.3 NA objective and a cooled CCD camera (Sensicam HQ; The Cooke Corporation) after staining with an anti-tubulin antibody (DM1A, anti–α-tubulin; 1:500; Sigma-Aldrich) and rhodamine phalloidin. For the mitotic index screen, mitotic index was determined by dividing the number of phosphohistone H3–positive nuclei (1:1,000; Upstate Biotechnology) by the total number of nuclei (determined by DAPI staining). These cells were imaged using a 20 or 10× air objective in either an ArraySCAN HCS System (Cellomics Inc.) or an automated microscope (ImageXpressMicro; Molecular Devices). In the follow-up experiments described in this paper, most assays were performed after 7 d of RNAi treatment as previously reported (). At the end of the RNAi treatments, cells were resuspended and seeded on Con A–coated coverglasses or dishes for 2 h before imaging or fixation. For colchicine treatment, cells were allowed to settle for 20 min, the media was removed and replaced with media containing 6 μg/ml colchicine, and imaging or fixation and staining was performed 4 h after treatment began. HeLa cells were maintained as previously described (). siRNA oligonucleotides were On-TARGETplus SMARTpools (Dharmacon), and transfections were performed using Dharmafect1 (Dharmacon) according to the manufacturer's instructions. Immunofluorescence was performed with affinity-purified rabbit anti-Dm Spindly (1:100), rabbit anti-Hs Spindly serum (1:100), chicken anti-Cid (1:200; provided by G. Karpen, Lawrence Berkeley National Laboratory, Berkeley, CA), rabbit anti-Rod (1:200; provided by R. Karess, Centre National de la Recherche Scientifique, Gif-sur-Yvette, France), mouse anti-DHC (1:100; provided by T. Hays, University of Minnesota, Minneapolis, MN), rabbit anti-p150 (1:200; provided by R. Giet, University of Rennes, Rennes, France), rat anti–α-tubulin (1:150; Serotec), mouse anti–CENP-A (1:2,000; Abcam), rabbit anti-DIC (1:500; provided by K. Vaughan, Notre Dame University, South Bend, IN), and rabbit anti-Mad2 (1:35; provided by C. Sunkel, (Instituto de Biologia Molecular e Celular, Porto, Portugal). Images were collected with either a confocal microscope (LSM510; Carl Zeiss MicroImaging, Inc.) using a 63× 1.4 NA objective or a microscope (Axioplan; Carl Zeiss MicroImaging, Inc.) outfitted with 40× 1.3 NA, 63× 1.4 NA, and 100× 1.3 NA objectives and a cooled CCD camera (Sensicam HQ; The Cooke Corporation). We cloned Spindly from an S2 cell cDNA pool and found that the sequenced cDNA clone lacks 27 amino acids from the predicted ORF. This ORF was cloned into the pENTR/D-TOPO vector (Invitrogen) and moved into N- or C-terminal Gateway GFP vectors under the control of the metallothionein promoter vector (N- and C-terminal fusions produced the same results). To observe the tip tracking, it was optimal to use cells without inducing GFP-Spindly protein expression with CuSO For observation of protein on kinetochores, GFP-Spindly expression was induced by incubating the cells with 20 μm CuSO for 18 h. S2 cells stably expressing GFP-tagged proteins were plated in dishes with coverslip bottoms (MatTek) that had been coated with Con A. Images were collected at 1–20-s intervals at room temperature using a cooled CCD (Orca II ERG; Hamamatsu Photonics) or iCCD (MEGA10; Stanford Photonics) camera attached to a spinning disk confocal scan head (Yokogawa Electric and Solamere Inc.) that was mounted on a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) outfitted with a 100× 1.45 NA objective. Images were collected using either MetaMorph software (Molecular Devices), QED (Media Cybernetics), or μManager (). For analysis of GFP-Spindly movement from the kinetochore to poles, cells were imaged on the spinning disk confocal microscope with 300-ms exposures taken every second. Image stacks were opened in ImageJ (National Institutes of Health), and spindles were oriented horizontally. A box was drawn that was wide enough to contain all of the kinetochores on one half of the metaphase plate and long enough to contain the proximal spindle pole. A stack of kymographs (each one representing a given one-pixel–thick line within the box) was then generated. These kymograph stacks were then combined into maximum intensity z projections, and particle velocities were determined by measuring the lengths of the lines created by particles moving toward or from the spindle poles (distance traveled) and then dividing that value by the displacement in the y direction (time). To determine statistical significance, datasets were analyzed using the test. A region of the Spindly gene corresponding to amino acids 451–780 was cloned into pET28a (Novagen), and protein expression was induced in BL21 DE3 cells (Invitrogen). Full-length Hs Spindly was also cloned into pET28a, and the protein was expressed in BL21 DE3 cells. The expressed proteins were purified and used for injecting rabbits (Covance). Anti-Dm Spindly antibodies were purified on an Affi-Gel 10 column (Bio-Rad Laboratories) containing the immobilized antigen. To isolate protein from S2 and HeLa cells after RNAi treatment, 100 μl laemmli sample buffer was added per well of cells in a 96-well plate. The sample was then processed for Western blotting as previously described (). The blot shown in Fig. S1 was pieced together from multiple lanes of a larger gel; the blot was cut between the 100- and 150-kD markers and blotted with the indicated antibodies (rabbit anti-p150; 1:500; provided by E. Holzbaur, University of Pennsylvania, Philadelphia, PA). The blot shown in was cut at the 250-kD marker and blotted with the indicated antibodies (mouse anti-DHC; 1:1,000; provided by T. Hays). Fig. S1 shows that the endogenous Spindly protein also enriches on unattached, unaligned, and anaphase kinetochores. Fig. S2 shows kymograph analysis of GFP-Spindly particles. Fig. S3 shows that Spindly depletion does not alter the targeting of endogenous dynactin to the kinetochore. Fig. S4 shows that Spindly is not required for the dynein-dependent reorganization of endosomes in S2 cells. Fig. S5 shows that the depletion of NP_060255 causes defects in chromosome alignment. Video 1 shows that GFP-Spindly tracks on the plus ends of microtubules in interphase cells. Video 2 shows that GFP-Spindly concentrates on lagging chromosomes and then diminishes after alignment at the metaphase plate. Video 3 shows that GFP-Spindly returns to kinetochores during anaphase, and Video 4 shows that GFP-Spindly traffics from kinetochores to centrosomes in a dynein- and Rod-dependent manner. Videos 5–8 show that the depletion of Spindly, dynein, or Rod slows the alignment of chromosomes on the metaphase plate. Online supplemental material is available at .

The ER plays a crucial role in several important aspects of eukaryotic cell physiology. It assists in the folding and maturation of all nascent secretory proteins and initiates their distribution to the broader secretory pathway (). In addition, the ER influences the overall composition of the cellular proteome by mediating the ER-associated degradation (ERAD) pathway, a pathway that destroys misfolded proteins and also responds to specific degradation signals to regulate the levels of certain native proteins (). The ER also houses many lipid biosynthetic enzymes, which impact the relative composition and overall abundance of lipids throughout the cell (). Genes involved in protein folding, protein trafficking, ERAD, and lipid metabolism are all transcriptionally activated by a conserved ER-initiated signal transduction pathway called the unfolded protein response (UPR; ; ; ; ; ). In budding yeast, the UPR pathway begins with an ER transmembrane protein, Ire1p (; ). The N terminus of Ire1p lies in the lumen of the ER, where it senses the ER's condition. When Ire1p detects a need for increased ER function, it transmits a signal across the ER membrane to activate its own cytosolic kinase and endoribonuclease domains (; ; ; ). Activated Ire1p then initiates the unconventional splicesome-independent splicing of mRNA (; ). Only the spliced form of mRNA can be translated, making the splicing step a critical point of regulation (; ). Upon translation, Hac1p localizes to the nucleus, where it acts as a transcription factor to up-regulate a wide array of UPR target genes (; ), thus increasing the ER's capacity to serve its many functions (). Northern analysis, which measures the relative abundance of spliced mRNA in the cell, is currently the most commonly used method of detecting UPR activation (). Using this technique, previous studies have detected UPR activation only during extreme conditions of ER stress. For example, mRNA splicing has been detected in cells treated with pharmacological agents that cause widespread protein misfolding (; ) or in cells overexpressing mutant proteins that fold improperly (). The inability to detect mRNA splicing during normal growth has led to the designation of the UPR pathway as a stress response pathway. However, it is likely that cellular demand for ER function is dynamic even during unstressed growth conditions. This evokes the intriguing possibility that in addition to responding to conditions of extreme stress, the UPR pathway manages the everyday challenges of fluctuating ER demand. This housekeeping function for the UPR has been previously unnoticed, perhaps because it induces a level of Ire1p activity that is too subtle to be detected by conventional Northern analysis. Because progression through the cell cycle requires dramatic molecular and cellular changes, we hypothesized that cell cycle progression requires fluctuations in ER capacity. To isolate a cell cycle event that requires particularly high ER functionality, we used ER stress as a tool to disrupt ER function. We then asked whether any particular cell cycle event was sensitive to this reduction in ER capacity. Most cell cycle events that we examined did not require exceptionally high ER activity, as they occurred normally during ER stress. However, cells experiencing ER stress were specifically defective in cytokinesis, suggesting that elevated ER functionality is required for cells to carry out efficient cytokinesis. Because cytokinesis required a greater ER capacity than other cell cycle events, we tested the possibility that the UPR plays a role in achieving an increased ER capacity during normal, unstressed cytokinesis. Indeed, we found that UPR-deficient cells were unable to carry out efficient cytokinesis even in the absence of external ER stress. This is the first time the UPR pathway has been shown to function in cells that are growing optimally, expressing no misfolded mutant proteins, exposed to no protein misfolding agents, and not differentiating into high volume secretory cells. Therefore, our study supports the concept of a UPR that continuously fine tunes the ER to accommodate everyday fluctuations in ER functional demand. Because previous Northern analysis has not uncovered HAC1 mRNA splicing in unstressed cells (), we performed Northern analysis with 30 μg RNA rather than the 10 μg RNA that is traditionally assayed. Under these conditions, we could clearly detect the spliced form of in unstressed optimally grown wild-type cells. This spliced form constituted 7.4 ± 0.6% of total mRNA (). Basal splicing was dependent, suggesting the presence of a bona fide UPR signal in unstressed cells. The results of our Northern analysis, which we confirmed by RT-PCR (Fig. S1, available at ), prompted us to seek a functional relevance for basal UPR induction. To determine whether this low level of UPR activity has a role in cell cycle progression, we used the temperature-sensitive allele to identify cell cycle stages that are sensitive to ER perturbations. In the yeast ER, the essential proteins Ero1p (ER oxidoreductin 1) and Pdi1p (protein disulfide isomerase 1) work together to catalyze oxidative protein folding (; ; ). For cells carrying the temperature-sensitive allele, growth at the restrictive temperature rapidly induces ER stress (). In asynchronous cultures, the restrictive growth of cells caused an accumulation of cells with a 2C or greater DNA content (Fig. S2, available at ). This suggests that ER stress delays cell cycle progression at a point subsequent to DNA replication. To specifically define this ER-sensitive stage of the cell cycle, we induced ER stress in α-factor synchronized cells (). When grown at the restrictive temperature, synchronized cells experienced severe ER stress, as measured by splicing (). Compared with wild-type cells, these ER-stressed cells proceeded normally through the initial stages of the cell cycle. By 30 min after the temperature shift, both cell types completed DNA replication, thus adopting a 90–95% 2C DNA content ( and quantitated in E). After 1 h of growth at 37°C, wild-type cells began to divide and reenter G1 phase. In contrast, only a small percentage of cells divided at 37°C. Instead, ER-stressed cells retained a 2C DNA content or began to acquire abnormally high amounts of DNA (). Microscopic examination of synchronized wild-type and cells revealed that ER-stressed cells were delayed with large buds and divided nuclei. After 30 min of 37°C growth, 90% of cells of each cell type had initiated bud formation (). After 45 min, both cell types remained budded, and, by this time, 60–70% of both cell populations had divided nuclei (). After 1 h, wild-type cells began to divide and become newly divided unbudded cells with a single nucleus. In contrast, cells did not divide but remained budded with divided nuclei for the remainder of the time course (), suggesting that ER stress slows the cell cycle at a point after nuclear division, probably during late M phase or cytokinesis. In fact, many cells began to adopt a multibudded morphology after 1.5 h of 37°C growth (). This multibudded morphology was never seen in wild-type cells. The appearance of extra buds coupled with the appearance of 3C/4C DNA peaks strongly suggests that cells initiate a new round of the cell cycle despite a block or delay in the previous cell division. To confirm that ER stress is specifically responsible for delaying the cell cycle in cells, we examined the effects of another well-characterized ER stress inducer, tunicamycin (Tm), on cell cycle progression. Tm inhibits N-linked glycosylation in the ER, which causes the accumulation of unfolded proteins. Consistent with previous studies (; ), we found that Tm inhibits the budding process when added immediately after α-factor release (Fig. S3, available at ). Budding inhibition is known to activate the morphogenesis checkpoint and induce a G2/M delay (), which would likely obscure a subsequent ER-induced delay. Therefore, we introduced Tm to synchronized cultures 30 min after G1 release, after cells had already initiated the budding process (Fig. S3). Tm treatment recapitulated the cell cycle effects of the mutation. As expected, Tm-treated cells displayed 90% mRNA splicing 1 h after α-factor release () and retained maximal UPR induction for the entire 3-h time course. Both Tm-treated and untreated synchronized cultures contained ∼90% 2C cells after 1 h, indicating that they had progressed through S phase and into G2/M phase (). After 1.25 h of growth, untreated cells began to divide, as indicated by the return to a 1C DNA content, and continued through the next cell cycle, ultimately losing synchronicity. Like cells, Tm-treated cells failed to divide and instead began to attain a 3C or 4C DNA content (). Untreated and Tm-treated cells were ∼90% budded after 1 h of synchronized growth (). After 1.5 h of growth, untreated cells divided and became unbudded before reentering the next cell cycle. Tm-treated cells remained 80–90% budded for the entire duration of the time course. Furthermore, after 1.75 h of growth, Tm-treated cells began to attain a multibudded morphology (). We also examined the timing and integrity of nuclear division in Tm-treated cells. In addition to following the segregation of DAPI bodies in these cells, we expressed a GFP fusion protein that localized to both copies of chromosome IV (see Materials and methods). This allowed us to visualize sister chromatids segregating to separate nuclei during nuclear division () to confirm that DNA segregation was occurring appropriately. We found that nuclear division occurred with the same kinetics in Tm-treated cells as in untreated cells, as both conditions allowed ∼45% of cells to divide their nuclei after 1 h of growth and ∼75% of cells to divide their nuclei after 1.25 h of growth (). After 1.5 h of growth, untreated cells divided to become unbudded cells with a single nucleus. Tm-treated cells continued to contain 70–80% divided nuclei for the remainder of the time course. Furthermore, we never observed DAPI bodies separating with improperly segregated sister chromatids, indicating that mitosis occurred properly in these Tm-treated cells (, white arrows denote GFP-marked chromosomes). Therefore, similar to cells grown at the restrictive temperature, cells experiencing ER stress as a result of Tm treatment were delayed with a budded morphology after nuclear division. Tm treatment and -restrictive growth had very similar effects on the cell cycle, strongly suggesting that these effects are the specific result of ER stress rather than ER-independent effects of Tm treatment or the allele. To verify that the cell cycle is sensitive specifically to ER stress, we examined the effects of Tm treatment on the cell cycle of synchronized Δ cells. Because is required for recovery from ER stress, Δ cells should be unable to recover from any specific effect of ER stress but should respond normally to ER-independent stimuli. Indeed, the absence of rendered cells incapable of recovering from the Tm-induced appearance of cells with a high DNA content. The percentage of 3C/4C cells in the wild-type Tm-treated populations peaked at 40% after 2 h of growth (see ) and then began to decline, reaching 25% after 3 h of growth. In contrast, Δ cells continued to be 40–45% 3C/4C for the entire 3-h time course. To distinguish between the possibilities of a late M-phase delay or a delay in cytokinesis, we examined the effect of ER stress on several mitotic events: Clb2p production/degradation, Cdc14p release, and mitotic spindle formation/depolymerization. Clb2p is a major regulator of cell cycle progression. Its levels increase as cells enter mitosis and decrease as cells exit mitosis. Cells delayed in mitotic exit typically display sustained high levels of Clb2p (). Directly after the temperature shift (0-h time point), both wild-type and cells contained very low levels of Clb2p (), which is consistent with most cells being in G1 or S phase. In both cell types, Clb2p levels began to increase 30 min after the temperature shift, marking mitotic entry 15 min before nuclear division ( and ). Similarly, Clb2p degradation, marking mitotic exit, occurred at the same time (60 min) in wild-type and cells. In wild-type cells, this Clb2p decrease correlated well with the onset of cytokinesis ( and ), but, in cells, cytokinesis did not occur. The key events of mitotic exit are signaled by the phosphatase Cdc14p, which is only active during anaphase. During all other times in the cell cycle, Cdc14p is kept inactive by virtue of its nucleolar localization. After nuclear division, Cdc14p is released into the nucleus and cytoplasm, where it signals multiple key cell cycle events, including the completion of Clb2 degradation, breakdown of the mitotic spindle, and cytokinesis (; ). 10 min after temperature shift, for both wild-type and cells, Cdc14p-GFP colocalized with a portion of the nucleus, which is consistent with the expected nucleolar localization of Cdc14p (). After 55 min of 37°C growth, both cell types released Cdc14p-GFP into their nucleus and cytoplasm, demonstrating that these conditions of ER stress did not delay Cdc14p release. Wild-type cells divided and resumed the nucleolar localization of Cdc14p by 70 min. Mutant cells also reabsorbed Cdc14p into the nucleolus at the 70-min time point but did not divide and eventually assumed a multibudded morphology ( and ). Finally, we used a GFP fusion gene () to examine the formation and breakdown of the mitotic spindle during ER stress. By 45 min after the temperature shift, both wild-type and cells exhibited fully formed mitotic spindles between their two spindle pole bodies, indicating that ER stress did not delay spindle formation. Spindle breakdown also occurred at the same time (75 min) in both cell types. Again, cells did not divide. In the absence of cell division, some cells rereplicated their spindle pole bodies, rebudded, and reformed a mitotic spindle, thus forming the unusual cells depicted in (150′ panel). We also examined Clb2 fluctuations, Cdc14p release, and mitotic spindle formation and breakdown in synchronized untreated and Tm-treated cells. We found that like , Tm had no effect on these mitotic markers (Fig. S4, available at ). Therefore, ER stress delays cell division but does not affect mitotic entry, mitosis, or mitotic exit, suggesting that ER stress specifically inhibits cytokinesis or cell separation. Cytokinesis creates a membrane barrier between mother and daughter cells. After cytokinesis, the septum continues to hold the two cells together; the septum must be degraded for cell separation to occur (). Experimentally, lyticase can be used to degrade the septum of delayed cells, thus differentiating between a cytokinesis defect and a defect in cell separation. Lyticase treatment demonstrated that ER-stressed cells fail to divide because of incomplete cytokinesis rather than incomplete cell separation. We collected cells 2.5 h after temperature shift as described in except that α factor was added back to the medium 45 min after G1 release to prevent the initiation of a second cell cycle. As before, most cells were delayed with a budded morphology at this time point. Their delay was clearly caused by a cytokinesis defect, as 79% of these budded cells were resistant to cell separation by lyticase treatment (). Confirming that lyticase treatment only separated cells that had completed cytokinesis, wild-type cells in M phase (collected 1 h after α-factor release) remained 96% budded after lyticase treatment. In addition, Δ cells, which are known to be defective in cell separation (), were 43% budded 1.5 h after α-factor release (unpublished data). Of the budded Δ cells, 86% were separated by lyticase (), confirming that the experimental conditions used here were sufficient to dissociate the majority of separation-defective cells. Successful cytokinesis requires that cortical actin patches become polarized to either side of the bud neck late in the cell cycle (; ; ; ; ). We followed actin patch localization in synchronized cells and found that wild-type and cells displayed bud-localized cortical actin patches throughout S, G2, and most of M phase (). Just before cytokinesis, the actin patches of cells redistributed to the bud neck in a manner indistinguishable from wild-type cells (). Therefore, the ER stress–induced cytokinesis defect is not caused by a delay or alteration in actin patch redistribution. The induction of ER stress in synchronized cell populations revealed that cytokinesis is highly sensitive to the state of the ER. This suggests that ER capacity increases during cell division, a process that might be facilitated by UPR signaling. To determine whether UPR signaling affects cytokinesis during normal cell growth, we examined cytokinesis in Δ strains. In the absence of any external ER stressor, wild-type cell populations never exhibited cells with a >2C DNA content. In contrast, after 1.5 h of normal synchronized growth, 15% of untreated Δ cells were >2C. This number increased to 20% after 2 h of growth and remained ∼20% until the end of the 3-h time course (). Untreated Δ cells were almost as cytokinesis deficient as wild-type cells treated with Tm (, compare wild-type +Tm to Δ −Tm). Furthermore, we examined Δ and Δ strains for the multibudded morphology that is indicative of cells with a cytokinesis defect. We found that a small percentage of cells (<1%) did display this multibudded morphology, whereas we never observed multibudded cells in wild-type populations (). A complete cytokinesis block should cause a much higher percentage of cells to attain multiple buds. Therefore, UPR mutants are delayed in cytokinesis rather than blocked. To further investigate the link between UPR signaling and the cytokinesis process, we measured basal UPR activity in various cytokinesis mutants using a 4× UPRE-GFP reporter construct (). , , , , and all participate in cytokinesis (see Discussion). Of the cytokinesis mutants tested, Δ and Δ strains did not exhibit basal UPR activity (). However, in the absence of any external ER stress induction, Δ, Δ, and Δ strains exhibited three- to sixfold UPR reporter gene expression compared with wild-type cells. This level of reporter activity reflects a true link between the UPR and cytokinesis, as Δ and Δ mutants, which are ERAD deficient and are known to induce functionally relevant levels of UPR activity (; ), exhibited similar levels of reporter gene expression. The finding that some cytokinesis mutants exhibit UPR activation is quite novel: the detection of basal UPR activity has been previously limited to mutants with specific ER defects. The UPR's role in cytokinesis, which is revealed in this study, represents a novel type of UPR activity, as it can be detected during optimal unstressed growth conditions. All previous studies of UPR mutants describe their inability to respond to unusually stressful growth conditions such as inositol starvation (; ), drug treatments that induce widespread protein misfolding (; ), overexpression of a misfolded mutant protein (; ; ), or development into a specialized secretory cell (; ; ; ). Each of these known UPR-requiring conditions imposes a massive load on the ER. The newly discovered importance of UPR signaling during normal cell growth uncovers a novel housekeeping function for the UPR pathway. In addition to responding to stressful growth conditions, the UPR must monitor and manage the cell's fluctuating ER requirements. The UPR's ability to serve a housekeeping function sheds new light on the mode of UPR activation. In theory, the UPR pathway might operate according to one of two modes of activation. It could activate in a manner similar to an on/off switch. In this case, the pathway remains “off” until a threshold level of stress is experienced, at which point the pathway turns “on” and becomes highly active. Alternatively, the UPR pathway might operate as a dimmer switch in which the off state and on state actually represent two extremes on a continuum. Previous studies have investigated the UPR pathway by inducing crisis levels of ER stress (; ; ; ). If the UPR pathway could fine tune the level of ER function, this could actually prevent such an ER crisis by allowing the gradual adaptation of ER capacity. Data from previous studies provide support for both modes of activation. In support of the on/off switch mode of activation, mRNA remains unspliced during normal cell growth but becomes rapidly and efficiently spliced upon treatment with DTT or Tm or upon the overexpression of misfolded proteins (). In addition, certain modest amounts of ER stress have been shown to not activate the UPR pathway at all. For example, expression of the misfolded mutant protein CPY* from its genomic locus does not activate UPR signaling, and ERAD of genomic CPY* does not require UPR components (). However, data are also accumulating to support the dimmer switch mode of UPR activation. For example, certain mutations in the ERAD pathway have been shown to induce intermediate levels of UPR activity (; ; ). Our study further supports the dimmer switch mode of UPR activation, as we have shown that subtle activation of the UPR pathway contributes to efficient cytokinesis. Although DNA replication, mitotic entry, spindle formation, nuclear segregation, Cdc14p release, mitotic exit, spindle disassembly, and actin patch repolarization all occur normally during ER stress, cytokinesis does not (summarized in ). Therefore, we have found that ER stress specifically disrupts cytokinesis, and we have ruled out the possibility that this disruption is caused by a defect in actin patch relocalization. This disruption could be caused by a stress-induced attenuation of any of the ER's many functions, including secretion, ERAD, or phospholipid metabolism. Despite the ER's well-characterized role in initiating protein secretion, it remains unknown whether ER stress inhibits the entire secretory pathway. If it does, there are several reasons that this may impact cytokinesis. Cytokinesis begins with the assembly and contraction of an actomyosin ring. In animal cells, it has been shown that membrane deposition at the cleavage furrow must accompany actomyosin ring contraction for proper cytokinesis to occur (; ). The extra membrane, which is delivered in the form of secretory vesicles, presumably relieves the tension created by membrane constriction. Perhaps, as in animal cells, the yeast secretory pathway assists in cytokinesis by providing new membrane to the site of ring contraction, and it is the lack of membrane at the bud neck that prevents cytokinesis under conditions of ER stress. Regardless of whether membrane addition itself is required for yeast cytokinesis, it is clear that Golgi-derived vesicles are targeted to the yeast bud neck at the end of the cell cycle and that these vesicles assist in the process of cytokinesis. First, vesicles carry cargo that is necessary for actomyosin ring contraction. Cells that are defective in vesicle fusion assemble an actomyosin ring normally, but the assembled ring is unstable and does not properly contract (). Second, during cytokinesis, secretory vesicles provide the yeast bud neck with the enzymes responsible for septum formation, a process that is essential for yeast cytokinesis (; ; ). Therefore, if ER stress disrupts vesicle trafficking, this could slow membrane deposition, ring contraction, and/or septation and, thereby, delay cytokinesis, thus explaining the results of our study. This explanation implies that during normal cytokinesis, the UPR manifests its housekeeping function by increasing the cell's secretory capacity, thus fulfilling the enhanced secretory requirements of cytokinesis. Despite expectations that ER stress would broadly inhibit secretion, some studies find that ER stress has a minimal impact, if any, on the overall secretory pathway (; ). This suggests that the ER might play a role in cytokinesis through one of its cellular functions besides protein folding and trafficking. This possibility is especially intriguing, as it implies that the UPR pathway can detect ER functional cues other than the simple accumulation of unfolded proteins in the ER. Although previous studies have not tested this prospect directly, UPR target genes represent the entire spectrum of ER functions (). In addition to functioning in protein folding and secretion, the ER has the task of regulating phospholipid metabolism. Because cytokinesis entails a membrane fusion event and the creation of a membrane barrier between mother cell and daughter cell, it is not surprising that certain phospholipids are necessary for its proper completion. Phosphatidylethanolamine and phosphatidylinositol 4,5-bisphosphate become locally concentrated to the cleavage furrow during cytokinesis in various eukaryotic cell types. Interfering with the production of either of these two phospholipids results in a cytokinesis defect (; ; ; ). Therefore, the disruption of cytokinesis by ER stress may be caused by the effects of ER stress on phospholipid metabolism. If this is the case, the UPR's role during normal cytokinesis may be to up-regulate genes involved in phospholipid metabolism. Three cytokinesis mutants, Δ, Δ, and Δ, exhibit constitutive UPR activity. Strains deleted for or , which are involved in the cytokinesis processes of actomyosin ring disassembly and septum formation, respectively (; ), did not activate the UPR. During yeast cytokinesis, promotes actomyosin ring assembly (), coordinates ring contraction with septum formation (; ), and mediates septum formation (). There is no indication that any of these mutants are defective in protein secretion or any other aspect of ER function. This is the first instance of UPR activity in mutants that are not directly defective in an ER-associated function. Furthermore, unlike previous cases of basal UPR activity in mutants, none of these three genes is a UPR target gene (). Therefore, the UPR induction in these mutants does not represent the cell's attempt to transcriptionally activate the specific gene that is absent. Increased UPR activity in Δ, Δ, and Δ strains probably helps these cells partially overcome their cytokinesis defect. This implies that the UPR pathway can directly or indirectly sense and modify the cell's cytokinesis efficiency. Yeast strains containing MNY numbers were in the W303 strain background, and strains containing RHY numbers were derived from the S288C strain background. All strains were generated using standard genetic methods and are listed in . MNY1008 and MNY1009 were constructed by integrating StuI-linearized pAFS125 () at the locus. All strains carrying the UPRE-GFP reporter were constructed by integrating StuI-linearized pJCI86-GFP () at the locus. Wild-type and CDC14-GFP strains were constructed using a one-step PCR-mediated technique (). All deletion strains were constructed by amplification of the Research Genetics heterozygous diploid collection followed by G418 selection and verification by PCR. For Western blot analysis, ∼3 × 10 cells were harvested by centrifugation at 4°C, washed with 1 ml HO, frozen with liquid N, and stored at −80°C. Pellets were resuspended in 100 μl of lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM EDTA, 1% NP-40, 1 mM sodium pyrophosphate, 1 mM PMSF, 1 mM sodium orthovanadate, 2 μg/ml pepstatin A, 2 μg/ml leupeptin, 20 mM NaF, 5 μg/ml aprotinin, and 1.75 mM β-glycerophosphate). 100 μl of acid-washed glass beads were added, and cells were vortexed at 4°C for 5 min. Lysates were centrifuged at 13,000 for 8 min at 4°C, and the supernatant was collected. Protein concentration was determined using a BCA protein assay kit (Pierce Chemical Co.). 30 μg of protein was denatured at 95°C in 2× loading buffer (125 mM Tris-HCl, pH 6.8, 2% SDS, 50% glycerol, 12% β-mercaptoethanol, and 0.02% bromophenol blue) and was loaded on a 10% SDS-polyacrylamide gel (Invitrogen). Clb2p was detected with a 1:1,000 dilution of anti-Clb2 antibody (Santa Cruz Biotechnology, Inc.) followed by anti–rabbit secondary antibody at a dilution of 1:10,000 (GE Healthcare) and ECL detection (GE Healthcare). RNA isolation and Northern blotting were performed essentially as described previously () and were quantified using a phosphorimager (Typhoon; GE Healthcare). Approximately 10 cells were collected by centrifugation at 4°C, washed with 1 ml of ice-cold HO, and resuspended in 400 μl of cold HO. 1 ml of ice-cold EtOH was added slowly, and cells were fixed at 4°C overnight or longer. After fixation, cells were collected by centrifugation, washed with 1 ml PBS, and treated with 1 mg/ml RNase A in 100 μl PBS at 37°C for 2–12 h. Cells were then treated with 5 mg/ml pepsin in 200 μl HO, pH 2.0, at 37°C for 20 min followed by washing and resuspension in 1 ml PBS. Cells were sonicated for 15 s at 15%. 100 μl of cells (10 cells) were stained with 1 μM Sytox green (Invitrogen) in PBS. Data were collected using a flow cytometer (FACSCalibur; BD Biosciences) and analyzed using FlowJo software (Tree Star). Cells were fixed in 4% PFA and sonicated briefly before analysis. Budding index was calculated as the number of cells with an obvious bud divided by the total number of cells counted. For visualization of nuclei, DAPI was added to a concentration of 0.04 μg/ml. Nuclear division was scored as positive when two separate DAPI bodies were present in a single cell. To visualize sister chromatid segregation, MNY1005 cells expressed a LacI12-GFP fusion protein and contained a Lac operon at the TRP1 locus. This caused both copies of chromosome IV to be GFP marked (). For the visualization of actin, cells were fixed in 4% PFA/PBS, washed with PBS, and incubated with 6.6 μM AlexaFluor546-phalloidin (Invitrogen). All cells were visualized using a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with a 100× 1.3 NA objective. Images were captured using a monochrome digital camera (Axiocam; Carl Zeiss MicroImaging, Inc.) and analyzed using Axiovision software (Carl Zeiss MicroImaging, Inc.). Cells were fixed in YPD/4% formaldehyde for 10 min followed by 1 h in 400 mM KHPO, pH 6.5, 500 μM MgCl, and 4% formaldehdye. Cells were then washed in 400 mM KHPO, pH 6.5, and 500 μM MgCl and resuspended in 400 mM KHPO, pH 6.5, 500 μM MgCl, and 1 M sorbitol. Fixed cells were sonicated (15% for 15 s) and treated with 80 U/ml lyticase at 37°C for 1 h. Fig. S1 shows by RT-PCR that spliced mRNA is present in unstressed wild-type cells but not in Δ cells. Fig. S2 shows that asynchronous cells accumulate with a 2C or greater DNA content when shifted to restrictive growth. Fig. S3 demonstrates that Tm inhibits budding when added directly after α-factor release but has no effect on DNA replication. This budding inhibition is bypassed when Tm is added 30 min after α-factor release. Fig. S4 shows that Tm treatment does not affect Clb2p production/degradation, Cdc14p release, mitotic spindle formation/depolymerization, or actin patch relocalization but still inhibits cytokinesis. Online supplemental material is available at .

Oxygen homeostasis is important for normal cellular function (). As oxygen levels decrease in the surrounding environment (hypoxia), cells respond by activating hypoxia- inducible factor (HIF) dependent gene transcription to facilitate cellular adaptation to hypoxia. HIF is a heterodimer of two basic helix-loop-helix/Per/Arnt/Sim domain proteins, HIF-α and the aryl hydrocarbon nuclear translocator (ARNT or HIF-β; ). Under normal oxygen conditions, ARNT is constitutively stable, whereas the α subunit is labile. In normal oxygen conditions, the α subunit is hydroxylated at proline residues by a family of prolyl hydroxylase enzymes (PHDs). Proline hydroxylation targets the protein for ubiquitination by the von Hippel-Lindau protein (pVHL)/E3 ubiquitin ligase and for subsequent proteasomal degradation (; ; ; ; ). The α subunit is also hydroxylated at an asparagine residue by the enzyme factor inhibiting HIF-1 (FIH-1) under normal oxygen conditions (; ,). Asparagine hydroxylation blocks the binding of the transcriptional coactivators p300 and CREB binding protein (CBP) to HIF-1 (; ). Under hypoxic conditions, the α subunit is not hydroxylated by the PHDs or FIH, resulting in the stabilization of the HIF-α protein, dimerization with ARNT, and association with p300/CBP to initiate gene transcription. The mechanism by which cells transduce the hypoxic signal to activate HIF is a subject of ongoing research. Previous studies indicate that mitochondria are involved in the transduction of hypoxic signals; however, the mechanism is not fully understood. One model proposes that mitochondria regulate the ability of the PHDs to hydroxylate HIF-1α protein because of their capacity to consume oxygen (; ). Mitochondrial oxygen consumption would generate a gradient of oxygen within the cytosol of the cell, thereby limiting the availability of oxygen, a necessary cosubstrate for PHD activity. Another model proposes that mitochondria increase the levels of cytosolic reactive oxygen species (ROS) during hypoxia to activate HIF (). Initial evidence to support this model came from observations that cells treated with mitochondrial electron transport inhibitors, and cells devoid of mitochondrial DNA (ρ cells), fail to activate HIF during hypoxia because of a lack of mitochondrially generated ROS (, ). Recently, three independent studies genetically confirmed the initial finding that mitochondria-generated ROS are required for hypoxic activation of HIF. Cells devoid of the cytochrome gene do not increase cytosolic ROS or stabilize HIF-1α in hypoxic conditions (). Cells with either transient or stable RNAi of the Rieske Fe-S protein, a component of the mitochondrial complex III (the bc complex), inhibits hypoxic increase of cytosolic ROS and stabilization of HIF-1α protein (; ). Although these studies indicate that ROS generated within the mitochondrial electron transport chain are required to relay the hypoxic signal, they do not indicate which complex of the electron transport chain is the site of ROS generation. The mitochondrial electron transport chain generates superoxide at complexes I, II, and III (). Complexes I and II generate superoxide within the mitochondrial matrix (; ; ; ; ; ; ). Complex III generates superoxide at the Q site, resulting in the release of superoxide into either the intermembrane space or the matrix (; ; ; ; ; ). Complex IV has not been reported to generate ROS; however, cytochrome was recently demonstrated to participate in the generation of hydrogen peroxide by providing electrons to p66 Shc (). The observations that loss of cytochrome gene or RNAi of the Rieske Fe-S protein prevent hypoxic stabilization of the HIF-1α protein implicate either complex III or cytochrome as the source of ROS generation required for hypoxic stabilization of the HIF-1α protein. In the present study, we examined which site within the mitochondrial electron transport chain is required for the generation of ROS and hypoxic stabilization of the HIF-1α protein independently of respiration. ). The electron flux from ubiquinol (QH) to cytochrome occurs through the ubiquinone (Q) cycle within complex III (). /cytochrome axis transiently, making the radical ubisemiquinone. The second electron from ubisemiquinone is transferred to cytochrome . However, ubisemiquinone does have the capability of transferring an electron to oxygen to generate superoxide. This allows for the generation of ROS at the Q site of complex III through the interaction between ubisemiquinone (Q) and molecular oxygen within the bc complex (). To explore the role of complex III in the stabilization of the HIF-1α protein, we used cells that are deficient in cytochrome . These cells are cybrids that were generated by repopulating 143Bρ cells with mitochondria that contain either a wild-type (WT) mitochondria DNA or a 4-base pair deletion of the cytochrome gene (). The cytochrome –deficient cells (ΔCyt ) do not consume oxygen, similar to ρ cells (). However, the ΔCyt cybrid cells retain the ability to stabilize HIF-1α protein under hypoxia (). These data indicate that the ability of cells to consume oxygen is not related to their ability to stabilize HIF-1α protein. Moreover, under hypoxia, the ΔCyt cybrids increase HO levels measured in the cytosol using Amplex red (). ρ cells did not display an increase in ROS in the cytosol during hypoxia, indicating that mitochondria are the major source of ROS production during hypoxia. These data indicate that that the ability of mitochondria to increase cytosolic ROS and stabilize HIF protein in hypoxia is independent of both cytochrome and oxygen consumption. Additionally, the levels of cytosolic antioxidant proteins Cu/Zn superoxide dismutase and catalase did not change drastically in hypoxic conditions (Fig. S1, available at ). To determine whether mitochondrial ROS generation was responsible for the increase in cytosolic ROS and stabilization of the HIF-1α protein in the ΔCyt cybrids, these cells were treated with the mitochondrial-targeted antioxidant MitoQ. MitoQ abolished the increase in cytosolic ROS during hypoxia in the ΔCyt cybrids (). Hypoxic stabilization of the HIF-1α protein was diminished in both WT and ΔCyt cybrid cell lines treated with MitoQ but not in the presence of dimethyloxalylglycine (DMOG; ). MitoQ diminished stabilization in a dose-dependent manner in both cell types (Fig. S2). These results are corroborated with the observation that EUK-134, a mimetic of both catalase and superoxide dismutase, inhibits hypoxic stabilization of HIF-1α protein in both the WT and the ΔCyt cybrids (). To further validate that the mitochondrial electron transport chain is required for hypoxic stabilization of the HIF-1α protein in the ΔCyt cybrids, we used short hairpin RNA (shRNA) against the mitochondrial transcription factor A (TFAM). TFAM is required for the proper transcription and replication of mitochondrial DNA (). In the absence of TFAM, cells become depleted of their mitochondrial DNA (ρ cells; ). Expression of TFAM shRNA in the ΔCyt cybrids lowered TFAM mRNA expression and mitochondrial copy number by 75% compared with cells expressing the control shRNA against HIF (dHIF; ). The cell containing TFAM shRNA diminished their ability to stabilize HIF-1α protein under hypoxia (). These data indicate that mitochondrial electron transport has an important role in hypoxic stabilization of HIF-1α protein. To determine whether ROS generation from the Q site is responsible for the increase in cytosolic ROS and stabilization of the HIF-1α protein, the ΔCyt cybrid cells were stably infected with retrovirus containing shRNA against the Rieske Fe-S protein. In the absence of the Rieske Fe-S protein, the Q cycle is not initiated and ROS are not generated at the Q site. It is theoretically possible that the Q site might generate ROS. However, activity of the Q site is abolished in cells deficient in the cytochrome protein (). Therefore, the data indicate that the Q site is dispensable for hypoxic increase in cytosolic ROS and stabilization of HIF-1α. Stably expressing a shRNA against the Rieske Fe-S protein in the ΔCyt cybrid cells decreases expression of the Rieske Fe-S protein (). These cells do not stabilize the HIF-1α protein when exposed to hypoxia but retain HIF-1α protein stabilization in the presence of DMOG (). As expected, neither the TFAM shRNA cells nor Rieske Fe-S shRNA cells were able to increase cytosolic ROS under hypoxic conditions (). To ensure that our results were not due to any adaptation to the loss of cytochrome protein or shRNA against Rieske Fe-S protein, we corroborated our genetic findings in WT cells using well-established pharmacological inhibitors of complex III. Incubating WT cells with the complex III inhibitor stigmatellin, which binds to the Q site, inhibits hypoxic stabilization of HIF-1α protein (). In contrast, the complex III inhibitor antimycin A, which preserves the ROS generation at the Q site of complex III, did not decrease hypoxic stabilization of HIF-1α in the WT cybrids (). Collectively, these data implicate the Q site of complex III as the primary site of ROS generation for hypoxic stabilization of HIF-1α protein. Under normal oxygen conditions, HIF-1α protein is hydroxylated by the PHDs, thereby facilitating ubiquitination and subsequent proteasomal degradation. Exogenous ROS are sufficient to stabilize HIF-1α protein under normal oxygen conditions (). Using an antibody that specifically recognizes HIF-1α protein hydroxylated on proline 564, we demonstrate that ROS inhibit the ability of the PHDs to hydroxylate HIF-1α protein. Quenching the increase in cytosolic ROS under hypoxia with MitoQ recovers hydroxylation of HIF-1α protein in both the WT and ΔCyt cybrids (). To test whether exogenous ROS are sufficient to prevent hydroxylation of the HIF-1α protein, cells were exposed to glucose oxidase, an enzyme that generates HO. Addition of 10 μg/ml glucose oxidase increases intracellular ROS to levels that are similar to those measured in hypoxic conditions (). These levels of ROS generated in normal oxygen conditions are sufficient to stabilize HIF-1α protein (). Addition of the antioxidant protein catalase in this experiment inhibits stabilization of HIF-1α protein, indicating that HO is responsible for the stabilization of the HIF-1α protein. The presence of glucose oxidase attenuates hydroxylation of HIF-1α protein, as assessed by reactivity with the hydroxylation-specific antibody (). The addition of catalase in the presence of glucose oxidase under normal oxygen conditions recovers hydroxylation of HIF-1α protein, indicating that the ability of the PHDs to hydroxylate HIF-1α protein is indeed regulated by ROS. Previous findings indicate that loss of cytochrome prevents the hypoxic stabilization of the HIF-1α protein (). To determine whether cytochrome –generated ROS are required for the hypoxic stabilization of the HIF-1α protein, we exposed ρ cells to normoxia or hypoxia in the presence of ,,′,′-tetramethyl--phenylenediamine (TMPD). ρ cells lack a functional complex III and IV, resulting in a loss of electron flux through cytochrome . TMPD donates electrons to cytochrome , thereby fully reducing cytochrome . The electrons from cytochrome could then be donated to p66 Shc in the absence of a functional complex IV, resulting in ROS generation and HIF stabilization. WT 143B or A549 cells stabilize the HIF-1α protein during hypoxia (1.5% O) or in the presence of DMOG (). In contrast, the 143Bρ or A549ρ cells do not stabilize HIF-1α protein during hypoxia. The addition of TMPD to either 143Bρ or A549ρ cells also did not stabilize the HIF-1α protein under normoxia (21% O) or hypoxia (1.5% O; ). This was not due to reduced levels of p66Shc or cytochrome in the ρ cells (). TMPD did reduce cytochrome , and under these conditions it did not generate ROS (Fig. S3, available at ). Previous results indicate that ROS due to electron transfer between cytochrome and p66shc is observed only during DNA damage and may not be the normal physiological response in healthy cells (). These results indicate that reduction of cytochrome is not sufficient for hypoxic stabilization of HIF-1α protein. Mammalian cells transduce signals that couple decreases in oxygen levels to initiate HIF-dependent gene expression. The mechanism of how cells transduce hypoxic signals is not fully understood. Mitochondrial electron transport chain has been proposed as part of the hypoxic signal transduction machinery. Indeed, previous genetic evidence indicates that loss of cytochrome or the Rieske Fe-S protein prevents the hypoxic stabilization of HIF-1α, indicating that these proteins are involved in the increase in cytosolic ROS during hypoxia (; ; ). RNAi of the Rieske Fe-S protein or loss of cytochrome prevents the formation of ubisemiquinone, thus preventing ROS generation at the Q site of complex III. The loss of cytochrome or RNAi of Rieske Fe-S protein would also not cause a reduction in cytochrome , thereby preventing cytochrome –dependent ROS generation. In the present study, we demonstrate that cytochrome is not the primary site of ROS generation in hypoxia. We demonstrate that fully reducing cytochrome levels is not sufficient to stabilize the HIF-1α protein during hypoxia or normoxia. Therefore, cytochrome does not contribute to hypoxic signal transduction through its ability to generate ROS via p66Shc. Our data indicate that the Q site of complex III is part of the hypoxic signal transduction machinery. Cells deficient in cytochrome protein are able to generate ROS at the Q site of complex III. During hypoxia, these cells stabilize the HIF-1α protein. Preventing ROS generation at the Q site in the cytochrome –deficient cells with MitoQ or by shRNA against the Rieske Fe-S protein prevents the increase in cytosolic ROS and stabilization of the HIF-1α protein during hypoxia. The present data also demonstrate that ROS regulate the ability of the PHDs to hydroxylate HIF in both normoxic and hypoxic conditions. Quenching ROS with MitoQ in hypoxic conditions allows for continued hydroxylation of HIF-1α protein, whereas addition of exogenous ROS in normal oxygen conditions inhibits the ability of the PHDs to hydroxylate HIF-1α protein. These data suggest that the Q site of the bc complex participates in hypoxic signal transduction via ROS generation to initiate HIF-mediated transcriptional responses that facilitate cellular adaptation to low oxygen. The present data are in contrast with other groups that have proposed that the ability of mitochondria to consume oxygen is the major requirement for stabilization of the HIF-1α protein in hypoxic conditions. Their model proposes that respiring mitochondria generate an oxygen gradient, preventing hydroxylation, and thereby increasing stabilization of the HIF-1α protein (; ). According to this model, in the absence of a functioning respiratory chain, the oxygen gradient would be reduced, resulting in hydroxylation and degradation of the HIF-1α protein. However, the cytochrome –null cells are respiratory incompetent and therefore unable to generate an oxygen gradient. Contrary to this model, these cells still retain the ability to stabilize the HIF-1α protein during hypoxia. Furthermore, the mitochondrial-targeted antioxidant MitoQ or the cytosolic antioxidant EUK-134 prevents stabilization of the HIF-1α protein in the cytochrome –deficient cells, indicating ROS involvement in HIF stabilization. MitoQ has been shown to prevent hypoxic stabilization of the HIF-1α protein in respiratory-competent cells, demonstrating the importance of ROS in HIF-1α protein stabilization (). However, there are instances when an oxygen gradient created by the mitochondria during normoxia can create a hypoxic environment within cells, causing HIF-1α protein accumulation (). For example, if metabolically active cells are cultured at high confluency, their demand for oxygen exceeds the supply of oxygen, resulting in a local hypoxia. Under these conditions, respiratory inhibition would result in restoration of the oxygen levels to normoxia within the cells, resulting in the degradation of the HIF-1α protein. Cells that are cultured at a high confluency under hypoxia (1–2% O) would experience anoxia (0% O). Under these conditions, respiratory inhibition would result in restoration of oxygen levels only to the hypoxic levels. If respiratory inhibition does not result in attenuating ROS generation, such as in the cytochrome –deficient cells, then cells would still be able to stabilize the HIF-1α protein in conditions of high confluency under hypoxia. Collectively, our data indicate that the ability of mitochondria to generate ROS and not an oxygen gradient is required for the stabilization of the HIF-1α protein during hypoxia. In summary, we demonstrate that the Q site of complex III is necessary to increase cytosolic ROS in hypoxic conditions, which results in the inhibition of the ability of the PHDs to initiate degradation of HIF-1α protein (). We also conclusively demonstrate that the ability to consume oxygen by mitochondria is not required for hypoxic stabilization of the HIF-1α protein. The link between ROS and the PHDs is currently unknown. As oxygen levels fall, the enzymatic activity of PHDs decrease (). It is possible that the link could be an oxidant-dependent signaling pathway in which a posttranslational modification of the PHDs, such as phosphorylation, turns off the catalytic activity. In fact previous studies have implicated multiple signaling molecules that are required for hypoxic activation of HIF-1 (; ; ; ; ). Alternatively, the link between ROS and the PHDs could be due to changes in the cytosolic redox state. The ROS may induce a shift in iron redox state from Fe to Fe as a result of the Fenton reaction, thereby limiting an essential cofactor of the PHDs, resulting in an inhibition of hydroxylation of HIF protein (). It could also be that the low oxygen levels decrease PHD activity and the ROS produced during hypoxia further decrease PHD activity to prevent hydroxylation of HIF-α protein. Furthermore, multiple factors affecting cellular redox state and metabolism are likely to affect hydroxylation of the HIF-1α protein (). Our study also suggests that the targeting of mitochondrial ROS could serve as a therapeutic target for many HIF-dependent pathological processes, including cancer. It will be of interest in future studies to examine whether the Q site of complex III serves as part of a signal transduction machinery for other hypoxia-initiated cellular events, such as calcium signaling. WT A549 and 143B cells were cultured in DME, whereas the ρ derivatives were cultured in DME supplemented with 100 μg/ml uridine. The ρ derivatives were made as previously described (). The WT and ΔCyt 143B cybrid cells (provided by I.F.M. de Coo, University Medical Center, Rotterdam, Rotterdam, Netherlands) were previously described by and were cultured in DME supplemented with 100 μg/ml uridine. Cells were cultured at 37°C in 5% CO humidified incubators for normoxic conditions. Hypoxic conditions (1.5% O) were achieved in the humidified variable aerobic workstation InVivo (Biotrace). Glucose oxidase, catalase, stigmatellin, and antimycin A were purchased from Sigma-Aldrich. The pSiren retroviral vector (CLONTECH Laboratories, Inc.) was used to express shRNA sequences for the Rieske Fe-S (5′-AAGGTGCCTGACTTCTCTGAA -3′), TFAM (5′-GTTGTCCAAAGAAACCTGT-3′), and HIF (5′-GCCTACATCCCGATCGATGATG-3′). Stable lines were generated using retroviral infection methods with the PT67 packaging cell line (CLONTECH Laboratories, Inc.). Intracellular ROS was measured using Amplex red (Invitrogen) according to manufacturer's protocol. In brief, cells were lysed in 100 μM Amplex red solution supplemented with 2 U/ml HRP and 200 mU/ml superoxide dismutase (OXIS International) and incubated in the dark for 30 min. Fluorescence was measured in a plate reader (SpectraMax Gemini; Molecular Devices) with excitation of 540 nm and emission of 590 nm. Alternatively, ROS were measured using 10 μM of the oxidant-sensitive fluorescent probe 5-(and-6)-chloromethyl-2′,7′-dichlorodihydrofluorescein diacetate acetyl ester (CM-HDCFDA; Invitrogen). Cells were plated at equal density in 60-mm plates. The next day, cells were incubated continuously with CM-HDCFDA for 2 h and exposed to different experimental conditions. Subsequently, cells were washed and lysed in 0.1% Triton X-100. Fluorescence was measured in the SpectraMax Gemini plate reader with excitation of 488 nm and emission of 530 nm. Total RNA was isolated using the Aurum Mini kit (Bio-Rad Laboratories), and cDNA was generated using the RNAqueous-4PCR system (Ambion) according to manufacturer's protocol. Prepared cDNA was analyzed for TFAM mRNA using SYBR Green Master Mix (Bio-Rad Laboratories). Cycle threshold (C) values for TFAM were normalized to C values for ribosomal protein L19, and data were analyzed using the method. Primers used were as follows: TFAM, 5′-AAGATTCCAAGAAGCTAAGGGTGA-3′ and 5′-CAGAGTCAGACAGATTTTTCCAGTTT-3′; and RPL19, 5′-GTATGCTCAGGCTTCAGAAGA-3′ and 5′-CATTGGTCTCATTGGGGTCTAAC-3′. Protein levels were analyzed in whole cell lysates obtained using lysis buffer (Cell Signaling), and 50 μg of samples were resolved on a SDS polyacrylamide gel. Gels were analyzed by immunoblotting with antibodies for HIF-1α, cytochrome , p66Shc (BD Biosciences), Rieske Iron Sulfur protein (Invitrogen), and hydroxylated HIF-1α (a gift from P. Ratcliffe, University of Oxford, Oxford, UK), and α-tubulin (Sigma-Aldrich) was used as a loading control. Cellular O consumption rates were measured in aliquots of 1–3 × 10 subconfluent cells removed from flasks and studied in a magnetically stirred, water-jacketed (37°C) anaerobic respirometer fitted with a polarographic O electrode (Oxytherm system; Hansatech Instruments). Oxygraph Plus software was used to determine oxygen consumption rate. The number of mitochondria was determined by analyzing the abundance of the mitochondrial DNA encoded gene cytochrome oxidase subunit 1 (COX1) relative to the nuclear gene 18S. Total DNA was isolated using the DNEasy Tissue kit (QIAGEN). 10 ng of total DNA was subjected to quantitative Real-Time PCR using Sybr Green Chemistry. Primers used were as follows: 18S forward, 5′-ACAGGATTGACAGATTGATAGCTC-3′; 18S reverse, 5′-CAAATCGCTCCACCAACTAAGAA-3′; COX1 forward, 5′-CCCACCGGCGTCAAAGTATT-3′; and COX1 reverse, 5′-TTTGCTAATA- CAATGCCAGTCAGG-3′. The data presented are means ± SEM. Data were analyzed by two-way analysis of variance using Graph Pad Prism 4. When the analysis of variance indicated a significant difference, individual differences were explored with paired test. Statistical significance was determined at the 0.05 level. Fig. S1 shows antioxidant protein profile of whole cell lysates from WT cybrids and ΔCyt cybrids subjected to either 21% O (N) or 1.5% O (H) for 4 h. Fig. S2 shows HIF-1α protein levels in WT and ΔCyt cybrids subjected to 21% O (N), 1.5% O (H), or 1 mM DMOG (D) for 4 h at various concentrations of MitoQ. Fig. S3 shows ρ cells treated with TMPD/ascorbate for 15 min and then subjected to either 21 or 1.5% O; cytochrome redox state in isolated mitochondria or ROS measurement via DCFH oxidation was then assessed in whole cells. Online supplemental material is available at .

Stem cells have the unique capacity to self-renew and generate committed, transit amplifying (TA) progenitors that differentiate into the cell lineages of the tissue of origin (; ; ; ). The most important function of TA cells is to increase the number of differentiated progeny produced by each stem cell division, thus enabling stem cells to divide infrequently, at least under normal tissue homeostasis. The cornea provides an ideal experimental system for studying stem cells of human stratified epithelia (). Human corneal stem cells are segregated in the basal layer of the limbus, which is the vascularized zone encircling the cornea and separating it from the bulbar conjunctiva. The corneal epithelium lies on the avascular Bowman's membrane and is formed by TA keratinocytes that migrate millimeters away from their parental limbal stem cells (; ; ; ). Clonal analysis of squamous human epithelia, including the cornea, has identified three types of clonogenic keratinocytes, giving rise to holoclones, meroclones, and paraclones in culture (; ). Holoclone-forming cells have all the hallmarks of stem cells, including self-renewing capacity (; ), telomerase activity (), and an impressive proliferative potential—a single holoclone can generate the entire epidermis of a human being (). Holoclone-forming cells generate all the epithelial lineages of the tissue of origin (; ; ; ), permanently restore massive epithelial defects (; ; , ; ), and can be retrieved from human epidermis regenerated from cultured keratinocytes years after grafting (). We have recently shown that a defined number of genetically corrected stem cells regenerate a normal epidermis in patients with genetic skin adhesion disorders (). The paraclone is generated by a TA cell, whereas the meroclone has an intermediate clonal capacity and is a reservoir of TA cells (; ). The p63 gene produces full-length (TAp63) and N-terminally truncated (ΔNp63) transcripts initiated by different promoters. Each transcript is alternatively spliced to encode three different p63 isoforms, designated α, β, and γ (). The p63 gene products are essential for the morphogenesis and the regenerative proliferation of stratified epithelia (; ). In particular, ΔNp63α sustains the proliferative potential of basal epidermal keratinocytes (; ; ; ). In the human corneal epithelium, high levels of ΔNp63α identify limbal stem cells both in vivo and in vitro, whereas ΔNp63β and ΔNp63γ correlate with corneal regeneration and differentiation (; ). In mammary gland epithelial cells, the CCAAT enhancer binding protein δ (C/EBPδ) transcription factor regulates cell cycle by inducing a G/Garrest. This effect is specific for epithelial cells and for the G/G phase, as C/EBPδ expression does not increase in other types of G/G-arrested cells or in mammary cells arrested at other stages of the cell cycle (; ). C/EBPδ is a member of a highly conserved family of leucine zipper transcription factors expressed in a variety of tissues and cell types and involved in the control of cellular proliferation and differentiation, metabolism, and inflammation (; ). At least six members of the family have been isolated and characterized (C/EBPα–C/EBPζ), with further diversity produced by the generation of different polypeptides by differential use of translational initiation sites, and extensive protein–protein interactions within the family and with other types of transcription factors (; ). In this paper, we show that C/EBPδ and ΔNp63α are coexpressed by human limbal stem cells in vivo and in vitro and that the expression of C/EBPδ is restricted to a subset of mitotically quiescent ΔNp63α/Bmi1 cells. Forced expression of a constitutive C/EBPδ or of a tamoxifen-inducible estrogen receptor (ER)–C/EBPδ fusion protein in human primary limbal keratinocytes shows that C/EBPδ is instrumental in regulating self-renewal and cell cycle length of limbal stem cells. Experiments were performed on four uninjured and five wounded corneas, referred to as resting and activated cornea, respectively (). We have previously shown that ΔNp63α is expressed by 10% of resting limbal basal cells endowed with stem cell properties and that activated ΔNp63α limbal cells contain ΔNp63β and ΔNp63γ, proliferate, and migrate to the central cornea to restore a wounded epithelium (). Immunofluorescence analysis on resting limbal sections revealed that C/EBPδ and ΔNp63α were coexpressed in the same patches of basal cells (, left). Both transcription factors were undetectable in suprabasal cell layers () and in the entire corneal epithelium (not depicted). Limbal cell nuclei were stained with DAPI to estimate the proportion of C/EBPδ/ΔNp63α cells in the basal layer. 1 mm of resting limbal epithelium contained a mean of 15 C/EBPδ/ΔNp63α cells, equivalent to ∼10% of the basal layer. Upon corneal wounding and limbal activation, ΔNp63α appeared in many basal and some suprabasal limbal cells, whereas C/EBPδ remained confined to ∼10% of the basal layer (, middle). Of note, C/EBPδ limbal cells invariably coexpressed ΔNp63α (). In activated limbus, patches of C/EBPδ/ΔNp63α basal cells flanked by C/EBPδ/ΔNp63α cells were commonly observed (, right), whereas neither resting nor activated central corneal epithelium expressed C/EBPδ (not depicted). C/EBPδ/ΔNp63α resting limbal cells did not express Ki-67, a proliferation-associated nuclear antigen present throughout the cell cycle but absent in G/G-arrested cells (not depicted). In activated limbus, proliferating Ki-67 limbal cells expressed ΔNp63α, but not C/EBPδ, whereas C/EBPδ cells contained ΔNp63α but not Ki-67 (). Thus, C/EBPδ and ΔNp63α are coexpressed by quiescent limbal basal cells, whereas ΔNp63α, but not C/EBPδ, is expressed in proliferating limbal cells. The cyclin/Cdk inhibitors p27 and p57 negatively regulate G progression. Nuclear levels of p27 are high in quiescent cells (). Mitogenic and/or oncogenic signals activate different kinases that phosphorylate p27 on serine and tyrosine residues, promoting its export from the nucleus and cytoplasmic proteolysis, thereby leading to cell proliferation (; ; ; ). Of note, p57, which inhibits cyclin D–Cdk4/6 complexes (), is highly expressed in mouse epidermal stem, but not TA, cells (). Immunofluorescence analysis on limbal sections revealed that C/EBPδ, p27, and p57 were coexpressed in the nucleus of the same patches of basal cells (, arrowheads). Such cells also expressed ΔNp63α (, arrowheads). C/EBPδ cells were flanked by C/EBPδ/ΔNp63α cells containing cytoplasmic, but not nuclear, p27 and p57 (, brackets). Finally, p27 and p57 were never detected in a fully activated limbus or in the corneal epithelium (not depicted). These data are consistent with the notion that p27 and p57 are localized in the nucleus of quiescent cells, appear in the cytoplasm at the G–S transition, and are not expressed by actively proliferating cells (), and confirm that C/EBPδ is expressed only by quiescent limbal basal cells. Immunofluorescence analysis showed that ΔNp63α is abundantly and uniformly expressed in holoclones (), is expressed in a subset of meroclone cells, and is not expressed in paraclones (). Western analysis showed that clonal evolution, i.e., the transition from holoclones to paraclones, is accompained by a progressive disappearance of ΔNp63α and a relative enrichment in ΔNp63β and ΔNp63γ (). Strikingly, C/EBPδ expression was detected exclusively in holoclones () and confined to a subpopulation of ΔNp63α cells (). C/EBPδ/ΔNp63α cells were not proliferating, as shown by the mutually exclusive expression of C/EBPδ and Ki-67 (). Of note, although Ki-67 and C/EBPδ were never expressed in the same cell (), large areas of the colony were formed by nonproliferating yet C/EBPδ-negative cells (, dots), suggesting that the expression of C/EBPδ was not merely related to the proliferative status of the limbal cell. C/EBPα and -β are the most commonly expressed and thoroughly studied isoforms of the C/EBP family (). In particular, C/EBPα and -β are known to positively regulate the program of squamous differentiation in the epidermis (; ). Accordingly, we found that C/EBPα and -β were contained in the suprabasal layers of both human limbal and corneal epithelium (unpublished data). Of note, however, although C/EBPβ was expressed in all limbal clonal types (), we could not detect C/EBPα in cultured limbal colonies. Gene profiling experiments have led to the identification of genes that are commonly expressed in adult stem cells. Among these genes, Bmi1, a member of the polycomb group of transcription factors, plays a crucial role in the renewal of hematopoietic and neural stem cells (; , ; ) and is expressed in clonogenic, multipotent, and self-renewing murine hair follicle stem cells (). Immunofluorescence performed on resting limbal sections revealed that C/EBPδ and Bmi1 were coexpressed by the same limbal basal cells (). In particular, the basal layer of palisades of Vogt, where limbal stem cells are thought to be concentrated, was formed by C/EBPδ/Bmi1 cells (). Accordingly, Western blot analysis showed that Bmi1 was expressed in holoclones but not in meroclones and paraclones (). Collectively, these data indicate that C/EBPδ, ΔNp63α, and Bmi1 colocalize in limbal stem cells of the resting corneal epithelium in vivo and in limbal holoclone-forming cells in vitro and that expression of C/EBPδ is restricted to a subset of ΔNp63α cells that are mitotically quiescent both in vivo and in vitro. Primary limbal cultures were infected with a lentiviral vector expressing either an epitope-tagged human C/EBPδ or a control protein (a truncated form of the p75 low-affinity NGF receptor [ΔNGFr]) under the control of a constitutive phosphoglycerokinase (PGK) promoter. Both vectors expressed GFP under the control of an internal ribosomal entry site (IRES) element (, RRL-δ-G and RRL-N-G). Transduction efficiency on clonogenic cells was ∼90%, as calculated by GFP expression. After 2 d of cultivation, the size of untransduced colonies increased nearly threefold (, red circles), whereas the size of C/EBPδ/GFP colonies increased only slightly (, yellow circles). Control cells reached confluency 5 d after plating (). In contrast, a 5-d culture of C/EBPδ/GFP cells showed well-defined colonies composed of small, tightly packed cells (). Replicative senescence and differentiation of keratinocytes are associated with increased levels of p16 and involucrin, which indicate irreversible exit from the cell cycle and onset of terminal differentiation, respectively (). C/EBPδ-transduced cells contained threefold less p16 and involucrin than ΔNGFr-transduced cells (). C/EBPδ-transduced cells contained four- and threefold more p27 and p57 than control cells, respectively (). A cell cycle profile revealed that ∼55, 35, and 10% of the control cells were in the G, S, and G–M phases, respectively (). In sharp contrast, most of the C/EBPδ-transduced cells were in the G phase of the cell cycle (). The amount of apoptotic cells was negligible in both C/EBPδ-transduced and control cells (). Finally, C/EBPδ-dependent growth inhibition was associated with neither increase of p21 or pRb expression () nor activation of the p53 checkpoint pathway (not depicted). These data indicate that the growth inhibitory effect of C/EBPδ was not due to replicative senescence, terminal differentiation, or apoptosis. To investigate whether the growth inhibitory effect of C/EBPδ was reversible, we transduced primary limbal cells with a lentiviral vector expressing an N-terminal fusion between C/EBPδ and a modified, 4-hydroxytamoxifen (4OHT)–inducible ligand binding domain of the human ER (; , RRL-ERδ-G). In mock-transduced (RRL-ER-G) cells, C/EBPδ was found predominantly in the nucleus (, ER, middle). In the absence of 4OHT, the ER-C/EBPδ chimeric protein was found in the cytoplasm of transduced limbal cells, and nuclear translocation was observed within 12 h from the addition of 1 μM 4OHT to the culture medium (). In contrast, C/EBPβ was present only in the nucleus, irrespective of the presence of 4OHT (). Members of the C/EBP family are known to form homo- and heterodimers. In the absence of 4OHT, ER-C/EBPδ sequestered also the endogeneous C/EBPδ in the cytoplasm, as indicated by the absence of nuclear immunofluorescent staining (, −4OHT), the absence of endogeneous C/EBPδ in nuclear extracts, and the presence of C/EBPδ in the corresponding cytoplasmic extracts (, middle, −4OHT) of RRL-ERδ-G–transduced cells. Colonies of cells transduced with the control vector showed a progressive and linear increase in their size, irrespective of the presence of 4OHT (). In sharp contrast, the growth of ER-C/EBPδ/GFP colonies was strictly dependent on the localization of the ER-C/EBPδ chimera (): addition of 4OHT at day 1 considerably slowed the growth of transduced colonies; removal of 4OHT at day 4 was promptly followed by a linear increase of the size of GFP colonies; and readdition of 4OHT at day 6 again induced a growth arrest. Of note, untransduced, ΔNGFr- and ER-transduced primary limbal cells duplicated every 17–19 h, whereas C/EBPδ-transduced cells showed a doubling time of 41 h (, green). These data show that C/EBPδ lengthened the limbal cell cycle by forcing cells into the G phase without altering their capacity for multiplication. Semiquantitative RT-PCR was performed on control and C/EBPδ-transduced cells using p27- and p57-specific primers. As shown in , we observed a 5–10-fold increase of both p27 and p57 transcripts in RRL-δ-G (C/EBPδ)–transduced limbal cells and RRL-ERδ-G–transduced cells treated with 4OHT (+4OHT), as compared with RRL-N-G (ΔNGFr)–, RRL-ER-G (ER)–, and RRL-ERδ- G–transduced cells not treated with 4OHT (−4OHT). To prove the role of p27 and p57 in mediating the effect of C/EBPδ on keratinocyte cell cycle, C/EBPδ-transduced limbal cells were transfected with siRNAs specifically targeted to the p27 and p57 mRNAs. Transfection efficiency was 84 ± 2 and 77 ± 3%, respectively (). Western blot analysis showed that siRNA-p27 and siRNA-p57 caused a strong decrease of the expression of p27 and p57, but not of C/EBPδ and a control protein (). Untransduced and C/EBPδ-transduced cells showed a doubling time of 18 and 41 h, respectively (). C/EBPδ-transduced cells transfected with either siRNA-p27 or siRNA-p57 showed a doubling time of 23.5 and 23 h, respectively. Of note, C/EBPδ-transduced cells transfected with both siRNAs simultaneously, showed a doubling time of 18.5 h, a value undistinguishable from that of untransduced control cells (). These data show that p27 and p57 mediate C/EBPδ-induced mitotic quiescence. To provide evidence for a direct contribution of C/EBPδ in regulating the expression of ΔNp63α, involucrin, p27, p57, and p16, we analyzed recruitment of C/EBPδ to these loci by a chromatin immunoprecipitation (ChIP) assay on cultured limbal keratinocytes three and five passages after transduction with either RRL-δ-G or the control, RRL-N-G vector. Protein–DNA complexes were immunoprecipitated with antibodies specific for C/EBPδ or the Flag epitope and with control IgGs. Immunoprecipitated chromatin DNA was analyzed by PCR with primers specific for different regions of the p63, involucrin, p27, p57, and p16 loci (, red arrowheads), containing evolutionarily conserved and/or putative C/EBP binding elements. In C/EBPδ-transduced cells, vector-derived (Flag-tagged) C/EBPδ was found associated to the p63 locus in intron 3 (at position +147873 to +148041) and in an evolutionarily conserved, keratinocyte-specific enhancer in intron 5 (+202579 to +202761; ). Primers designed to amplify other sequences from the p63 locus detected the correct fragment only in the input samples (). Binding of Flag-tagged C/EBPδ was also observed to a region upstream of the involucrin promoter (−421 to −119) containing a C/EBP responsive element previously characterized in keratinocytes (; ) and upstream of the p27 (−227 to +14), p57 (−622 to −398), and p16 (−1020 to −871) loci (). Binding to all these sites was observed specifically in C/EBPδ-transduced cells and was more pronounced at the fifth than at the third passage (). The signals obtained with the anti-Flag antibody were always weaker than those obtained with the anti-C/EBPδ antibody, probably reflecting a lower immunoprecipitation efficiency. In control, ΔNFGr-transduced cells, a weak but specific signal was observed at the p63, involucrin, and p27 loci in chromatin immunoprecipitated with the anti- C/EBPδ but not the anti-Flag antibody. Binding was observed at the third but not at the fifth passage (), most likely as a result of the presence of endogenous C/EBPδ activity in a subset of early passage cells, which is lost in later passages. Chromatin from the same cells was also immunoprecipitated with antibodies specific for all isoforms or only the α isoforms of p63. Binding of ΔNp63α was observed in the intron 5 enhancer of the p63 locus () in both C/EBPδ-transduced and control cells. Binding was more pronounced in C/EBPδ than in control cells, reflecting either an increased recruitment of ΔNp63α to the enhancer or simply the increased proportion of cells expressing ΔNp63α in these cultures. Interestingly, ΔNp63α and C/EBPδ appear to bind the same regions in the p63 and p27 loci (). These results suggest that the p63, involucrin, p27, p57, and p16 loci might be direct targets of C/EBPδ activity, in some cases in combination with ΔNp63α. Serially cultivated, untransduced, or ΔNFGr-transduced limbal cells showed a progressive decrease of their clonogenic capacity () and ceased to proliferate after 60–75 d (or 9–11 passages) in culture (). Replicative senescence occurs because of clonal evolution, as indicated by the progressive increase of aborted, paraclone-type colonies () and by the replacement of ΔNp63α with ΔNp63β and ΔNp63γ expression ( and ). In sharp contrast, both clonogenic ability () and proliferative capacity () of C/EBPδ-transduced cells were maintained indefinitely. This effect was due to the capacity of enforced C/EBPδ expression to promote self-renewal and halt clonal evolution in holoclones, as indicated by the following evidence: serially cultivated C/EBPδ-transduced cells showed no increase in the number of paraclones () or replacement of ΔNp63α with ΔNp63β and ΔNp63γ (); statistical analysis of cell size (), a major marker of clonogenic stem cells (), showed that C/EBPδ-transduced cells were nearly 10-fold smaller than control cells (325.93 vs. 3,035.25 μm); clonal analysis revealed that the percentage of holoclone-forming cells decreased and eventually set to zero in serially cultivated control cells but remained constant in C/EBPδ-transduced cells (10–15% of inoculated cells); and ER- C/EBPδ was able to fully sequester also endogeneous C/EBPδ in the cytoplasm of limbal cells in the absence of 4OHT (). Such cells ceased to express ΔNp63α (not depicted) and underwent replicative senescence in only two passages as compared with 9–11 passages of control untransduced cells (). To investigate whether C/EBPδ was able to rescue TA cells from their terminal fate, we transduced different single cell–derived clones. As expected, holoclone, meroclone, and paraclone type clones displayed a progressive decrease in clonogenicity and ΔNp63α content (). Forced expression of C/EBPδ was able to sustain the self-renewal of holoclones and meroclones and, hence, of clones still containing ΔNp63α cells, but not that of ΔNp63α paraclones (). All cells in transduced holoclones and meroclones expressed ΔNp63α (not depicted), further suggesting that C/EBPδ is able to foster self-renewal only of ΔNp63α cells and to maintain expression of ΔNp63α in such cells. These data prompted us to investigate whether forced expression of ΔNp63α was sufficient to sustain limbal cell self-renewal. Primary limbal cultures and single cell–derived clones were infected with a lentiviral vector expressing the ΔNp63α isoform (, RRL-ΔNα-G). ΔNp63α-transduced holoclones underwent regular clonal evolution and ceased to proliferate after 11 passages, a value identical to control untransduced cells ( and ). ΔNp63α was therefore unable to sustain limbal stem cell self-renewal both in primary cultures (unpublished data) and in clones. Finally, simultaneous infection with lentiviral vectors expressing C/EBPδ and ΔNp63α was unable to rescue clonogenic ability and self-renewal in paraclones, suggesting that loss of self-renewal is an irreversible process, at least in limbal keratinocytes. Exceptional progress has been made in understanding the molecular mechanisms regulating keratinocyte stem cells. The role of transcription factors, such as p63, tcf3, CCAAT displacement protein, and GATA-3, and of adhesion and signaling molecules, such as integrins, Wnt/β-catenin, c-Myc, Notch, hedgehog, Sgk3, and bone morphogenic proteins, in controlling hair follicle and epidermal development and stem cell fate has been highlighted (; ; ; ). Molecular phenotyping of some of the keratinocyte stem cell niches helped explain how stem cells interact with the microenvironment to maintain their properties (; ). Little is known, however, on the regulation of perhaps the most important property of epithelial stem cells, that is, their capacity to self-renew. It has been shown that the Rho guanosine triphosphatase Rac1 sustains murine epidermal stem cell renewal and human epidermal stem cell clonogenicity by negatively regulating MYC (). However, differences exist between different lining epithelia and among animal species. For instance, Rac1 stimulates differentiation and not self-renewal in the intestinal epithelium (), whereas the CD34 antigen identifies murine but not human hair follicle stem cells (). We took advantage of the availability of human corneas to carry out genetic manipulation experiments on primary, clonogenic limbal stem cells and show that C/EBPδ plays a key role in regulating their cell cycle and self-renewal properties. Our findings are graphically summarized in . According to this model, a defined number of mitotically quiescent limbal stem cells coexpress Bmi1, ΔNp63α, and C/EBPδ under normal homeostasis. Coexpression of Bmi1, ΔNp63α, and C/EBPδ therefore identifies limbal holoclones and is part of the genetic program maintaining stem cell identity. Bmi1 fosters self-renewal of haematopoetic and neural stem cells through regulation of the p16 and p19 pathways (; , ; ; ) and might play a similar role also in limbal stem cells. ΔNp63α sustains the proliferative potential of stem cells in several stratified epithelia, including the cornea (; ; ; ; ; ). We show here that C/EBPδ regulates mitotic quiescence of limbal keratinocytes by forcing cells in the G/G phase of the cell cycle. Even under culture conditions specifically designed to promote keratinocyte proliferation, forced C/EBPδ expression greatly increases the cell cycle length through activation of the cell cycle inhibitors p27 and p57. The growth inhibitory effect of C/EBPδ is not due to replicative senescence or terminal differentiation, as confirmed by the down-regulation of p16 and involucrin. Perhaps more important, C/EBPδ promotes the self-renewal of ΔNp63α limbal stem cells, as suggested by the block of clonal evolution and the indefinite maintenance of the number of holoclones during serial cultivation of C/EBPδ-transduced limbal keratinocytes. Stem cells are capable of shifting from a homeostatic state of relative quiescence to rapid proliferation under specific conditions (activation). In the ocular surface, this shift occurs upon central corneal wounding (; ). This explains the apparently opposing actions of C/EBPδ and ΔNp63α. On one hand, C/EBPδ induces mitotic quiescence (through a positive regulation of p27 and p57) and self-renewal of limbal stem cells; on the other, it preserves their proliferative potential (essential for stem cell–dependent tissue regeneration) through a positive regulation of ΔNp63α. In this way, when some limbal stem cells are released from C/EBPδ-dependent mitotic constraints, as in a corneal damage, they can unchain their remarkable p63-dependent proliferative capacity, multiply, and migrate to repair a corneal wound. This process is, however, irreversible and leads to limbal stem cell terminal differentiation (). Our data therefore strengthen the notion that proliferation and self-renewal capabilities are two related, albeit distinct, processes. At least in human limbal stem cells, proliferation potential relies on the expression of ΔNp63α, whereas self-renewal requires also C/EBPδ. Similarly, Bmi1 is essential for the self-renewal of neural stem cells but does not influence the proliferative capacity of their committed progeny (). The notion that ΔNp63α induces the expression of growth factor receptors and adhesion molecules regulating survival and motility of epithelial cells () is consistent with our proposed model. Our data establish an interesting parallel with the hematopoietic system, where quiescence and self-renewal of stem cells have been recently shown to be linked and regulated by p27, p57, and Mad1 (; ; ). Indeed, loss of p27 allows relatively quiescent hematopoietic stem cells to rapidly enter the cell cycle to restore haematopoiesis (). Finally, we show that C/EBPδ is directly associated in vivo, alone or in combination with ΔNp63α, to chromatin-surrounding promoters or regulatory elements of the p63, p27, p57, and p16 loci, suggesting a direct role of this transcription factor in determining the genetic program of self-renewing stem cells. The role of C/EBPδ described here is intriguing. Indeed, C/EBPs have been mainly related to cellular differentiation. C/EBPα, -β, and -δ are instrumental in regulating adipogenesis, whereas C/EBPα, -ɛ, and -β orchestrate myeloid differentiation into mature neutrophils, atypical neutrophils, and macrophages (; ), and C/EBPδ regulates learning and long-term memory in the central nervous system (; ). The importance of the C/EBP family in cellular differentiation also extends to other cell types, including hepatocytes, ovarian luteal cells, intestinal epithelial cells, and epidermal keratinocytes. For instance, it has been shown that C/EBPα and -β induces cell cycle exit in normal keratinocytes and positively regulates the program of squamous differentiation in the epidermis (; ). However, C/EBPβ promotes keratinocyte proliferation and skin tumor formation in the presence of oncogenic Ras or in response to carcinogens (; ) and fosters hepatocyte proliferation during liver regeneration after partial hepatectomy (). Mammary epithelial cells from C/EBPβ-deficient mice have a proliferation defect that leads to impaired ductal morphogenesis and a failure to lactate (; ), and ectopic C/EBPβ expression in human mammary epithelial cells induces hyperproliferation and a partially transformed phenotype (). Finally, C/EBPδ induces late differentiation events in epidermal keratinocytes () and is indeed detected in the subrabasal layers of the human epidermis (unpublished data). Therefore, the biological effects of C/EBPs appear to be highly species and cell context specific, suggesting that role that C/EBPδ exerts in the human corneal epithelium might not necessarily be observed in other squamous epthelia. The mechanisms controlling C/EBPδ expression and function in the limbus, as well as the downstream mediators of C/EBPδ activity in controlling stem cell quiescence and self-renewal, remain to be determined. The expression of the C/EBPs has been found to change markedly during several physiological and pathophysiological conditions through the action of extracellular signals. C/EBPs are subject to extensive species- and tissue-specific posttranscriptional regulation and phosphorylation-mediated changes in DNA binding activity and nuclear localization (). Furthermore, the different C/EBP proteins are able to form heterodimers in all intrafamilial combinations and to associate with other factors (). A combination of biochemical, cellular, and genetic experiments is necessary to acquire a more comprehensive description of upstream regulators and downstream targets of C/EBPδ and to elucidate the networks of protein interactions and regulatory pathways that control its activity in human limbal stem cells. Swiss mouse 3T3-J2 cells (a gift from H. Green, Harvard Medical School, Boston, MA) were grown in DME supplemented with 10% calf serum. Keratinocytes were cultivated on a feeder layer of lethally irradiated 3T3-J2 cells, and colony forming efficiency (CFE) assays and calculation of the number of cell generations and population doublings were performed as described previously (; ). Clonal analysis was performed from subconfluent primary cultures as described previously (, ). In brief, single cells were inoculated onto multiwell plates containing a feeder layer of 3T3 cells. Clones were identified after 7 d of culture under an inverted microscope and transferred to replicate dishes. One dish (1/4 of the clone) was fixed 9–12 d later and stained with rhodamine B for clonal type classification (; ). The second dish was used for further experiments and analyses. In selected experiments, 100 limbal cells were plated in 100-cm dishes and cultured for 1 wk. Colonies were then examined under a microscope (Axiovert 200 M; Carl Zeiss MicroImaging, Inc.): large round colonies with smooth and regular borders and formed entirely by small cells with scarce cytoplasm were classified as holoclones () and were subjected to immunofluorescence. For cell cycle analysis, subconfluent keratinocyte cultures were trypsinized and fixed in 70% ethanol at 4°C. Samples (10 cells) were rehydrated in PBS/1% FCS at room temperature for 10 min and stained with 20 μg/ml propidium iodine for 30 min at 4°C. Flow cytometry was performed using a LSR II FACScan (Becton Dickinson). The following antibodies were used: rabbit anti-C/EBPδ, anti-Ras, anti-Rb, and p57 purified IgG (Santa Cruz Biotechnology, Inc.); 4A4 pan-p63 mAb (BD Biosciences); p16, p21, and p27 mAbs (Exalpha Biologicals, Inc.); involucrin and Ki67 mAbs (Novocastra); Bmi1 mAb (Upstate Biotechnology); rabbit anti-p63α unconjugated and FITC-conjugated purified IgG raised against a synthetic peptide (NH-DFNFDMDARRNKQQRIKEEGE-COOH) comprising the C terminus post-SAM domain of p63α (Primm; ). Secondary rhodamine- or FITC-labeled antibodies were obtained from Santa Cruz Biotechnology, Inc. For immunofluorescence analysis, keratinocyte colonies were fixed (3% paraformaldehyde/2% sucrose in PBS, pH 7.6), permeabilized (0.5% Triton X-100 in PBS), and coated with 0.5% BSA/PBS for 1 h at RT. Paraformaldehyde-fixed corneal samples were embedded in OCT, frozen, and sectioned. Immunofluorescence was performed on fixed colonies and 5–7-μm corneal sections as described previously (). Confocal analyses were done with a confocal analyzer (LSM510-META; Carl Zeiss MicroImaging, Inc.). Multitrack analysis was used for image acquisition. For immunoblots, mass or clonal cultures were extracted on ice with RIPA buffer (0.15 mM NaCl/0.05 mM Tris/HCl, pH 7.5/1% Triton X-100/1% sodium deoxycholate/0.1% SDS). Nuclear and cytoplasmic protein extraction was performed using the NE-PER Nuclear and Cytoplasmic Extraction kit (Pierce Chemical Co.) following conditions supplied by the manufacturer. Equal amounts of samples were electrophoresed on 7.5% SDS-polyacrylamide gels and transferred to polyvinylidene difluoride filters (Immobilon-P; Millipore). Immunoreactions were performed as described previously () using antibodies at a 1:500 dilution. Immobilon bound antibodies were detected by chemiluminescence with ECL (GE Healthcare). Total RNA was extracted from keratinocyte cultures, purified with RNase Micro kit (QIAGEN), and quantified by spectrophotometry. RT-PCR was performed using the One Step RT-PCR kit (QIAGEN). cDNAs were synthesized from 0.5–2 μg of total RNA, and PCR reactions were performed using 20, 24, 28, 32, 36, and 40 cycles. β-Actin was used for normalization. Ethidium bromide–stained agarose gels were visualized with an Image Station 440 CF (Kodak). Quantification was performed using 1D 3.5 software (Kodak). Primer sequences for p63 isoforms and annealing temperatures were as described previously (). The following primers and annealing temperatures were used for p27, p57, and β-actin RT-PCR: p27, 5′-AGTGTCTAACGGGAGCCCTA-3′ and 3′-GTCCATTCCATGAAGTCAGC-5′ (annealing temperature 60°C, 829 bp); p57, 5′-CACGATGGAGCGTCTTGTC-3′ and 3′-CTTCTCAGGCGCTGATCTCT-5′ (annealing temperature 60°C, 699 bp); and β-actin, 5′-GAGCGCAAGTACTCCGTGT-3′ and 3′-ACGAAGGCTCATCATTCAAA-5′ (annealing temperature 58°C, 548 bp). The human C/EBPδ cDNA was cloned by RT-PCR from total RNA extracted from the THP1 cell line using specifically designed primers containing EcoR1 recognition sequences. To generate a N-terminal Flag epitope–tagged C/EBPδ sequence, the EcoR1 C/EBPδ cDNA fragment was subcloned into the pFlagCMV-2 plasmid (Sigma-Aldrich), after inserting an EcoRV restriction site at position 913 by PCR (for all primers, see Table S1, available at ). To generate a Flag-tagged ER-C/EBPδ fusion sequence, a MfeI–EcoRI PCR–amplified fragment containing the modified ligand binding domain of the human ER () was fused to the C terminus of the Flag epitope before inserting the C/EBPδ cDNA. The Flag-C/EBPδ, Flag-ER- C/EBPδ, and Flag-ER cassettes were extracted as EcoRV fragments and cloned downstream the human PGK promoter and upstream an IRES-EGFP cassette into the blunted SmaI–BamHI sites of the pRRL.ppt.PGK.IRES.GFP.WPRE lentiviral vector (), to obtain the RRL-δ-G, RRL-ERδ-G, and RRL-ER-G vectors. The control RRL-N-G vector was generated by cloning an NcoI–EcoRV fragment encoding a truncated form of the low-affinity, p75 NGFr (ΔNGFr; ) into the same vector backbone. The RRL-ΔNα-G vector was obtained by inserting the cDNA of ΔNp63α isoform () into the same vector backbone. Lentiviral stocks pseudotyped with the vesicular stomatitis G protein (VSV-G) were prepared by transient cotransfection of 293T cells using a three-plasmid system (the transfer vector and the helper plasmids pCMVΔR8.74, encoding Gag, Pol, Tat, and Rev, and pMD.G, encoding VSV-G), as previously described (). Viral titers were determined by transduction of HeLa cells with serial dilution of the vector stocks and ranged from 10 to 10 TU/ml. Transduction efficiency was evaluated by scoring GFP and/or ΔNGFr transgene expression by flow cytometry. Subconfluent primary or clonal limbal cultures were trypsinized. 2 × 10 cells were resuspended in 1 ml of culture medium containing 8 mg/ml polybrene and transduced with lentiviral vector stocks at a MOI of 25, overnight at 37°C. Gene transfer efficiency was assessed 4 d after transduction by scoring GFP cells by confocal fluorescence microscopy (LSM510-META; Carl Zeiss MicroImaging, Inc.). In selected experiments (), 1 μM 4OHT was added every 12 h for 3 d to RRL-ERδ-G–transduced cultures. 4OHT was then removed from the culture medium for 2 d and readded until control cultures reached confluency. These experiments were performed using the siRNA-p27 and siRNA-p57 RNAi Human/Mouse Starter kit (QIAGEN). The siRNA duplexes were designed using the HiPerformance Design Algorithm licensed from Novartis AG, integrated with a stringent in-house homology analysis tool. Double-stranded RNAs were synthesized by QIAGEN. Primary cultured human limbal epithelial cells previously transduced with a lentiviral vector carrying the C/EBPδ cDNA (RRL-δ-G) were plated into 24-well plates at 4 × 10 cells/cm. 48 h later, cells were transfected using fluorescently labeled p27 (Alexa Fluor) and p57 (Cy3) siRNAs at a final concentration of 67 nM (siRNA to Hiperfect reagent ratio 1:6) and incubated under normal growth conditions (37°C and 5% CO). Transfection efficiency was determined 24 h after siRNA addition through laser-scanning confocal microscope (LSM510-META) analysis. Nonsilencing siRNAs (Alexa Fluor) were used as negative controls at the same conditions of transfection (67 nM; 1:6 ratio) and cell density. ChIP assays were performed essentially as described previously (). Chromatin was prepared from 10 limbal keratinocytes at the third and the fifth passage after transduction with either the RRL-δ-G or the RRL-N-G vector. Nuclear extracts were sonicated to obtain DNA fragments ranging from 400 to 800 bp in length. The equivalent of ∼5 × 10 cells was immunoprecipitated with rabbit anti-CEBPδ (Santa Cruz Biotechnology, Inc.), mouse anti-Flag (Sigma-Aldrich), mouse anti-p63 (4A4 pan63; BD Biosciences), and rabbit anti-p63α antibodies. Immunoprecipitations with mouse and rabbit IgGs (BD Biosciences) were included as controls. Immunoprecipitated DNA was analyzed by PCR with primers spanning regions containing known or putative CEBPδ and p53/p63 binding motifs within the genomic loci of p63 (from position +70720 to +70951, +147873 to +148041, +151191 to +151408, +202579 to +202761, and +234830 to +235057 from the transcription start site), involucrin (−421 to −119), p27 (−227 to +14), p57 (−622 to −398), and p16 (−1020 to −871). Specific primers are listed in Table S1. Table S1 gives the primers used in this study. Online supplemental material is available at .

Myelination allows the rapid propagation of action potentials along the axon and is an important prerequisite for the normal function of the nervous system. Schwann cells (SCs) are the myelin-forming cells of the peripheral nervous system (PNS). During development, neural crest–derived SC precursors populate outgrowing axon bundles, where they proliferate and differentiate into immature SCs (for review see ). Possibly as a result of increasing cell density (), these cells extend cytoplasmic processes into axon bundles to segregate and establish one-to-one relationships with individual axons in a process referred to as radial sorting (; ). During this process, large caliber fibers are sorted and later become myelinated, whereas the remaining small caliber axons are engaged by nonmyelinating SCs (). To sense their environment and regulate their intrinsic developmental program accordingly, SCs need to interpret instructive cues originating within the extracellular environment, among which growth factors and proteins of the ECM are essential components. This process is likely to involve GTPases of the Rho subfamily. Cdc42, rac1, and rhoA are the best-characterized family members. They are well known for their roles in regulating signaling pathways linking extracellular stimuli to the assembly and organization of the actin cytoskeleton (). In addition, Rho GTPases control microtubule dynamics, cell polarity, membrane trafficking, and gene transcription (; ). Rho GTPases are expressed by SCs (). In vitro experiments using dominant-negative and constitutively active forms of rac1 and cdc42 suggested that these small GTPases together with FAK may serve to link growth factor activation with SC motility (). More recently, , again using an in vitro approach, suggested that cdc42 and rac1 regulate the JNK signaling cascade to enhance migration of SCs in response to TrkC tyrosine kinase receptor activation by endogenous neurotrophin-3. Further work by the same investigators proposed that TrkC activation by neurotrophin-3 stimulates SC migration through two parallel signaling pathways involving cdc42 or rac1 (). Precise control of rac1 activity also regulates SC morphology and promotes normal axonal interaction (). Cells derived from SC tumors (schwannomas) in which rac1 activity is deregulated showed a disorganized cytoskeleton () and failed to interact with axons (). Reestablishing normal rac1 activity levels in these cells restored SC spindle morphology and their capacity to interact with axons (). In this study, we examined the role of cdc42 and rac1 signaling in peripheral nerve development using tissue-specific conditional gene ablation specifically in the SC lineage. Ablation of either or impairs radial sorting of axons. However, these proteins play different roles during SC development. Our results indicate that although cdc42 is required for SC proliferation, rac1 is necessary for correct SC process extension and stabilization. Furthermore, we show that cdc42 activation can be induced by neuregulin-1 (NRG1) and is critically involved in SC proliferation. Rac1 is activated by β1 integrin signaling and regulates SC process extension and stabilization. To study the role of cdc42 and rac1 signaling in SCs, we conditionally ablated or by expressing Cre recombinase (Cre) under the control of the desert hedgehog () gene regulatory sequences (). In this setting, Cre is active in SC precursors from embryonic day (E) 12 (). To identify the recombined cells, we bred also the conditional reporter allele from the reporter mouse () into control, mutant, and mutant mice. Recombination of the conditional or alleles () led to a strong reduction in cdc42 and rac1 protein in lysates obtained from the sciatic nerves of postnatal day (P) 1 and mutant mice, respectively (). The low residual cdc42 and rac1 protein levels detected in the mutant lysates are likely due to the presence of endoneurial and perineurial fibroblasts as well as some unrecombined SCs. In acute SC cultures obtained from P1 sciatic nerves, –mediated recombination of the conditional allele was found in 87% of the SCs (400 of 460 counted cells) as assessed by X-gal staining (). At P14, the sciatic nerves of and mutant mice were thinner and more transparent (). Both and mutants displayed a progressive hind limb accentuated paresis. At around P30, mutants developed hind limb paralysis. Although less affected, mutants died of unknown causes at ∼P40. Thus, for both mutants, our analysis was performed at P24 or earlier. During postnatal development, SCs segregate and myelinate individual large caliber axons from the axon bundles formed during embryogenesis (). We compared control and mutant sciatic nerves at different developmental stages. At E17.5, SCs were present between bundles of tightly apposed axons in the sciatic nerves of control, mutant, and mutant mice (). Although there were still a few unsorted bundles of axons in control mice at P5 (), the majority of large caliber axons were already engaged in a one-to-one relationship with SCs. By P24, practically all large caliber axons were sorted and myelinated (). In contrast, in the mutant nerves, axon sorting and myelination was impaired and bundles of axons containing large caliber axons persisted. This feature was more pronounced in mutant nerves () than in mutant nerves, where a considerable number of larger caliber axons progressively became sorted and myelinated over time (). Very few axons were myelinated in mutant nerves (). As the mutant SCs also carried the conditional allele (), we could follow the fate of recombined cells by detecting the activity of the β-galactosidase by Bluogal staining. To this end, we performed Bluogal EM analysis on P24 and mutant nerves. Bluogal precipitates were detected in promyelinating and myelinating SCs of both P24 and mutant nerves (Fig. S1 A, available at ). This suggests the concomitant recombination of the or conditional alleles in these cells. The Bluogal EM data also indicated that the improvement of the radial sorting phenotype observed in P24 mutant nerves is not the result of cellular compensation by unrecombined, rac1-positive SCs. This conclusion was further supported by Western blot analysis of protein lysates obtained from sciatic nerves of P24 mutant and control mice. The levels of rac1 in mutant lysates were strongly reduced in relation to control lysates (Fig. S1 B), comparable to the one previously detected in P1 mutant nerves (). To understand why axon sorting was affected in mutant nerves, we performed EM analysis of P5 and P24 control and mutant nerves. At P5, this analysis revealed SCs at different stages of differentiation, including immature SCs associated with axon bundles, promyelinating SCs in a one-to-one relationship with large caliber axons, and myelinating SCs that had started to form a compact myelin sheath. In control () and mutant nerves (), immature SCs extended long processes that fully enveloped axon bundles, a normal feature of SCs at this stage of differentiation. This, however, was not the case in mutant nerves (), where immature SC processes were shorter and often failed to envelope axon bundles. At the promyelination stage, SC-axon profiles in control () and mutant nerves () were surrounded by an apposed basal lamina (BL; , arrows). In contrast, mutant SC-axon profiles contained abnormal cytoplasmic protrusions that extended in various directions (, arrowheads). These protrusions were surrounded by an apposed BL covering their entire surface (, black arrows) and often displayed empty loops of redundant BL (, white arrows). At the sites where they emerged, these loops were continuous with the apposed BL. Retracting protrusions may have detached from their original BL and, while leaving behind these loops of empty BL, produced a new apposed BL layer. At P24, radial sorting of large caliber fibers in control nerves was virtually completed and mature SCs formed a compact multilayered sheath of myelin around single large caliber axons (). Small caliber axons were engaged by nonmyelinating SCs (, arrowhead). SC-axon profiles at the myelinating stage were surrounded by an apposed BL (, arrows). In P24 mutant nerves, hypomyelination was pronounced, resembling the situation at P5. Axon bundles in mutant nerves were still not enveloped by immature SC processes (). SC-axon profiles at the myelinating stage () did not show any of the protrusions seen at the promyelinating stage (, arrowheads), but several loops of empty BL in the shape of the protrusions remained attached to the apposed BL (, white and black arrows, respectively). In mutant nerves, some naked axons were surrounded by several layers of BL, suggesting that SC processes repeatedly engaged and retracted from these axons, leaving behind empty BL layers (, arrows). P24 mutant nerves contained many axon bundles (), which as in P5 nerves were fully enveloped by immature SC processes (, arrowheads). mutant nerves contained also a small number of SC-axon profiles at the myelinating stage (), which were surrounded by an apposed BL (, arrows). Interestingly, and in contrast to what was described by in SC-specific conditional β1 integrin mutant nerves, and more recently by in SC-specific conditional FAK mutant nerves, we have not observed perineurial cells in contact with axon bundles in or in mutant nerves. We conclude that in mutant nerves, SC process extension and stabilization is severely affected in a timely and spatially controlled manner. It is likely that such deficits are the reason why radial sorting is delayed. In contrast, and despite their severe radial sorting defects, SC process extension in mutant nerves appears to be normal, suggesting that the mechanisms associated with radial sorting failure in the two mutants are intrinsically different. To further confirm that the differentiation of and mutant SCs was severely delayed or even blocked, we used immunofluorescence analysis to examine the expression of the transcription factor Oct6 on transverse sections of control and mutant P5 and P24 sciatic nerves and Western blotting to assess the expression levels of Krox20 in mutant and control sciatic nerves at P2 and P14. The expression of Oct6 peaks at the promyelinating and early myelinating stages and is down-regulated at later stages of myelination (). At P5, control, mutant, and mutant nerves contained similar numbers of Oct6-positive SCs, whereas at P24, Oct6-positive SCs were only present in Rac1 or Cdc42 mutant nerves but not in control nerves (). Krox20 expression is required for down-regulation of Oct6 and for the activation of myelin genes, such as peripheral myelin protein 22, myelin protein zero, and myelin basic protein (). At P2, Krox20 expression was detected by Western blotting in control but only at very low levels in and mutant nerves (). Later, at P14, Krox20 was expressed at similar levels in control and mutant nerves while it was reduced in mutant nerves (). These findings corroborate our morphological data showing developmental SC abnormalities (). Insufficient numbers of SCs, as a result of reduced proliferation or survival, could potentially explain the persistence of axon bundles, in particular in mutants. Thus, we compared the total numbers of DAPI-stained nuclei present within transverse sections of P5 and P24 sciatic nerves of mutant, mutant, and control mice (). Although at both ages the numbers of DAPI-positive nuclei in mutant nerves were significantly lower compared with control nerves (), no differences were found between mutant and control nerves (). Transverse sections obtained from P5 and P24 mutant nerves were smaller in size than those obtained from control nerves (). This reduction in size is likely the result of the severe hypomyelination in mutant nerves. To analyze potential proliferation defects directly, we compared the percentage of proliferating Ki67-positive cells in control, mutant, and mutant nerves at E17.5, P0, P5, and P14. As expected, proliferation was normal in the absence of rac1 () but was significantly diminished in mutant nerves at E18 and P0. At P5, there was no difference between control and mutant cell proliferation, and at P14, cell proliferation in mutant nerves was even higher than in control nerves (). Despite this continuing late proliferation, most likely due to delayed SC differentiation, the total number of cells present in P24 mutant nerves was still less than half of those present in control nerves (). To investigate whether the reduction in cell numbers was influenced by increased cell death, we compared the percentage of TUNEL-positive cells in P0 and P5 in control, mutant, and mutant nerves. No significant differences were found in all experimental settings examined (). We conclude that the loss of cdc42 leads specifically to a reduction in SC proliferation but does not affect cell survival. Similar to and mutant nerves, conditional ablation of the β during SC development also results in the impairment of radial sorting (). This and the knowledge that in a variety of different cell types, integrin signaling can regulate the activity of Rho GTPases (), including rac1 and cdc42, prompted us to investigate whether β1 integrin regulates the activity of these two proteins in SCs. Therefore, we performed Rho GTPase activity assays (see Material and methods) in sciatic nerve lysates obtained from P5 ′,′′ β mutant and control mice (). In these mice, the Cre recombinase is under the control of the CNP gene regulatory sequences and is expressed in SCs from E12 on (; unpublished data). Western blot analysis of protein lysates obtained from β mutant nerves confirmed that the expression of β1 integrin was strongly reduced (). The following Rho GTPase activity assays showed that in β mutant nerve lysates, rac1 activity was significantly reduced compared with controls (). In contrast, the cdc42 activity showed a slight tendency to increase but, interestingly, the total expression level of cdc42 was reduced (). We conclude that in SCs, rac1 activity is dependent on β1 integrin signaling, whereas cdc42 activity is not, suggesting that activation of cdc42 and rac1 relies, at least partially, on different pathways. This hypothesis is consistent with our findings that cdc42 is required for SC proliferation, a process that is not dependent on β1 integrin () or rac1 signaling (this work). SC proliferation is, however, modulated by growth factors such as NRG1 (). As Rho GTPase activity can also be regulated by growth factors, we hypothesized that in SCs, exposure to NRG1 may preferentially activate cdc42 rather than rac1. We tested this hypothesis by examining the levels of activated cdc42 and rac1 in cultured rat SCs after 15, 30, and 60 min of exposure to NRG1. As expected, we found that exposure to NRG1 significantly increased the activation of cdc42 in relation to control non–NRG1-treated SCs (). Cdc42 activation was elevated after 15 min of NRG1 stimulation, peaked after 30 min, and returned to almost control levels after 60 min. In contrast, the increase in rac1 activation was modest and only significantly higher than control levels after 30 min of NRG1 stimulation (). The increase in cdc42 and rac1 activity was accompanied by a reduction in the expression levels of both molecules (). Our morphological data demonstrated that process extension and stabilization in mutant SCs is defective. SC process extension is thought to require ECM–integrin interactions and mutations in genes, and the conditional ablation of β in SCs appears to affect the extension and targeting of SC processes, resulting in impaired radial sorting (; ). Therefore, we hypothesized that when plated on laminin-2, a component of SC BL and a ligand for SC laminin integrin receptors containing the β1 integrin subunit, rac1 but not cdc42 mutant SCs would behave as β1 integrin mutant cells and produce shorter processes. Similar assays and rationale were applied before to study the role of ECM–integrin signaling and downstream molecules in the regulation of oligodendrocyte process extension (; ; ; ). Thus, we plated primary SCs obtained from sciatic nerves of neonate control, mutant, and mutant mice on laminin-2 substrates and grew the cells in defined media for 16 h, sufficient time for SCs to extend several thin processes (). SCs prepared from sciatic nerves of β mutant mice were treated in the same way. Visualization of the actin and microtubule cytoskeleton by immunofluorescence allowed us to measure and compare the length of the SC processes formed. β and mutant SCs produced significantly shorter processes than their control counterparts, whereas the length of the processes produced by mutant SCs was not significantly different from controls (). Collectively, these results indicate that rac1 activation is required for normal SC process extension and support the view that the specific radial sorting defects in the nerves of mutant mice are a consequence of deficient SC process extension. #text Conditional mutant () and mutant () mice have been described. mice (hereafter called mutant mice) and (hereafter called control mice). To follow the fate of recombined cells by detection of β-galactosidase expression, we also bred the conditional allele from the reporter mouse strain () into control and mutant mice (provided by P. Soriano, Fred Hutchinson Cancer Research Center, Seattle, WA). Genotypes were determined by PCR on genomic DNA. Mutant and control mice were produced using the same breeding strategy as for mice. The generation of β mutant and control mice has been described (; ). mice were also recently used successfully to target the excision of a conditional FAK allele to SCs (). Mice were deeply anesthetized with Pentobarbital (150 mg/kg, i.p.; Nembutal; Abbott Laboratories) and then perfused with 0.1 M phospate buffer, pH 7.4, followed by buffer containing 3% glutaraldehyde and 4% PFA. Fixed tissues were postfixed in 2% osmium tetroxide, dehydrated through a graded acetone series, and embedded in Spurrs resin (Electron Microscopy Sciences). Semithin sections were stained with toluidine blue for analysis at the light microscope, and ultrathin sections were contrasted with 3% uranyl acetate and 1% lead citrate before observation in a transmission electron microscope (H-600; Hitachi) at 75 kV. After fixation with 4% PFA, sciatic nerve sections were blocked for 1 h with 10% goat serum and 0.1% Triton X-100 in PBS. Incubation with primary antibodies was performed overnight at 4°C. Rat monoclonal antibodies against myelin basic protein (1:50; Serotec) and Ki67 (1:50; DakoCytomation), mouse monoclonal antibodies against NF160 (1:150; Sigma-Aldrich), and polyclonal antibodies against Oct6 () were used. On the following day, tissue sections were washed in PBS and incubated with the appropriate secondary antibodies for 1 h at room temperature. Secondary antibodies conjugated to fluorescein or rhodamine were obtained from Jackson Immunoresearch Laboratories. The sections were mounted in Citifluor (Citifluor Ltd) containing DAPI to stain the nuclei. Mouse SC cultures were fixed in 4% PFA in MP buffer (65 mM Pipes, 25 mM Hepes, 10 mM EGTA, and 3 mM MgCl, pH 6.9) for 10 min at room temperature. The cells were permeabilized with 0.2% Triton X-100 in MP buffer for 5 min at room temperature. Primary monoclonal antibody against α-tubulin (1:100; Sigma-Aldrich) and Alexa Fluor 488 phalloidin (1:100; Invitrogen) were incubated in 1 mg/ml BSA in PBS overnight at 4°C, followed by incubation with Cy3-conjugated anti-mouse (1:200; Jackson ImmunoResearch Laboratories) at room temperature. Cells and tissue sections were visualized in a fluorescence microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) equipped with 10×/NA 0.30, 20×/NA 0.50, and 40×/NA 0.75 Plan-Neofluar objectives (Carl Zeiss MicroImaging, Inc.). Images were acquired using a charge-coupled device camera (MRm Axiocam; Carl Zeiss MicroImaging, Inc.) connected to a PC running the Axiovision 4 acquisition software (Carl Zeiss MicroImaging, Inc.). Images were further processed (levels adjustment) using Photoshop 7.0 software (Adobe). was acquired as a black-and-white image, and artificial color was added using Photoshop. Primary mouse SC cultures were obtained from P0–P2 sciatic nerves. In brief, nerves were digested in enzyme buffer (0.7 mg/ml collagenase type I and 0.25% Trypsin [Invitrogen] in HBSS [Invitrogen]). After trituration, cells were grown on poly--lysine/laminin–coated dishes in minimal medium plus 0.5% FCS. Minimal medium is DME/F12 (Invitrogen), containing 100 μg/ml human apo-transferrin, 60 ng/ml progesterone, 5 mg/ml insulin, 16 μg/ml putrescine, 400 ng/ml -thyroxin, 160 ng/ml selenium, 10 ng/ml triiodothyronine, and 300 mg/ml BSA (Fluka). Supplements were obtained from Sigma-Aldrich unless stated otherwise. Rat SCs were isolated from Wistar rats and grown in SC medium (DME; Invitrogen), containing 10% FCS, 50 μg/ml gentamicin (Sigma-Aldrich), 100 μg/ml crude GGF (BioReba Biotechnology, Inc.), and 2 μM forskolin (Sigma-Aldrich) as described previously (). Before the Rho GTPase activity assays and exposure to NRG1, freshly plated rat SCs were starved for 6 h in DME without forskolin, growth factors, and serum. Sciatic nerve tissue was homogenized with a chilled mortar and pestle in lysis buffer (0.1% SDS, 10 mM TrisHCl, 150 mM NaCl, 50 mM NaF, 1 mM NaVO, 1 mM EDTA, 0.5% sodium-deoxycholate, and protease inhibitor cocktail [Sigma-Aldrich]). Extracts were processed using standard SDS-PAGE and Western blotting procedures. The following antibodies were used: Krox20 (1:100; Babco), β-actin (1:1,000; Sigma-Aldrich), α-tubulin (1:1,000; Sigma-Aldrich), GAPDH (1:20,000; HyTest), rac1 (1:1,000; BD Biosciences), and cdc42 (1:1,000; Santa Cruz Biotechnology, Inc.). Secondary antibodies were obtained from Pierce Chemical Co. and Santa Cruz Biotechnology, Inc. Immunoreactive proteins were detected using ECL (GE Healthcare). Densitometry and quantification of the relative levels were performed on scanned images of Western blots using Quantity One software (Bio-Rad Laboratories, Inc.). The GST-PAK-CD construct was provided by J. Collard (The Netherlands Cancer Institute, Amsterdam, Holland). Rac1 or cdc42 activity was measured essentially as described previously (). In brief, for each independent experiment, sciatic nerves from three P5 mutant mice, three P5 control mice, or rat SCs were pooled, homogenized in FISH buffer (10% glycerol, 50 mM Tris-HCl, pH 7.4, 100 mM NaCl, 1% NP-40, 2 mM MgCl, and protease inhibitor cocktail), and centrifuged for 5 min at 21,000 at 4°C. Aliquots were taken from the supernatant to determine the total protein amounts. The supernatant was incubated with bacterially produced GST-PAK-CD fusion protein, bound to glutathione-coupled Sepharose beads (GE Healthcare) at 4°C for 30 min. The beads and proteins bound to the fusion protein were washed three times in an excess of FISH buffer, eluted in Laemmli sample buffer, and analyzed for bound GTP-cdc42 or GTP-rac1 by Western blotting. Data show the mean ± SD. Statistical significance was determined using test. Significance was set at *, P < 0.05; **, P < 0.01; ***, P < 0.001. represents the number of independent experiments, which was three unless stated otherwise. Fig. S1 shows Bluogal precipitates in premyelinating and myelinating SCs of P24 and mutant nerves and that in P24 mutant nerves the levels of rac1 protein were reduced. Online supplemental material is available at .

Myelin optimizes conduction of nerve impulses and is formed by multiple membrane wraps of glial cells (for review see ). In the peripheral nervous system, Schwann cells (SCs) are the glial cells that associate with axons to form myelinated and nonmyelinated fibers. SCs originate from neural crest and migrate longitudinally along bundles of growing axons. Then, SCs send processes radially within bundles, to segregate out axons destined to be myelinated (axonal sorting), obtain a one-to-one relationship with them, and wrap them with sheets of inwardly spiraling membrane (for review see ). Axonal sorting is regulated by signals from axons and from the extracellular matrix. Mice lacking laminins have a block in axonal sorting, resulting in bundles of “naked” axons (; ). SCs express laminin receptors, including α6β1, α6β4 integrin, and dystroglycan (for review see ). Among these, β1 integrins play a pivotal role in radial sorting, as its absence in SCs causes a defect similar to that of laminin mutants (). The signaling cascades activated by β1 integrins to promote sorting are poorly known. Small Rho GTPases, such as Rac, Cdc42, and RhoA, are signaling molecules that cycle between an active (GTP bound) and an inactive (GDP bound) state. They influence cell shape by regulating actin upon activation from various stimuli, including integrin engagement (; ; , ). Rac1 promotes actin polymerization to produce lamellipodia and ruffles. Low levels of active Rac1 produce axial (at the two extremities of the main cell axis) lamellae, favoring directional cell migration, whereas higher levels of Rac1 produce radial (around the whole cell perimeter) lamellae (). Cdc42 regulates the formation of filopodia, whereas RhoA leads to the assembly of stress fibers and of focal adhesions (). Small GTPases are active in the peripheral nervous system (). Studies in and in vitro proposed a role for Rac1 in glial migration and oligodendrocyte differentiation and for RhoA in internodal and nodal organization (; ; ; ). Little is known on the role of small GTPases in mammalian peripheral nerves and during sorting and myelination. Here, we first determine that β1 integrin–null SCs display normal cytoskeletal dynamics during migration and elongation on axons, but cannot produce radial lamellipodia, similar to cells with reduced levels of active Rac1. Second, we show that the levels of active Rac1 are reduced in nerves lacking β1 integrin in SCs and that Rac1 is not targeted to the membrane of β1 integrin–null SCs. Third, we generate a mouse with specific Rac1 deletion in SCs and show that Rac1 regulates radial lamellipodia, segregation of axons, and myelination. Finally, we show that exogenous activation of Rac1 in β1 integrin–null nerves ameliorates the sorting defects. We conclude that SCs longitudinally oriented and elongated on axons produce radial processes that segregate and then myelinate axons upon β1 integrin–mediated activation of Rac1. Perturbation of β1 integrin in SCs impairs interactions with axons during radial sorting and precludes myelination (; ). To test whether this is due to the inability of β1 integrin–null SCs to reorganize the cytoskeleton during axonal interactions, we used organotypic cultures of DRG from wild-type (wt) or β1 integrin/P0-Cre–conditional null mice. These mice lose β1 integrin expression in SCs after embryonic day (E) 17.5 (). We first characterized mutant DRG cultures explanted at E14.5. Mutant DRG reached a maximum of 60% β1-negative SCs after 4 wk in culture (), in contrast to postnatal nerves, where the extent of P0-Cre–mediated recombination was nearly complete (). β1-null SCs migrated at distances from DRG (). The number of SC nuclei was not reduced in mutant cultures (), although mutant SCs have a slight increase in the fraction of apoptotic nuclei (not depicted). Because of the presence of both β1-negative and -positive SCs, the absence of β1 integrin was always confirmed by direct or retrospective staining (Fig. S1, available at ). To validate β1 integrin–null cultures, we asked if they could myelinate after ascorbic acid supplement. The number of myelin internodes stained with anti–myelin basic protein (MBP) antibody was significantly lower in mutant cultures (P < 0.05; ). Defective myelin formation was not due to a reduced number of SCs or axons, as demonstrated by DAPI (cigar-shaped SC nuclei) and neurofilament staining (). Thus, cultures from β1 integrin–null mice showed defective myelination and could be used as a model to analyze the cytoskeleton of mutant SCs as they interact with axons. To evaluate whether mutant SCs maintained proper orientation and association with axons, we analyzed these parameters at 3–5 wk in culture. Cells were classified based on their relationship with axons (), and their proportions were quantified. β1 integrin–null SCs presented a slight decrease in the percentage of cells associated with axons (). Despite the statistically significant difference, this decrease is probably not sufficient to explain the dramatic impairment of mutant SCs to form myelin. To evaluate the dynamic interactions between mutant SCs and axons, we used time-lapse microscopy. Previous experiments showed that SC tips remodel resembling growth cones, with filopodia and lamellipodia (). We asked whether mutant SCs manifest this behavior near axons in the presence or absence of ascorbic acid. In both conditions, mutant and wt SCs behaved similarly, as they organized dynamic processes at one extremity that interacted with and moved around the axon (Fig. S2 and Videos 3 and 4, available at ). The time frame during which these growth cone–like processes were dynamically reorganized was similar in both genotypes (Fig. S2 and Videos 1–4). To visualize the ability of mutant SCs to extend radial processes, we asked if they were able to spread on laminin, vitronectin, or poly--lysine (PLL). β1 integrin–null SCs spread less when plated on laminin (). In contrast, the spreading of mutant and wt cells plated on vitronectin and PLL was similar. This is consistent with the fact that β1 integrin is contained in all laminin receptors expressed by cultured SCs except dystroglycan (; ; ; ), whereas β1 integrin–null and wt SCs express the vitronectin receptor αvβ3 (unpublished data) and PLL promotes non–receptor-specific cell adhesion. Thus, deficient spreading by β1-null SCs results from loss of β1 integrin binding to laminin. Next, we counted the number of lamellipodia (, asterisks). The number of lamellipodia was reduced in β1 integrin–null SCs plated on laminin, but not PLL or vitronectin (). It was proposed that directed migration and elongation in other cell types depend on the ability to produce axial lamellipodia (). Axial lamellae are present within of a 20° angle of the extremities of the cell-long axis (, arrows), whereas peripheral or radial lamellae are outside this zone (, arrowheads; ). Notably, although the number of axial lamellae was similar between wt and mutant SCs plated on laminin, the number of peripheral/radial lamellae was significantly reduced (P < 0.0001; ). The length of extension of both axial and radial lamellipodia was reduced in mutant SCs (). Both talin and phalloidin (not depicted) were used to quantify lamellipodia. The relative sparing of axial lamellipodia can explain why mutant SCs migrate and elongate normally on axons (as this process uses axial extension at the edge of the cell-long axis; ) and suggests that mutant SCs are instead unable to insert radial processes within axons. It was recently shown that the formation of radial/peripheral versus axial lamellae in other cell types depends on levels of activation of the small GTPase Rac1 (). To ask directly if the levels of Rac1 affect the formation of radial and peripheral lamellae in SCs, we inhibited Rac1 using the specific NSC23766 inhibitor (). NSC23766 inhibited the levels of active Rac in rat SCs in a dose-dependent way (). In the absence of inhibitor, SCs spreading on laminins had both radial and axial lamellipodia (). At 100 μM of inhibitor (intermediate levels of active Rac1), the number of radial lamellipodia was reduced (P < 0.0001; = 100), whereas the number of axial lamellipodia was unchanged (P = 0.6; = 100; ). Decreasing further the activity of Rac1 reduced the number of both radial and axial lamellipodia (). Few axial, and nearly no radial, lamellipodia were produced on vitronectin or PLL. Thus, laminins and high levels of Rac1 are required for the formation of radial lamellipodia in SCs. Having shown the effect of NSC23766 on SC lamellipodia, we asked if Rac1 inhibition also affected myelination by SCs in DRG explants. Low doses of inhibitor were sufficient to almost completely inhibit myelination, as shown by the absence of MBP-positive internodes (), without affecting axons and SC number (). Next, we asked if the activity of Rac1, or other small GTPases, is affected in β1 integrin–null SCs and promotes axonal sorting. Both axons and SCs express small Rho GTPases, myelinating SCs at high levels (). We studied expression and activation of Rac1, Cdc42, and RhoA in developing sciatic nerves, where SCs constitute the predominant cell type. The total protein levels of the three GTPases decreased in the adult (), whereas the active, GTP-bound fraction of the three small GTPases in sciatic nerve lysates did not change substantially during postnatal development (). To ask if mutant SCs had diminished Rac1 signaling, we compared the amount of active, GTP-bound Rac1 in sciatic nerves deriving from mutant and control mice. Relative Rac1 activity of β1 integrin–null sciatic nerves was decreased to ∼60% of wt nerves (). One way by which integrins regulate Rac1 is by promoting the translocation of GTP-bound Rac1 to the cell membrane, allowing interaction with the effector p21-activated kinase (PAK) 1 (). To test if GTP-bound Rac was able to target to the SC membrane in the absence of β1 integrin, we added purified GST–PAK binding domain (PBD) to wt or mutant SCs spreading on laminin and visualized its localization using anti-GST antibodies. We favored internalization of PBD-GST using saponin. Addition of GST alone resulted in no immunostaining (unpublished data). As expected, PBD was enriched at the surface of lamellipodia in wt cells, suggesting that active Rac1 is recruited at the leading edge of SC spreading on laminin (). In contrast, PBD-GST was not enriched in processes of mutant SCs (). The fraction of processes with membrane enrichment was significantly lower in β1 integrin cells (121 out of 208 processes in 47 wt cells and 33 out of 114 processes in 31 null cells; P < 0.005 by χ analysis). This suggests that Rac1 translocation is inhibited in the absence of β1 integrin. PBD can bind also Cdc42. To evaluate whether PBD could be interacting with Cdc42 in lamellipodia, we stained wt and mutant SCs for Cdc42. Cdc42 is excluded from lamellipodia of SCs spreading on laminin (), and its localization is similar between β1 integrin–negative and –positive cells (). Thus, Cdc42 does not colocalize with PAK during lamellipodia formation on laminin. We conclude that SCs recruit active Rac1 to the membrane of lamellipodia when spreading on laminin and that this translocation is impaired in the absence of β1 integrin. We next assayed the activity Cdc42 and RhoA in mutant nerves. The levels of active Cdc42 were not substantially different (), consistent with the normal localization of Cdc42 in β1-null SCs. This suggests that Cdc42 is activated independently of β1 integrins in SCs. RhoA activity was markedly reduced in β1 integrin–null and control sciatic nerves (). Because RhoA in other cell types controls the formation of focal adhesion and stress fibers (), we visualized them in wt and mutant SCs using phalloidin and antibodies against talin and paxillin. We could not detect any qualitative difference in focal adhesion and stress fiber formation between wt and mutant cells (Fig. S3, available at ). Because the absence of β1 integrin in SCs causes a reduction in lamellipodia formation and Rac1 activity and Rac1 promotes radial lamellipodia formation in SCs, we hypothesized a role for Rac1 in the formation of radial processes by SCs during axonal sorting (). To test this, we generated mice with an SC-specific deletion of the Rac1 gene, by crossing mice bearing a conditional loxP-flanked allele of Rac1 (Rac1F) with P0-Cre transgenic mice (; ). Rac1 conditional null mice were viable but developed severe clenching and occasional paralysis of posterior limbs (). We analyzed recombination of the Rac1 locus by PCR on genomic DNA from different tissues of Rac1//P0-Cre mice. As shown for other floxed loci (; ; ), P0-Cre mediated specific and robust inactivation of Rac1 () in sciatic nerves, but not liver and kidney. Recombination was also observed in brain, as previously reported for some floxed loci using this P0-Cre (). By Western blot analysis, the levels of Rac1 protein were reduced in mutant nerves (). We could not demonstrate the complete absence of Rac1 protein or activity in SCs by Western blot, pull-down assay, or immunohistochemistry of nerves ( and not depicted), likely because Rac1 is expressed in axons and because all available anti-Rac1 antibodies also recognized the highly related Rac3 (). To ask whether Rac1 affected SC number, we counted the number of nuclei in P45 Rac1 mutant nerves. The number of nuclei per sciatic nerve segment was not reduced (). Instead, the number of nuclei was increased, as well as their density, as the mutant nerve was smaller. Cdc42 and RhoA activities were not substantially different between mutant and wt mice (). To ask whether SCs lacking Rac1 are able to generate lamellipodia, we isolated them from postnatal day (P) 5 mutant sciatic nerves and plated them on different substrates. As observed before, wt cells generated more radial lamellipodia when plated on laminin than PLL or vitronectin. Similar to SCs lacking β1 integrin, SCs lacking Rac1 could not generate lamellipodia, but neither radial nor axial (). This is consistent with the fact that levels of active Rac are much lower than those of β1 integrin–null cells () and agrees with the dose-dependent model proposed by and our data using 300 μM NSC23766. Lack of lamellipodia was present also on vitronectin and PLL. Use of the Rac1 inhibitor prevented myelination in wt DRG (). To eliminate a possible neuronal intrinsic effect, we asked whether mutant DRG, lacking Rac1, specifically in SCs, could myelinate normally. The number of MBP-positive internodes was reduced significantly in mutant DRG (P < 0.01; ). We next analyzed sciatic nerve morphology by semithin and ultrathin sections during the postnatal development of Rac1//P0-Cre mice and controls. In wt animals, P5 nerves contain few bundles of unsorted axons, axons in a one-to-one relationship with SCs and many thinly myelinated fibers (). In contrast, in P5 Rac1 mice, many axons were unsorted in bundles, some axons were in a one-to-one relationship, and no myelinated fibers were present (). Ultrastructural analysis at P5 confirmed that many axons were unsorted in more numerous and larger bundles in mutant than in wt littermates (, asterisks). In P10 wt nerves, occasional unsorted bundles were present, and myelin becomes thicker (). In contrast, several unsorted bundles of axons remained in Rac1-null nerves (, arrows), which contain axons >3 μm (, double asterisks). Many SCs began to segregate large caliber axons in a one-to-one relationship (promyelinating SCs), and several thinly myelinated fibers appeared (). Thus, developing Rac1-null nerves present a delay in axonal sorting and myelination. Many promyelinating SCs in Rac-null nerves showed bizarre and disoriented cytoplasmic processes directed away from axons (), strikingly similar to those observed in β1 integrin–null nerves (). By P28, unsorted axons were not detectable in mutant nerves, and there was a progressive increase in the number of myelinated fibers (). However, many large caliber axons were devoid of myelin or had thin myelin sheaths. Thus, Rac1 is required for timely radial sorting of axons by SCs, but the sorting defect can be overcome, possibly by compensation or redundancy with other GTPases such as Rac3, Cdc42, or RhoG. After reaching the promyelinating stage, albeit with delay, many Rac1-null SCs are still unable to myelinate. The inability to produce radial sheets of membranes as a result of deficient Rac1 activation seems to be one of the important defects of β1 integrin–null SCs. Thus, forced Rac1 activation should improve the radial sorting phenotype of β1 integrin–null nerves. To test this, we injected an adenovirus expressing constitutively active (CA) Rac1 in the endoneurium of P10 β1 integrin–null sciatic nerves. This procedure creates minimal damage to the developing nerve () and allows for distant spreading of the injected virus. Contralateral nerves were injected with saline solution or adenoviruses expressing reporter genes. Remarkably, nerves treated with active Rac1 showed an improvement of sorting, as compared with control-injected nerves (, compare A–D with E–H). The effect was most evident near the site of injection (−1 and +1), where many naked bundles of axons were reduced in size, and they contained fibers in a one-to-one relationship with promyelinating SCs or thin myelin sheaths (; compare enlarged magnifications in L with K). Statistical analysis performed on optic and electron microscopic sections of the −1 level confirmed that radial sorting was significantly improved, as the size of unsorted bundles decreased (P < 0.01) and the proportion of promyelinating SCs (fibers in a one-to-one relationship) increased (P < 0.02). In contrast, the proportion of myelinated fibers did not change, suggesting that the levels of active Rac1 achieved were sufficient for radial sorting but not for myelination. The active Rac construct is fused to an HA tag, which allowed us to confirm its expression in injected nerves by Western blotting (). To ask whether the abnormalities in Rac-null nerves could be due to basal lamina defects, we compared the basal laminae of adult mutant and wt nerves by immunohistochemistry and electron microscopy. Staining with antibody to α2, α4, and γ1 laminin chain and to collagen IV did not reveal abnormalities in mutant nerves. A subset of fibers in Rac-null sciatic nerves showed higher α4 laminin immunoreactivity, probably because of their immaturity (Fig. S5 F, available at ). Basal laminae appeared normal ultrastructurally (Fig. S5, I and J). Similarly, we asked if Rac activation in rescued β1 integrin–null nerves had an effect on the basal lamina. β1 integrin–null nerves contain both fibers with normal and redundant basal lamina (Fig. S5, K and M; ). Similarly, CA-Rac rescued nerves contained both normal (Fig. S5, L′) and redundant (Fig. S5, N′) basal laminae around nerve fibers. Thus, basal laminae abnormalities do not seem to account for the delay in myelination seen in Rac-null nerves or for the rescue by Rac1 seen in β1 integrin–null nerve. #text Experiments with animals followed protocols approved by the Institutional Animal Care and Use Committee of San Raffaele Scientific Institute. mP0TOTA (P0-Cre), floxed β1 integrin (β1; obtained from U. Mueller, The Scripps Research Institute, La Jolla, CA), and floxed Rac1 (Rac1) were described previously (, ; ). P0-Cre, β1 and β1 mice were N15-N18 generations congenic in C57BL/6, whereas Rac1 P0-Cre mice and littermate controls were F2. Crosses to generate conditional null mice were either Y//P0-Cre × Y or Y//P0-Cre × Y. Genotypes were identified by PCR analysis of tail genomic DNA (). Primers for recombination of the floxed Rac1 allele were as follows: 5′-ATTTTGTGCCAAGGACAGTGACAAGCT-3′, 5′-GAAGGAGAAGAAGCTGACTCCCATC-3′, and 5′-CAGCCACAGGCAATGACAGATGTTC-3′, which amplify a 175- (recombined), 300- (wt), or 334- (floxed allele) nucleotide band. Given the multiplex nature of the PCR amplification, the band intensity cannot be used to judge the amount of recombination. Mouse E14.5 DRG were dissected as described previously () and maintained in N/D Sato medium (modified from ). Myelination was induced with 50 μg/ml ascorbic acid (Sigma-Aldrich). Time-lapse analysis was performed on glass-bottomed 35-mm plates (Mat-Tek), using an inverted microscope (Axiovert; Carl Zeiss Microimaging, Inc.). Images were captured every 60 s, for a maximum of 2 h, by a charge-coupled device camera (Hamamatsu) and analyzed with ImageJ. Recorded cells were identified by drawing on the plate with a diamond tip and retrospectively stained for β1 integrin. SCs were purified from DRG cultures by the cold jet technique (), plated in N/D Sato, and stained after 24 h. Mouse SCs were isolated from P5 sciatic nerves stripped of perineurium using 4% collagenase and 2.5% trypsin, plated on coverslip in defined medium () with 10 ng/ml β-neuregulin-1 (R&D Systems) and 2 μm forskolin (Calbiochem), and analyzed after 24 h. Rat SCs were plated on laminin in DME and 10% fetal calf serum for 2 h, treated with NSC23766 (; provided by Drug Synthesis and Chemistry Branch, Developmental Therapeutic Program, Division of Cancer Treatment and Diagnosis, Bethesda, MD) in defined medium for 30 or 120 min, and stained with phalloidin or assayed for Rac1 activity. For myelination, NSC23766 was added to mouse DRG during ascorbic acid treatment. Immunohistochemistry was performed as described previously (). Immunocytochemistry was performed on glass coverslips, coated with PLL alone (Sigma-Aldrich), 0.1 mg/ml, or followed by coating with 10 μg/ml vitronectin or laminin 1 (Sigma-Aldrich). Purified mouse SCs or DRG were fixed in 4% PFA or in dodecyl-trimethylammonium chloride + 1% PFA () and permeabilized with 0.1% Triton X-100 (Sigma-Aldrich) or 100% methanol. To reveal active Rac1 localization, recombinant PAK1 PBD-GST protein was detected with an anti-GST antibody. PAK1 PBD-GST was added at a final concentration of 0.01 μg/μl. After 5–6 h, cells were fixed with 4% PFA for 15 min at RT, permeabilized with 0.2% saponin (Sigma-Aldrich)/0.2% gelatin gepulvert (Merck), and reexposed to PAK PBD-GST at 0.01 μg/μl diluted in 0.1% saponin/0.2% gelatin overnight at 4°C. Primary and secondary antibodies were diluted in 0.1% saponin/0.2% gelatin. GST alone was used as a negative control. Negative controls for β1 integrin staining were β1 integrin–deficient “TKO” embryoid cells (). The following antibodies were used: rabbit anti–Neurofilament H (Chemicon); rat anti-Neurofilament (TA-51; a gift from V. Lee, University of Pennsylvania, Philadelphia, PA); mouse anti–Neurofilament M (Roche and Chemicon); rabbit anti–β1 integrin (a gift from K. Rubin, University of Uppsala, Uppsala, Sweden); goat anti-paxillin antibody, goat anti-talin antibody, rabbit anti-Rac1, mouse anti-RhoA, and rabbit anti-Cdc42 (Santa Cruz Biotechnology, Inc.); goat anti-GST (GE Healthcare); mouse anti-Rac1 (Upstate Biotechnology and BD Biosciences); mouse anti–β-tubulin (Sigma-Aldrich); rabbit anti–S-100 (DakoCytomation); rat anti–laminin α2 (a gift from L. Sorokin, Lund University, Lund, Sweden); rabbit anti–laminin α4 (a gift from J. Miner); and peroxidase-conjugated anti-HA antibodies (Roche). Secondary antibodies were conjugated with FITC, TRITC, or Cy5 fluorochromes (Jackson ImmunoResearch Laboratories and Sothern Biotechnology Associates, Inc.). GST pull-down assays for Rho GTPases were performed for Rac and Cdc42 as described previously () using a pGEX-cRac1A plasmid (a gift from G. Bokoch, The Scripps Research Institute, La Jolla, CA), and for RhoA using Rhotekin RBD-GST plasmid (a gift from C. Laudanna, University of Verona, Verona, Italy) as described by , with few modifications. Lysates from sciatic nerves at P1, P5, and adult mice or from SCs were triturated in 20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% Triton X-100, and 5 mM MgCl with protease and phosphatase inhibitors. After centrifugation, 500 μg of lysates were incubated either with 17 μg PAK1 PBD-GST or 35 μg Rhotekin RBD-GST bound to glutathione agarose for 60 min at 4°C. Lysates activated or inactivated with GTP-γs or GDP-βs were used as positive and negative controls, respectively. Beads were washed once in lysis buffer and, together with total lysates (10 μg for Rac1 and Cdc42 and 20 μg for RhoA), heated for 5 min at 100°C in reducing sample buffer and processed for Western blotting by standard methods. For quantification, films were digitalized and analyzed using ImageQuant v1.2 for Mac software (Molecular Dynamics). Images for comparison were always on the same gel. 1.5 μl adenovirus-Rac1CA (10 virions/ml; a gift from M. Resh [Memorial Sloan-Kettering Cancer Center, New York, NY]; ), saline, or adenoviruses expressing GFP of LacZ were injected in the endoneurium of P10 sciatic nerve of anesthetized mice (seven animals total). Nerves were collected after 12 d. To quantify the rescue, the mean area of all bundles of unsorted axons present in transverse semithin section of the complete −1 levels from seven different experiments were measured using ImageJ. Promyelinating and myelinating axons were counted as a proportion of total nerve fibers in 10 random electron microscopic fields from the −1 levels in seven CA-Rac1–injected and seven control nerves. Morphological analyses of nerves were conducted as described previously (). Images were acquired using confocals (UltraView ERS spinning disk confocal microscope [PerkinElmer] equipped with a Plan Apochromat 63×/1.4 oil-immersion objective and using the UltraView acquisition software; TCS-SP5 [Leica] equipped with a Plan Apochromat 63×/1.4 oil-immersion objective and using the LCS confocal acquisition software [Leica]; or Confocal-MRC 1024 laser-scanning confocal microscope [Bio-Rad Laboratories, Inc.] equipped with a Plan Neofluar 40×/1.3 oil-immersion objective and using the Laser Sharp 2000 acquisition software) or a camera (DFC480 R2; Leica) mounted on a fluorescence microscope (DM 5000 B; Leica) equipped with N Plan 10×/0.25, HC PL Fluotar 20×/0.50, and HCX PL Fluotar 40×/0.75 objectives, using Firecam software (Leica). Videos were acquired using a camera (Orca II; Hamamatsu) mounted on a microscope (Axiovert S100 TV2; Carl Zeiss MicroImaging, Inc.) equipped with a Plan Neofluar 40×/1.3 oil-immersion objective at 37°C in Eagle's minimum essential medium with 10% FCS and 25 mM Hepes, without phenol red, with or without ascorbic acid, and using the Image Pro-Plus 4.5 acquisition software. Electron microscopy sections were visualized and photographed using a transmission electron microscope (EM900; Carl Zeiss MicroImaging, Inc.). Films were digitalized using a scanner (Arcus II; Agfa-Gevaert). Image processing and quantification was performed using Photoshop 7.0 (Adobe) or ImageJ (v1.33u). Adjustment of brightness or contrast was used in some cases but without obscuring, eliminating, or misrepresenting information. Statistical analysis was performed using Excel (Office X; Microsoft), Stat View (v5.0), and SPSS v11. Videos 1 and 2 show a wt and a β1 integrin–null SC, respectively, migrating on a DRG axon in a qualitatively similar manner. Videos 3 and 4 show a wt and a β1 integrin–null SC, respectively, interacting with axons in a similar fashion. Fig. S1 shows that the cells in Videos 2 and 4 were negative for β1 integrin. Fig. S2 shows still frames from Videos 1–4. Fig. S3 shows that stress fibers and focal adhesions appear normal in β1 integrin–null SCs. Fig. S4 shows the method used to quantify PAK PBD-GST enrichment in the cell membrane. Fig. S5 shows that basal laminae are normal in Rac1-null nerves and in CA-Rac rescued β1 integrin–null nerves. Online supplemental material is available at .

The formation of synapses requires a complex interchange of signals between presynaptic nerve terminals and postsynaptic cells. This is best illustrated at the mammalian neuromuscular junction (NMJ), where a signaling cascade mediating postsynaptic differentiation has been well characterized (; ; ; ; ). At the center of this signaling cascade are agrin, a proteoglycan derived from the terminals of presynaptic motoneurons (; ), and MuSK, a muscle-specific receptor tyrosine kinase activated by agrin (; ). Activation of MuSK leads to the assembly of the postsynaptic complex, and genetic studies in mice have shown that both agrin and MuSK are required for the formation of the NMJ (; ; ). In particular, MuSK is required for all aspects of postsynaptic differentiation, including the initial agrin and nerve-independent clustering of nicotinic acetylcholine receptors (AChRs) (; ; ), thus functioning as the major organizer of synapse formation at the NMJ. It is less clear how the activity of MuSK is regulated to ensure proper development and homeostasis of synapses. Increasing evidence suggests that protein ubiquitination plays an important role in regulating synaptic development, maintenance, and plasticity (). Protein ubiquitination is mediated by the sequential actions of three enzymes: an E1 ubiquitin activating enzyme, E2 ubiquitin conjugating enzyme, and E3 ubiquitin ligase. Of these, E3 ubiquitin ligases directly bind substrates and render substrate specificity to the ubiquitination reaction, thus acting as key modulators of the ubiquitin system. Consistent with this idea, genetic studies in and have demonstrated the importance of E3 ubiquitin ligases in synaptic development and function. Most notably, mutations of , a putative E3 ubiquitin ligase in , or its orthologue in (), lead to aberrant presynaptic development (; ; ; ). Recent studies in also suggest that E3 ubiquitin ligases play an important role in regulating the surface abundance of glutamate receptors at the postsynaptic membrane (; ). Although ample evidence supports the role of protein ubiquitination in synaptic development and plasticity in vertebrates (; ; ; ; ), the specific molecular mechanisms underlying these effects remain to be elucidated. In the present study, we report the functional characterization of PDZRN3, a protein containing both RING and PDZ domains, as a synapse-associated E3 ubiquitin ligase at the mammalian NMJ. PDZRN3 (PDZ domain containing RING finger 3) was named based on its sequence similarity to PDZRN1 and 2 (). It has also been named LNX3 and SEMCAP3 based on its sequence similarity to LNX1/LNX2 (Ligand-of-Numb protein X) and SEMCAP1/SEMCAP2 (M-SemF cytoplasmic domain-associated protein), respectively. A very recent study showed that PDZRN3 is expressed in muscle (). In culture, the expression of PDZRN3 is increased during differentiation of myoblasts to myotubes and may play a role in myoblast fusion (). We find that PDZRN3 mRNA is enriched in the synaptic region of the muscle and that PDZRN3 protein is concentrated at the NMJ. Coimmunoprecipitation shows that PDZRN3 interacts with MuSK in heterologous cells and in myotubes, and that this interaction is enhanced by agrin stimulation. Functionally, PDZRN3 promotes ubiquitination of MuSK and down-regulates cell surface levels of MuSK through its E3 ubiquitin ligase domain. Both gain- and loss-of-function studies in cultured myotubes reveal an important role for PDZRN3 in regulating agrin-induced AChR clustering. Furthermore, transgenic overexpression of PDZRN3 in vivo perturbs the growth and maturation of the NMJ. Our findings demonstrate an important role for PDZRN3 in regulating the growth and maturation of the NMJ. PDZ domain–mediated protein–protein interactions play important roles at synapses (). Many synaptic proteins at the mammalian NMJ, including MuSK, ErbB2, neuregulin-1, and Eph receptors, contain conserved PDZ binding motifs at the C termini (), suggesting that PDZ domain proteins may also play a role at the NMJ. Our previous yeast two-hybrid screen using the last 25 amino acids of erbB2 and neuregulin-1 as baits identified Lnx1 as an ErbB2-interacting, PDZ domain–containing E3 ubiquitin ligase that is exclusively expressed in perisynaptic Schwann cells at the NMJ (). We then searched databases for additional RING and PDZ domain–containing proteins that may play a role at the NMJ and identified PDZRN3 as a potential candidate. Structurally, PDZRN3 (also called SEMCAP3 and LNX3; ; ) contains a RING domain, a zinc finger domain, one (PDZRN3A) or two (PDZRN3B) PDZ domains depending on alternative splicing, and a PDZ domain binding motif (TTV) at the C terminus (). Northern blot analysis indicates that there are two species of PDZRN3 mRNA with molecular sizes of 5.5 and 4.6 kb. Both splicing variants are highly expressed in skeletal and cardiac muscle of adult mice. Lower levels of expression are also seen in the brain, spinal cord, kidney, and lung (). Examination of mRNA expression by in situ hybridization of the diaphragm shows that PDZRN3 mRNA is expressed throughout muscle fibers with a distinct enrichment at the central region of the diaphragm muscle where synapses and synaptic nuclei are localized (), suggesting that PDZRN3 mRNA is preferentially transcribed by synaptic nuclei and thus may play a role at the NMJ. Because MuSK contains a conserved PDZ binding motif at its C terminus () we tested whether PDZRN3 interacts with MuSK by coexpressing MuSK and PDZRN3A in COS-7 cells. As shown in , MuSK can be specifically coimmunoprecipitated with PDZRN3A, suggesting a direct interaction between MuSK and PDZRN3A. As a control, we found that Cbl, a well-characterized RING-type E3 ubiquitin ligase (), does not interact with MuSK under the same conditions (). To test whether PDZRN3 interacts with MuSK under physiological conditions, we immunoprecipitated MuSK from cultured myotubes and found that PDZRN3 was coimmunoprecipitated with MuSK, indicating that they form a complex in myotubes (). Moreover, agrin stimulation significantly increased the association of MuSK and PDZRN3 in cultured myotubes (), suggesting a dynamic regulation of PDZRN3 and MuSK interaction. To determine which domain(s) in the PDZRN3 protein mediates the interaction with MuSK, we made deletion constructs of PDZRN3A that lacked either the RING and zinc finger domains or the PDZ domain. When coexpressed with MuSK in COS-7 cells, PDZRN3A lacking the RING and zinc finger domains retained the ability to interact with MuSK, whereas PDZRN3A lacking the PDZ domains failed to interact with MuSK (). This suggests that the PDZ domain in PDZRN3A is required for its interaction with MuSK. However, deletion of the PDZ binding motif of MuSK only partially decreased its interaction with PDZRN3A (), suggesting that the PDZ domains of PDZRN3A may interact with internal sequences as well as the C-terminal motif on the cytoplasmic domain of MuSK. To further dissect the interaction of MuSK with PDZRN3A, we coexpressed full-length PDZRN3A with a series of deletion constructs of MuSK lacking the PDZ binding motif (last 3 amino acids). Coimmunoprecipitation data showed that, in the absence of the PDZ binding motif, a large portion of the kinase domain of MuSK is required for binding to PDZRN3A (). However, inclusion of the terminal PDZ binding motif (TTV) in deletion constructs of MuSK is sufficient to restore the binding with PDZRN3A (), indicating that the PDZ binding motif of MuSK is sufficient but not necessary for its binding to PDZRN3A in a heterologous system. To examine the localization of PDZRN3 proteins, we generated a rabbit polyclonal antibody against the N-terminal RING domain (PDZRN3-10B) and a guinea pig polyclonal antibody against an internal unique sequence (PDZRN3-3B). Both affinity-purified antibodies specifically recognized a single band from C2C12 myotube lysates which corresponded to the size of the shorter PDZRN3 isoform (PDZRN3A in ) from transfected COS-7 cells (), suggesting that PDZRN3A is the predominant form in myotubes. Consistent with the Northern blot result, PDZRN3 proteins are detectable in skeletal muscle, heart, spinal cord, and brain (). To determine the localization of PDZRN3 proteins in the neuromuscular system, we performed coimmunostaining of adult mouse muscle with anti-PDZRN3 antibodies and α-bungarotoxin (α-BTX), which specifically binds nicotinic acetylcholine receptors (AChRs). Both anti-PDZRN3 antibodies stain the NMJ brightly with a pattern that precisely matches the staining of AChRs with α-BTX (). The staining is specific as it can be blocked by preincubation of the antibodies with cognate antigens (). To determine whether PDZRN3 is localized postsynaptically, we denervated hindlimb muscles in adult mice by transection of the sciatic nerve on one side of the body, and compared the expression of PDZRN3 in denervated and non-denervated muscles using triple labeling of α-BTX, anti-synaptophysin, and anti-PDZRN3. 5 d after nerve resection, presynaptic nerve terminals had degenerated as revealed by the lack of staining for the presynaptic vesicle protein synaptophysin, whereas staining for AChRs () and MuSK (Fig. S1 [available at ]; ) remained. PDZRN3 staining persisted at the denervated NMJ and colocalized with AChR staining (), indicating that PDZRN3 is localized at the postsynaptic site of the NMJ, consistent with a potential role in regulating MuSK signaling. The developmental profile of PDZRN3 expression at the NMJ was examined by immunostaining of muscle sections from various developmental stages. PDZRN3 staining at the NMJ first appears at approximately embryonic day (E) 16.5 (), a time point when NMJs have just formed. As the NMJ grows and matures, the staining intensity and complexity of PDZRN3 increases in concert with that of AChRs (), a developmental profile common to known postsynaptic regulators at the NMJ, such as MuSK and rapsyn (; ). The presence of a classic RING domain with conserved key cysteine and histidine residues in the PDZRN3 protein suggests that PDZRN3 likely functions as a RING-type E3 ubiquitin ligase. To test this, we used an in vitro ubiquitination assay to examine whether the RING domain of PDZRN3 contains E3 ubiquitin ligase activity. The RING domain of PDZRN3 was fused to GST and expressed in . Glutathione bead-purified GST fusion proteins were tested for self-ubiquitination in the presence of added E1 and E2 enzymes. Currently, more than two dozen E2 enzymes have been identified, and they show varying degrees of specificity when interacting with E3s (). Our initial tests of various E2s showed that the RING domain of PDZRN3 exhibits ubiquitin ligase activity in the presence of E1 and the UbcH5 family of E2 enzymes (). Consistent with this finding, GST pull-down assays showed that PDZRN3 directly interacts with the UbcH5B ubiquitin conjugating enzyme (). Furthermore, mutation of either one of the two key cysteines in the RING structure completely abolished the ubiquitin ligase activity of PDZRN3 (). Together, these data demonstrate that PDZRN3 is a catalytically active RING-type E3 ubiquitin ligase. We next tested whether PDZRN3 promotes the ubiquitination of MuSK. MuSK was cotransfected with PDZRN3A, PDZRN3AΔRING (PDZRN3A lacking the RING domain), or PDZRN3AΔPDZ (lacking the PDZ domains) into cultured COS-7 cells. Transfected cells were harvested in the presence of 2% SDS and incubated at 4°C for 1 h to minimize nonspecific association of proteins with MuSK. MuSK was immunoprecipitated with anti-MuSK antibodies and analyzed by Western blot for ubiquitination using an anti-ubiquitin antibody. In the absence of PDZRN3A, MuSK from transfected COS-7 cells showed low levels of ubiquitination (). When MuSK was coexpressed with PDZRN3A, the levels of MuSK ubiquitination were significantly increased (). Deletion of the RING domain, which contains the E3 ubiquitin ligase activity, or the PDZ domain, which is required for the interaction with MuSK, completely abolished the effects of PDZRN3A on the ubiquitination of MuSK (). On the other hand, coexpression of PDZRN3A and TrkB, a functionally unrelated receptor tyrosine kinase, did not lead to increased ubiquitination of TrkB (). These data strongly suggest that MuSK is a specific substrate of PDZRN3 ubiquitin ligase activity. To determine whether endogenous MuSK in myotubes is ubiquitinated, we immunoprecipitated MuSK from agrin-treated and nontreated myotubes and probed for MuSK ubiquitination using an anti-ubiquitin antibody. In the absence of agrin stimulation, a low level of MuSK ubiquitination was observed. However, the level of ubiquitination of MuSK was significantly increased upon agrin stimulation (). Together with the results showing the enhancement of MuSK and PDZRN3 interaction upon agrin stimulation, these data suggest that the ubiquitnation of MuSK is dynamically regulated and PDZRN3 may play an important role in this regulation. To determine the functional consequences of MuSK-PDZRN3 interaction, we tested whether PDZRN3 directly regulates the level of MuSK on the cell surface. We coexpressed PDZRN3A and MuSK in COS-7 cells and measured cell surface levels of MuSK by membrane nonpermeable biotinylation of surface proteins and subsequent quantification of biotinylated MuSK by Western blot. Coexpression of PDZRN3A with MuSK significantly reduced the surface level of MuSK (). This effect was seen with both wild-type and constitutively kinase-active MuSK. Interestingly, PDZRN3A did not down-regulate the surface level of kinase-dead MuSK, which retains partial ability to interact with PDZRN3A (), suggesting that the effect of PDZRN3A on surface MuSK levels depends on the kinase activity of MuSK. This is consistent with our observation that agrin stimulation, which activates MuSK kinase activity, enhances the MuSK-PDZRN3 interaction and MuSK ubiquitination. This is also consistent with other studies showing that E3 ubiquitin ligases are often activated by the kinase activity of the receptor tyrosine kinases they regulate (; ; ). Because the RING domain contains the ubiquitin ligase activity, we examined whether the RING domain of PDZRN3 is required for the down-regulation of the surface level of MuSK using a PDZRN3A construct lacking the RING domain. As shown in , deletion of the RING domain abolishes the effect of PDZRN3 on the surface level of MuSK, suggesting that the E3 ubiquitin ligase activity of PDZRN3 plays a critical role in this regulation. To test the specificity of the effect of PDZRN3 on MuSK surface expression, we coexpressed PDZRN3A with erbB2, another receptor tyrosine kinase present at the NMJ. We found that PDZRN3A did not affect the surface level of erbB2 (Fig. S2 A, available at ). Moreover, the surface level of MuSK is not down-regulated by Cbl (Fig. S2 B), a RING-type E3 ubiquitin ligase that has been shown to down-regulate the signaling of several receptor tyrosine kinases such as PDGF and EGF receptors (; ). These data strongly suggest that PDZRN3 specifically regulates the surface levels of MuSK through the combination of its specific binding to MuSK and its E3 ubiquitin ligase activity. The ubiquitin system is best known for its function of adding polyubiquitin chains to target proteins, leading to the degradation of polyubiquitinated proteins through the 26S proteasome. However, a more recently discovered function of the ubiquitin system is to regulate membrane protein endocytosis through mono- or poly-ubiquitination (; ; ). To test whether PDZRN3 regulates MuSK activity through endocytosis, we compared the relative endocytosis of MuSK in the absence or presence of PDZRN3. To measure the endocytosis of MuSK, surface proteins from two identical sets of cells were biotinylated at 4°C. One set of biotinylated cells was kept on ice and used to measure initial surface levels of MuSK. The other set of biotinylated cells was moved to 37°C for 10 min to allow endocytosis to occur, and the remaining biotinylated surface proteins were then cleaved. Surface and endocytosed MuSK were detected by Western blot. The relative endocytosis was calculated as the ratio of endocytosed versus initial surface MuSK. As expected, expression of PDZRN3A reduced the surface level of MuSK (). This reduction of surface expression of MuSK was accompanied by increased endocytosis of MuSK (). In the presence of PDZRN3A, the ratio of endocytosed MuSK to initial surface MuSK is twice that of MuSK alone (). Ubiquitination of membrane proteins generally leads to endocytosis and subsequent degradation through the lysosome (; ). Consistent with our observation that PDZRN3 down-regulates the surface level of MuSK by increasing its endocytosis, we found that the addition of chloroquine (CLQ), a lysosome inhibitor that blocks lysosomal degradation and membrane recycling (; ; ; ), largely abolished the effect of PDZRN3 on MuSK surface expression. In contrast, blocking 26S proteasome function with MG-132 did not affect the PDZRN3-mediated down-regulation of MuSK surface expression (Fig. S3, available at ). To investigate the functional significance of PDZRN3 in regulating MuSK signaling, we examined MuSK-dependent AChR clustering in cultured myotubes. We used cultured C2C12 myoblasts, which differentiate into myotubes upon serum deprivation. The clustering of AChRs on these myotubes can be efficiently induced by the addition of exogenous agrin (), thus providing a system to dissect agrin–MuSK signaling without the interference of other factors from nerve terminals. Agrin stimulation of wild-type myotubes leads to the formation of numerous large clusters of AChRs (). However, agrin-induced AChR clustering is significantly attenuated in myotubes overexpressing either form of PDZRN3 (). Importantly, the deletion of the RING domain largely abolishes this effect (). Furthermore, coexpression of MuSK with PDZRN3 in myotubes blocks the effect of PDZRN3 on agrin-induced AChR clustering (), suggesting that overexpression of MuSK can antagonize PDZRN3. These data are consistent with our observation that PDZRN3 down- regulates the surface levels of MuSK when coexpressed in COS-7 cells and point to a potential role for PDZRN3 as an E3 ubiquitin ligase regulating postsynaptic development at the NMJ. Myotubes are multinucleated cells fused from individual myoblasts. To maximize the number of myoblasts expressing the siRNA constructs, we transfected myoblasts with siRNA constructs containing a Neo cassette in the vector. We then pooled all stably transfected myoblasts using G418 selection. These stably transfected myoblasts were cultured to fuse into myotubes, and they were found to fuse in a fashion indistinguishable from that of wild-type C2C12 cells (Fig. S4, available at ). Western blot analysis of endogenous PDZRN3 showed that myotubes expressing siRNA constructs have significantly reduced levels of endogenous PDZRN3 compared with myotubes expressing scrambled siRNA constructs (). To determine the effect of knockdown of PDZRN3 on MuSK signaling, we first measured the surface level of MuSK using biotinylation assays. As shown in , the surface level of MuSK in myotubes expressing the siRNA construct is significantly increased, indicating that endogenous PDZRN3 plays an important role in regulating the surface level of MuSK on myotubes. We next tested whether the increased surface level of MuSK in PDZRN3 knockdown myotubes also leads to an enhanced response to agrin stimulation. When stimulated with a saturating concentration of agrin (1 nM), myotubes expressing the PDZRN3 siRNA construct and the scrambled siRNA construct showed a similar response as measured by the number of AChR clusters (last data point in ), suggesting that myotubes stably transfected with siRNA constructs are capable of forming AChR clusters when stimulated with agrin. We then determined the dose–response curve of agrin stimulation. We found that at low concentrations of agrin, myotubes expressing the siRNA construct showed an enhanced response as indicated by the increased number of AChR clusters (). Furthermore, increased MuSK ubiquitination induced by agrin stimulation was not detected in myotubes expressing the PDZRN3 siRNA construct (). Together with the results from the gain-of-function studies, these data suggest a mechanism by which the activity of MuSK is regulated by the synapse-associated E3 ubiquitin ligase PDZRN3 to regulate postsynaptic development. To test whether modulation of PDZRN3 levels affect the development of the NMJ in vivo, we generated transgenic mice which overexpress PDZRN3 in skeletal muscle. We used the well-characterized promoter/enhancer elements from the myosin light chain (MLC) 1f/3f gene to drive HA-tagged PDZRN3A expression (). MLC 1f/3f promoter/enhancer elements have been shown to drive high levels of transgene expression in skeletal muscle but not in nonmuscle cells (; ). RT-PCR showed that HA-PDZRN3A transcripts are specifically expressed in the skeletal muscle (). The expression of mRNA and protein of HA-PDZRN3A could be detected as early as E15.5 (). Quantification of proteins from muscles of P21 mice indicates that there is a 1.8-fold increase of PDZRN3 proteins in transgenic mice compared with wild-type littermates (182 ± 6.9% of wildtype controls, = 3; ). Immunostaining showed that, like endogenous PDZRN3, HA-PDZRN3A is concentrated at the NMJ as revealed by co-staining with anti-HA antibodies and α-BTX (). To determine the effects of overexpression of PDZRN3A on the development of the NMJ, we first examined MuSK expression at the NMJ in MLC-PDZRN3A transgenic mice by immunostaining with a polyclonal anti-MuSK antibody (). In muscles of wild-type mice, MuSK is highly concentrated at the NMJ and precisely colocalized with AChRs (). In the MLC-PDZRN3A transgenic mice, a similar pattern of MuSK staining was observed. However, the intensity of the immunostaining of MuSK in the MLC-PDZRN3A transgenic mice was only 40% of that in wild-type littermates (), indicating that overexpression of PDZRN3A leads to reduced MuSK expression at the NMJ. We next examined the development of the NMJ using wholemount staining of gluteus muscle with anti-synaptophysin/anti-neurofilament antibodies and α-BTX. In wild-type mice, NMJs grow from “plaque-like” at P0 to “donut-like” by the end of the first postnatal week (). At 3 wk of age, NMJs in wild-type mice are well developed with complex patterns of multi-bifurcated presynaptic nerve terminals and “pretzel-like” perforated postsynaptic AChR clusters (). In the MLC-PDZRN3A transgenic mice, however, NMJs were much smaller in size at all three stages (P0, P8, and P21) examined (). Morphologically, NMJs in the transgenic mice remained plaque-like at P8. This is even more profound at P21; both presynaptic nerve terminals and postsynaptic AChR clusters showed much less complexity () compared with the wild-type littermates (). Postsynaptic AChR clusters of many NMJs were still plaque-like with few perforations, a hallmark of immature NMJs. To measure the complexity of the NMJ quantitatively, we counted the perforations of each NMJ and used this number as a “complexity index”. As shown in , NMJs in MLC-PDZRN3A transgenic mice contained many fewer perforations. These data strongly suggest that overexpression of PDZRN3A in muscle severely perturbs the growth and maturation of the NMJ. Although the importance of ubiquitin-dependent pathways in vertebrate synaptic development and plasticity has been well recognized, the specific E3 ubiquitin ligases involved in the ubiquitination of synaptic proteins are largely unknown (; ). In this study, we identified PDZRN3, a synapse-associated RING-type E3 ubiquitin ligase, as an important regulator of synaptic growth and maturation at the NMJ. We show that PDZRN3 is highly expressed in muscle and specifically localized to the postsynaptic site of the NMJ. PDZRN3 directly interacts with MuSK, and this interaction is enhanced by agrin stimulation. Coexpression of PDZRN3 and MuSK in heterologous cells promotes the ubiquitination of MuSK, and leads to enhanced endocytosis and reduced surface expression of MuSK. These effects require the RING domain of PDZRN3, which contains the E3 ubiquitin ligase activity. Overexpression of PDZRN3 leads to attenuated MuSK-dependent AChR clustering in cultured myotubes. Knockdown of endogenous PDZRN3 by RNAi in myotubes results in increased surface levels of MuSK and enhanced AChR clustering in response to agrin stimulation. Furthermore, overexpresion of PDZRN3 in muscle of transgenic mice perturbs the growth and maturation of the NMJ. Together, these data provide strong evidence that PDZRN3 functions as a synapse-associated E3 ubiquitin ligase to regulate the postsynaptic development of the NMJ. MuSK is essential for all aspects of postsynaptic differentiation of the NMJ (; ; ). Therefore, modulation of MuSK level/activity might serve as a central point of regulation in the formation and maintenance of the NMJ. In cultured myotubes, agrin-induced MuSK phosphorylation is transient, suggesting the existence of mechanisms rapidly regulating MuSK level and/or phosphorylation (; ). Transgenic expression of a constitutively active MuSK in mice leads to the formation of extrasynaptic clusters of AChRs (). Furthermore, patients with a missense mutation of MuSK that leads to low levels of MuSK expression have reduced AChR clusters and show signs of myasthenic syndrome (). This demonstrates the importance of precise control of MuSK level/activity in the development and function of the NMJ. Although several MuSK-interacting proteins have been identified that may serve as points of regulation (, ; , ; ; ; ; ), our study presents a molecular mechanism of direct regulation of MuSK activity through the modulation of its surface levels. The enrichment of PDZRN3 mRNA in the synaptic region of the muscle and the concentration of PDZRN3 protein at the NMJ are consistent with our findings of its role in regulating MuSK signaling. However, a recent study also showed that PDZRN3 is expressed in cultured myoblasts and that its expression is up-regulated during the differentiation of C2C12 myoblasts into myotubes (). Knockdown of PDZRN3 in cultured C2C12 cells inhibits the formation of myotubes, though overexpression of PDZRN3 has no effect on the differentiation of myoblasts into myotubes (). This is in contrast with results of our RNAi experiments, which show no effects on the formation of myotubes. In our study, C2C12 cells expressing the RNAi construct for PDZRN3 grow and fuse in a fashion indistinguishable from wild-type C2C12 myotubes. Furthermore, these myotubes have a similar response to high concentrations of agrin stimulation as control myotubes. Currently, the reason for this discrepancy is not clear. One obvious difference is the method of expressing RNAi constructs in C2C12 cells. used a transient expression approach by transfecting the same batch of C2C12 cells for three consecutive times at 24-h intervals to reach high transfection efficiency. We selected stably transfected C2C12 cells to enrich cells expressing the RNAi construct. In general, the expression level of exogenous gene is lower in stably transfected cells than in transiently transfected cells due to lower copy numbers. We saw ∼75% reduction of endogenous PDZRN3 proteins in myotubes expressing the RNAi construct for PDZRN3. Future gene knockout studies in mice will likely resolve this difference. In mice, NMJs start to form around E14.5 (). We have shown that PDZRN3 protein is first detected at the NMJ around E16.5, suggesting that the role of PDZRN3 at the NMJ is unlikely to be in the initial formation of the NMJ, but rather in the growth and/or maturation of the NMJ. This role is further supported by our study of transgenic mice. Overexpression of PDZRN3 in skeletal muscle leads to reduced MuSK expression at the NMJ and smaller and less mature NMJs. Because PDZRN3 contains multiple protein–protein interaction domains, we cannot exclude the possibility that overexpression of PDZRN3 may affect additional signaling pathways in muscles. However, our results are consistent with the proposed role of synaptic E3 ubiquitin ligases in regulating homeostatic growth of synapses in and , in which mutations of E3 ubiquitin ligases lead to aberrant growth of synapses (; ; ; ). Although ubiquitination-mediated protein degradation through the 26S proteasome is the best characterized function of the ubiquitin system, increasing evidence suggests that ubiquitination-mediated endocytosis and lysosomal degradation play an important role in regulating cell surface abundance of membrane proteins (). Initial studies from yeast indicate that monoubiquitination of membrane proteins is sufficient for triggering endocytosis (). However, recent studies suggest that polyubiquitination, rather than monoubiquitination, is required as a signal for endocytosis in mammalian cells (; ). Our data strongly suggest that PDZRN3 functions as an E3 ubiquitin ligase to regulate surface levels of MuSK through ubiquitination-dependent endocytosis. Ubiquitination-mediated endocytosis has been shown to be a key mechanism of down-regulating receptor tyrosine kinase activity in many cellular processes (). Recently, a putative Ariadne-like ubiquitin ligase (PAUL) has been found to be present at the NMJ and to interact with MuSK (). Together, these synaptically localized E3 ubiquitin ligases may play an important role in postsynaptic development at the NMJ through the regulation of surface levels of MuSK. PDZRN3AΔRING, PDZRN3AΔRZ, and PDZRN3AΔPDZ were generated by deleting amino acids 1–51, 1–157, and 246–338 of PDZRN3A, respectively. Kinase-dead MuSK was generated by replacing Lys608 with Ala, and kinase-active MuSK was generated by deleting amino acids 99–492 from the ectodomain of MuSK (). C2C12 myotubes or transfected COS-7 cells were harvested and lysed in lysis buffer containing PBS, 1% Triton X-100, and Complete protease inhibitors (Roche). Immunoprecipitation was performed by incubating samples with appropriate antibodies and 25 μl of protein A agarose beads (Invitrogen) for 2–4 h. Proteins were eluted by the addition of loading buffer and analyzed by Western blot. Tibialis anterior muscle was denervated, sectioned, and stained as described previously (). Fluorescent images were taken using an Axioskop2 Plus microscope (Carl Zeiss MicroImaging, Inc.) fitted with an AxioCam CCD camera (Carl Zeiss MicroImaging, Inc.) through a 20× objective (NA = 0.75) with acquisition software AxioVision 3.1 (Carl Zeiss MicroImaging, Inc.). Images were then adjusted with Adobe Photoshop 6.0 (cropping and brightness/contrast adjustments). Final figures were mounted and labeled using Adobe Illustrator 9.0. C2C12 myoblasts were plated in 6-well plates and transfected with a total of 2 μg DNA and 3 μl FuGENE 6 (Roche) per well. 36–48 h after transfection, myoblasts were switched to fusion medium (DME with 2% horse serum) to allow myoblasts to fuse into myotubes for 4–5 d. After stimulation with 1 nM agrin (R&D Systems) for 12–16 h, myotubes were fixed and stained with appropriate antibodies. C2C12 myotubes or transfected COS-7 cells were washed with PBS and surface biotinylated as described previously (). For the endocytosis assay, two sets of identically transfected COS-7 cells were prepared and surface biotinylated. One set of samples was moved to 37°C for 10 min to allow endocytosis to occur. The remaining surface biotin was then cleaved with glutathione cleavage buffer (50 mM glutathione, 75 mM NaCl, 10 mM EDTA, 1% BSA, and 0.075 N NaOH). The other set was kept on ice without subsequent cleavage and used to determine the initial surface levels of MuSK. The relative endocytosis of MuSK from the cell surface was then calculated by determining the ratio of biotinylated MuSK after cleavage to the initial surface-biotinylated MuSK. The in vitro ubiquitination assay was performed by incubating GST-FLAG-RING (GST and FLAG tagged RING domain of PDZRN3A) with 5 nM yeast E1, 100 nM of the indicated E2, and ubiquitin (Boston Biochem) in ubiquitination buffer (50 mM Tris-HCl, pH 7.5, 4 mM MgCl, 1 mM DTT, 1 mM ATP, 10 mM creatine phosphate, and 16 IU/ml creatine phosphokinase) to a final volume of 40 μl. Samples were incubated at 30°C for 2 h and subsequently analyzed by Western blot. To detect ubiquitination of MuSK, transfected COS-7 cells or C2C12 myotubes stimulated with 10 nM agrin for 2 h were harvested in lysis buffer containing 2% SDS. After incubating at 4°C for 1 h, samples were diluted 1:20 using lysis buffer without SDS, and MuSK was immunoprecipitated and analyzed by Western blot with an anti-ubiquitin antibody (Sigma-Aldrich). GST or GST-UbcH5B proteins were immobilized on glutathione-sepharose beads (Novagen). Whole cell lysate from PDZRN3A transfected COS-7 cells was added and incubated with these beads at 4°C for 1–2 h. Proteins were then eluted and analyzed by Western blot. The full-length mouse PDZRN3A sequence was copied into the siRNA design center of DHARMACON () and four oligo sequences against PDZRN3A were generated. These oligo sequences were inserted into pSuper vector and cotransfected with PDZRN3A into COS-7 cells. Western blot analysis showed that one oligo (sequence 5′-GTCGGTGACTACTGTATAA-3′) had the most dramatic effect in knocking down expression of PDZRN3A. C2C12 stable cell lines expressing either this oligo or another oligo with a scrambled sequence were then generated by transfection of C2C12 cells with subsequent G418 selection. The MLC-PDZRN3A transgene was constructed by inserting HA-tagged mouse PDZRN3A cDNA in the vector containing the promoter and enhancer elements from the rat MLC 1f/3f gene (). Transgenic mice were generated by injection of DNA into the fertilized oocyte using standard pronuclear injection techniques. Transgenic founders were subsequently backcrossed to C57BL6/J mice for two to four generations before analysis. Genotypes were determined by PCR, and wild-type littermates of transgenic mice were used as controls. Fig. S1 A shows weak staining of extrasynaptic muscle membrane by anti-PDZRN3 antibody; B and C shows that MuSK staining persists after denervation. Fig. S2 shows how regulation of surface levels of MuSK by PDZRN3 is specific. Fig. S3 describes how down-regulation of surface levels of MuSK by PDZRN3 is blocked by the lysosome inhibitor chloroquine, but not the proteasome inhibitor MG-132. Fig. S4 shows C2C12 cells stably transfected with PDZRN3 siRNA fuse in a fashion indistinguishable from controls. Online supplemental material is available at .

Parkinson's disease is characterized by the selective degeneration of dopamine-producing neurons that comprise the and the presence of proteinaceous inclusion bodies (Lewy bodies) in the affected neurons (). The principal component of Lewy bodies is α-synuclein (α-syn), which is an intrinsically unfolded protein of unknown function (; ). Wild-type (WT) α-syn and two mutants, A30P and A53T (; ), associated with early onset PD have been linked to a plethora of defects, including proteasomal and mitochondrial dysfunction (; ), the accumulation of reactive oxygen species (ROS) (), blockage of ER to Golgi traffic (), and histone acetylation inhibition (). Using a genetic screen that exploits the super sensitivity of α-syn–expressing yeast cells to killing by HO (), we discovered that , which is an essential gene of unknown function, suppresses the toxicity of the mutant α-syn A30P. Because humans have several possible orthologues, understanding the function of this gene may shed light on how human neurons protect themselves from α-syn. Given that codes for an α-synuclein protective protein, was named . Until quite recently, other than its deletion kills cells (), scant information existed about . A recent study using a technique to probe the spectrum of synthetic genetic interactions among essential genes revealed that interacts genetically with , , , , and (). Other studies have indicated that functions in MAPK pathways (; ; ). Ypp1p-GFP localizes in a punctate pattern around the plasma membrane (), although there is no indication from its sequence that Ypp1p is a membrane protein. Although information exists about its synthetic genetic interactions and localization, Ypp1p's function has remained obscure. Here, we show that Ypp1p binds A30P (but not WT or A53T) and mediates a sequence of events in which A30P is encapsulated into vesicles at the plasma membrane, and the vesicles then transit to and merge with the vacuole, where the A30P protein is proteolytically degraded. A high copy yeast genomic library was used to identify suppressors of the super sensitivity of A30P expressing cells to killing by hydrogen peroxide (). Herein, we describe the characterization of one suppressor, , which is an essential gene of unknown function. gives the strains and plasmids used in this study. codes for a protein with a theoretical molecular mass of 95.4 kD. Cells expressing α-syn and transformed with a plasmid harboring a myc-tagged fusion indeed expressed myc-Ypp1p, as judged by Western blot analysis and staining with an anti-myc antibody (). Specifically, myc-Ypp1p was detected in cells that coexpressed WT α-syn or A30P (lanes 2 and 4); however, in each case, a ladder of bands ranging from 95 to >200 kD was observed. We attribute the band at 95 kD to myc-Ypp1p and suggest that the bands at higher molecular mass are posttranslationally modified forms of Ypp1p. The four major bands of myc-tagged Ypp1p were absent in lysates of cells that did not harbor the plasmid (lanes 1 and 3). The effect of Ypp1p overexpression on the viability of cells expressing the various α-syns was also evaluated. A viability assay () was conducted using the dye FUN1 on cells induced for 12 h. Dead cells stained green; metabolically active, and hence viable cells, stained red; and a small percentage of cells (∼10%) failed to stain. The percentage of red cells indicated viability. For cells expressing A30P with Ypp1p overexpression, 90% of the cells were viable, whereas only 10% of cells were viable when A30P was expressed without Ypp1p overexpression (). In contrast, for cells expressing WT α-syn or A53T, with or without Ypp1p overexpression, only ∼10% of the cells were viable (). This viability assay demonstrated that in high copy specifically enhances the viability of A30P expressing cells but not of cells expressing the other two α-syns (WT or A53T). A signature feature of α-syn is that it induces oxidative stress in various types of cells (; ; ). Given that in high copy suppresses the super sensitivity of A30P expressing cells to killing by hydrogen peroxide, we expected that Ypp1p overexpression would abolish ROS production in A30P expressing yeast cells (), but not in cells expressing WT or A53T α-syn. To test this hypothesis, cells expressing α-syn (WT, A30P, or A53T) with decreased, endogenous, or increased levels of Ypp1p were stained with the cell-permeant dye DHR 123. This dye enters cells, and when oxidized by free radicals yields a fluorescent product (). For these experiments a strain from the “Hughes collection” of titratable promoter alleles was used () in which a kan-tetO-TATA cassette is integrated into the promoter of . Repression is controlled by adding doxycycline, which has no appreciable effect on global gene expression at the concentrations (20 μg/ml) used for the promoter shut off. shows ROS accumulation in cells expressing WT α-syn, A30P, or empty vector controls. ROS accumulation in cells expressing WT α-syn (or A53T) was insensitive to variations in the level of Ypp1p (compare the three vertical panels labeled DHR in ). In contrast, ROS accumulation in cells expressing A30P was exquisitively sensitive to the level of Ypp1p: 100% of the cells exhibited intense red fluorescence when the level of Ypp1p was decreased (+doxy), whereas only ∼5% of the cells exhibited red fluorescence when Ypp1p was overexpressed (compare the three vertical panels labeled DHR in ). Identically treated control cells harboring the plasmid with no insert exhibited no appreciable ROS accumulation (). These experiments revealed that Ypp1p, when overexpressed, abolished ROS accumulation in cells expressing A30P—but not in cells expressing WT α-syn or A53T (). Coimmunoprecipitation experiments were performed to determine whether Ypp1p physically associated with A30P. Coimmunoprecipitations of cells overexpressing Ypp1p with WT α-syn, A30P, or A53T coexpressed are shown in (lanes 1–6). For each sample, one lane contained the lysate and the other lane contained the myc-Ypp1p pull down. The myc-Ypp1p (∼95-kD band) was visualized with an anti-myc antibody and the α-syns (∼19 kD) with an anti-α-syn antibody. A comparison of the two blots showed that Ypp1p pulled down A30P (lane 4), but not WT α-syn or A53T (lanes 2 and 6). The experiments revealed that, when overexpressed, Ypp1p associates with A30P. We also found that in high copy permitted normal growth of cells expressing A30P but not of cells expressing WT or A53T (Fig. S1, available at ), and that in high copy failed to protect yeast cells from hydrogen peroxide-induced ROS (Fig. S2). is thus unlikely to code for an enzyme that inactivates hydrogen peroxide. To gain insight into the mechanism of suppression, fluorescence microscopy studies were conducted to determine the effect of Ypp1p overexpression on the localization of the various GFP-tagged α-syns. We found that Ypp1p overexpression altered the localization of each of the three GFP-tagged α-syns (WT, A30P, or A53T), but in subtly distinct ways compared with control cells without Ypp1p overexpression (). For example, cells expressing GFP-A30P with Ypp1p overexpression exhibited 3 to 6 inclusions per cell at 3 h of induction, whereas at 12 h the inclusions had coalesced into 1 to 2 larger inclusions (). Without Ypp1p overexpression, GFP-A30P expressing cells exhibited diffuse green fluorescence at 3 and 12 h. In contrast, Ypp1p, when overexpressed, even drove GFP-WT α-syn and GFP-A53T from the plasma membrane into inclusions, but in each case the inclusions failed to efficiently merge with one another (). Notice that control cells expressing GFP exhibited diffuse green fluorescence that was unaffected by increased levels of Ypp1p (). The experiments revealed that Ypp1p when overexpressed drives the various α-syns into inclusions. That Ypp1p concentrated GFP-A30P into 1 to 2 inclusions per cell raised the possibility that Ypp1p also drives A30P into a compartment such as the vacuole. To test the hypothesis that Ypp1p drives A30P to the lysosome/vacuole, two-color fluorescence microscopy experiments were conducted using cells expressing the various GFP-tagged α-syns and the lipophilic dye FM4-64, which stains vacuoles (). At various times (3 or 12 h) after induction, cells expressing GFP-α-syn were incubated with the FM4-64 dye and microscopy images were acquired (). If GFP-α-syn enters the vacuole then the green fluorescence from GFP and the red fluorescence from FM4-64 should overlap. No overlap between any of the GFP-α-syn inclusions with the red structures occurred after 3 h of induction, whereas almost every one of the GFP-A30P inclusions overlapped with the red structures after 12 h of induction. No such overlap occurred for GFP-WT α-syn or GFP-A53T structures with vacuoles. However, because vacuoles were not prominent in the DIC images (12 h), these experiments do not prove that Ypp1p drove A30P into the vacuole. For this reason, the following biochemical analysis was conducted. Ypp1p-mediated transport of A30P to the vacuole was tested using Western blot analysis to monitor the level of the A30P protein. Cells transformed with the various plasmids were pregrown in noninducing media to mid-log phase, shifted to inducing media, inhibited with 10 μM cycloheximide to halt protein synthesis, and incubated for 12 h. Aliquots were removed at the indicated times and extracts were prepared, subjected to SDS-PAGE followed by Western blot analysis. First, we found that A30P was rapidly degraded (t ∼1 h) in cells overexpressing Ypp1p (), whereas in cells expressing A30P, but with no Ypp1p overexpression, no such degradation of A30P occurred (). Second, if A30P is degraded in the vacuole, then A30P should fail to be degraded in a strain that lacks , which is a gene that codes for a protease that activates a variety of vacuolar proteases (). To test this hypothesis, the level of A30P was monitored in a Δ deletion strain, and the A30P protein showed no appreciable degradation in this strain (). Third, to rule out the possibility that Ypp1p mediates the proteasomal degradation of A30P, experiments were also conducted using the proteasome inhibitor MG132 (50 μM). With protein synthesis inhibited and with proteasome function inhibited A30P was still degraded (), albeit not with the exact kinetics as observed when only cycloheximide was used. The ability of the deletion to halt A30P degradation, combined with the lack of appreciable inhibition of A30P degradation by a proteasome inhibitor, indicated that Ypp1p mediates the transport of A30P to the vacuole, where it is proteolytically degraded. Transmission electron microscopy was conducted to gain insight into the nature of the α-syn inclusions in +A30P/+Ypp1 cells. A30P expression induced dramatic morphological changes in cells at only 3 h of induction compared with control cells (). Changes included granulation of the cytosol and chromatin condensation. In contrast, control cells containing two empty plasmids (−A30P/−Ypp1) had a well-delineated vacuole and cytoplasm. Cells expressing A30P with Ypp1p overexpressed exhibited smaller vacuoles than cells with the two empty plasmids and had numerous mitochondria per slice. Vesicles budding off of the plasma membrane were also observed in these cells. These vesicles were absent in cells expressing only A30P. Such a vesicle can be seen in the bottom left image in (denoted by the white arrow). At higher magnification the vesicle appears to be emerging from the plasma membrane (). After 12 h of induction, cells expressing only A30P were characterized by a granulated cytosol and even more extensive chromatin condensation compared with 3 h induction. Comparing this TEM image to the images of the −A30P/−Ypp1 control cells and +A30P/+Ypp1 cells revealed the following differences: control cells with two empty plasmids exhibited a granulated cytosol and chromatin condensation (). These changes did not occur when the strain was incubated in glucose or sucrose for 12 h. The changes occurred because of the abrupt shift from sucrose to galactose. It turns out that the wt S288c strain, but not the Resgen deletion collection, which is derived from the S288c strain, has a mutation in . After ∼15 generations WT S288c cells begin to lose viability. Using the FUN1 assay we found that 90% and 60% of the WT S288c cells were viable after 3 h and 12 h in galactose (unpublished data), respectively. It should be pointed out that in high copy suppressed A30P-induced ROS at 12 h in strains FY23, S288c, and various S288c gene deletion strains. The +A30P/+Ypp1 cells were characterized by a normal nucleus, numerous mitochondria per optical slice, and three large vacuoles (denoted by white arrows; ). More than 90% of the EM sections examined contained such vacuoles. Immunogold labeling was also performed to visualize the subcellular location of A30P. Cells expressing A30P and overexpressing Ypp1p were characterized by numerous clusters of gold particles in association with the plasma membrane (). In contrast, cells expressing A30P only, or harboring the two control plasmids, did not have gold particles in association with the plasma membrane. Therefore, we attributed the membrane-associated clusters to A30P molecules that were being packaged by Ypp1p into endocytic vesicles. Note that the membrane-associated clusters are smaller than the budding vesicle (compare ); this could be because of the difficulty of preserving membrane structure for immunogold labeling and also because only a small portion of a vesicle contains A30P. Clusters of gold particles were also evident inside the cells (). The TEM results showed that in high copy protects cells from A30P. To begin to define the pathway by which suppresses A30P toxicity, a targeted screen of 116 nonessential genes involved in autophagy, cytosol-to-vacuole transport (Cvt), endocytosis, and the vacuole protein sorting pathway (vps) was conducted ( and Table S1 for entire list, available at ). We reasoned that any gene that cooperates with to protect against A30P, when deleted, should increase ROS accumulation. The ROS assay using the DHR 123 dye was conducted in 96-well plates on cells induced for 3 h using a plate reader with fluorescence detection. The various deletion strains were transformed with the pTF302 (A30P) and pTF602 () plasmids. The vacuolar protein sorting pathway (vps) is a network of genes involved in the transport of newly synthesized proteins from the late Golgi to the vacuole (). 56 deletion strains were analyzed, and of these 56 strains 16 gave statistically significant increases in ROS (). Curiously, 11 of the 16 hits clustered in the class E genes. Many of the class E genes code for ESCRT () proteins that function to sort membrane-bound proteins into lumenal multivesicular body vesicles (MVB). Our interpretation of these findings is that the deletion of certain class E genes resulted in the failure of A30P to be sorted into lumenal MVB vesicles, and this prevented A30P from being delivered to and degraded within the vacuole. Failure to degrade A30P resulted in the accumulation of ROS. Data obtained from selected deletion strains associated with actin organization, autophagy, Cvt, endocytosis, vacuolar biogenesis, and vesicle–vacuole fusion, are presented in and Table S1. First, five deletion strains gave very large (>200,000 units; P < 0.001) increases in ROS compared with the WT control (25,818 ± 7005 units) (). These five strains were: Δ, Δ, Δ, Δ, and Δ. Sla1, Sla2, and End3 proteins form a complex on the lumenal side of the plasma membrane (; ), where they regulate actin dynamics and proteins required for endocytosis (). The Ccz1p–Mon1p complex is required for nearly all membrane-trafficking pathways where the terminal acceptor compartment is the vacuole (). Second, six deletion strains gave large (>100,000 units; P < 0.001) increases in ROS compared with the WT control (). These strains were: Δ, Δ, Δ, Δ, Δ, and Δ. In general, these genes have roles in vesicle trafficking and actin filament organization. Third, eleven genes involved in the autophagy and Cvt pathways, when deleted, failed to increase ROS (Table S1). The broad picture that emerged from this screen was that suppressed A30P toxicity via a pathway involving , , and , class E genes, and and . Given that these genes are involved in endocytosis and vesicle trafficking to the vacuole, such a finding, together with the fluorescence and TEM data (– ), indicate a role for in the endocytic pathway. Latrunculin A, a known inhibitor of actin (), disrupts the trafficking of endosomal vesicles from the plasma membrane to the vacuole (). Latrunculin A (25 μM)–treated cells expressing A30P and overexpressing Ypp1p indeed exhibited a large increase in ROS signal (284,979 ± 10,816) compared with identically treated cells that lacked latrunculin A (25,818 ± 7005) (). Our interpretation of this result is that latrunculin A inhibits the formation of actin filaments which in turn blocks the budding of endocytic vesicles containing A30P and delivery of A30P to the vacuole. To verify results obtained from the ROS plate reader assay, the Δ and Δ strains were selected for further analysis. The S288c control strain expressing GFP-A30P and overexpressing Ypp1p exhibited 4 to 5 inclusions per cell at 3 h, and these inclusions had coalesced into 1 to 3 inclusions per cell at 12 h (). The Δ strain expressing GFP-A30P and overexpressing Ypp1p also exhibited 4 to 5 inclusions per cell at 3 h (), but at 12 h multiple inclusions per cell were still evident. Given its role in catalyzing the fusion of vesicles with the vacuole (), deletion of would be expected to hinder fusion of A30P-containing vesicles with the vacuole, and the microscopy images are consistent with such a defect. Notice that adding back on a plasmid rescued the inability of vesicles to coalesce at 12 h in the Δ strain (). These experiments confirmed that is involved in the pathway by which rids cells of A30P. The effect of the Δ deletion on Ypp1p-mediated trafficking of GFP-A30P was also analyzed. After 3 or 12 h of induction Δ cells expressing GFP-A30P and overexpressing Ypp1p displayed green fluorescence throughout the cell (). These images are very different from the images acquired from the S288c and Δ strains. Our interpretation of the Δ images is that deletion of prevents the packaging of GFP-A30P into vesicles. Notice that adding back on a plasmid resulted in the formation of ∼0–3 inclusions per cell after 3 h of induction; whereas, after 12 h of induction the inclusions had coalesced yielding 1–2 inclusions per cell (). These images from 12 h are remarkably similar to the images obtained at 12 h from the WT strain. Accordingly, we concluded that deletion of resulted in a failure of cells to package A30P into vesicles. An issue is whether A30P transits to the vacuole via the endocytic pathway, as proposed, or the secretory pathway. ER stress occurs when unfolded proteins accumulate in the ER, and such proteins are typically retrotranslocated out of the ER to the proteasome for degradation; this process is called ERAD (endoplasmic reticulum–associated degradation) (). does not facilitate retrotranslocation and ERAD because A30P was still degraded when the proteasome was inhibited (). However, yeast have another pathway for the degradation of soluble, unfolded proteins that transit through the ER and Golgi, and this occurs via receptor-mediated forward transport of unfolded luminal proteins to the vacuole (). The receptor for this process is coded for by . Because deletion of caused intense ROS accumulation in cells expressing A30P (), perhaps is an enhancer of the forward transport of luminal proteins to the vacuole. On the other hand, although WT α-syn and A53T use the secretory pathway and cause ER stress (Cooper et al., 2006), such findings have not been reported for A30P. Instead, evidence exists that A30P does not use the secretory pathway (). We examined whether loss of function of one early-acting and two late-acting genes, i.e., , , and , respectively, affected -mediated trafficking of A30P to the vacuole. Sec12p controls transport vesicle budding from the ER (). Sec1p controls the fusion of secretory vesicles with the plasma membrane (), and when exocytosis is shut down actin regulation also becomes disrupted (). The Sec5 protein is a component of the essential exocyst complex, which tethers secretory vesicles to the plasma membrane (). If mediates A30P trafficking through the ER-Golgi to the vacuole, then this trafficking should be disrupted in the but not in the or strains. Conversely, if mediates A30P trafficking to the vacuole via endocytosis, then this trafficking should be disrupted in the and but not in the strains. The Hughes collection of titratable promoter alleles () was used in these experiments. Decreasing the level of the essential Sec12 protein had no appreciable effect on -mediated trafficking of A30P to the vacuole. Specifically, cells expressing GFP-A30P and overexpressing Ypp1p, but with a decreased level of Sec12p, exhibited 1–2 large GFP-A30P inclusions per cell, which coincided with the vacuole (). In contrast, decreasing the level of the essential Sec1 protein abolished -mediated trafficking of A30P to the vacuole. Specifically, cells expressing GFP-A30P and overexpressing Ypp1p, but with a decreased level of Sec1p, exhibited no GFP-A30P inclusions after 3 or 12 h of induction (). Similar results were obtained for cells with a decreased level of the Sec5 protein (unpublished data). Because decreasing the level of Sec1p, or Sec5p, or deleting eliminated -mediated transport of A30P to the vacuole, we concluded that mediates the trafficking of A30P to the vacuole via the endocytic pathway. Several reports have indicated that functions in a MAPK pathway (; ; ). Because the yeast mating response is a MAPK pathway () that involves receptor-mediated endocytosis (RME), we asked whether functions in the yeast mating response (). In yeast, the response to pheromone is a well characterized example of RME: α-factor pheromone binds to Ste2p (on Mat cells) and triggers a sequence of events in which Ste2p and the α-factor are encapsulated into endosomes, which then transit along actin cables to the vacuole, where the pheromone and its receptor are proteolytically degraded (). The hypothesis that participates in RME was tested by incubating α-factor with a Mat haploid strain containing an integrated allele replacing the WT allele (). The Mat haploid strain expressing Ypp1p-GFP exhibited no change in the localization of Ypp1p-GFP upon incubation for 20 min, and no change in localization was detected even after hours of incubation. Weak, diffuse green fluorescence occurred throughout the cytosol in these cells, and somewhat more brightly fluorescent puncta appeared around the periphery of the cells (). The enhanced staining around the periphery was consistent with Ypp1p-GFP associating with actin cortical patches, which are thought to be the sites of exocytosis and endocytosis. In contrast, treatment of this strain with α-factor resulted in a rapid change in the localization of Ypp1p-GFP and a parallel increase in the fluorescence signal (). Cells exhibited numerous vesicles that appeared to coalesce into larger structures after 20 min. The green inclusions of Ypp1p-GFP merged with the vacuolar structures in the DIC images. These experiments showed that the endosomes formed in response to pheromone contained Ypp1p-GFP, and that such Ypp1p-GFP–containing endosomes then merged with the vacuole. This data strongly supported our conclusion that Ypp1p is involved in endocytosis. Intrigued by the ability of pheromone to drive Ypp1p into vesicles, the hypothesis that pheromone itself could protect cells from the toxicity of the A30P protein—by driving A30P into endosomes, which then traffic to the vacuole, where A30P is degraded—was tested. The experiment was conducted by transforming the FY23 Mat haploid strain with plasmids for the various α-syns (WT, A30P, or A53T), pregrowing cells in noninducing media to mid-log phase, and shifting cells to inducing media for 3 h. The α-factor pheromone (5–10 μM) was added upon the shift into inducing media; the DHR 123 dye was added after 2 h in inducing media; and the images were acquired after 3 h in inducing media. The α-factor had no effect on ROS accumulation in cells expressing WT α-syn (or A53T), as judged by red fluorescence (). In contrast, in identically treated cells expressing A30P α-factor decreased the percentage of red staining A30P-expressing cells to 26.3 ± 4.2% from 69.2 ± 8.5% (P = 1.1 × 10) (). No appreciable amounts of ROS occurred in vector control cells with or without pheromone (). Our interpretation of these results was that pheromone triggered rapid RME, and because of Ypp1p's ability to bind A30P at the plasma membrane, A30P was encapsulated into the endocytic vesicles, which then trafficked to the vacuole. The biochemical, cell biological, and genetic experiments in this study have demonstrated a role for the essential gene in endocytosis. We conducted a PSI-blast analysis of , using the Genome Database, and found six potential functional homologues (orthologues) in humans. One of these, human TTC7B (Q86TV6), has 15% identical residues with . The TTC7B protein contains multiple tetratricopeptide repeat (TPR) protein–protein interaction domains (), and it is likely that the Ypp1 protein also contains such domains. Other than TPR domains, Ypp1p has no recognizable domains. Next, several issues are discussed relating to the role of Ypp1p in endocytosis and protection it affords against A30P. Notice that A30P expression in the Δ strain with in high copy caused no appreciable toxicity, as judged by ROS accumulation (). We propose that in high copy, whether in WT cells or Δ cells, mediates the transport of A30P to vacuole. The lack of toxicity of A30P in Δ cells with in high copy is due to the removal of A30P from the plasma membrane and cytosol and sequestration in vacuoles. Suppression of A30P toxicity could be linked more to its sequestration in the vacuole rather than its degradation. If Ypp1p functions in the endocytic pathway, why does Ypp1p select A30P, which appears to be a cytosolic protein, and target it to the vacuole instead of WT α-syn or A53T, which are membrane bound? Points addressing the selectivity of Ypp1p for A30P are: A30P is not only a cytosolic protein; it also associates with membranes, albeit not as strongly as WT α-syn or A53T, and permeablizes the membranes to which it binds (; ). Therefore, membrane-associated A30P should be able to associate with membrane-associated Ypp1p. When overexpressed, Ypp1p rapidly packaged each of the α-syns (WT, A30P, and A53T) into inclusions/vesicles, albeit only A30P-containing vesicles then merged with the vacuole (, , and ). The transport of A30P-containing vesicles to the vacuole is no doubt casually connected to the binding of A30P to Ypp1p. Such results suggest two separable functions for Ypp1p. The incomplete endocytosis of WT and A53T α-syn—which is proposed to be due to their inability to bind Ypp1p—results in the accumulation of Lewy body–like structures. It is important to recognize that a spectrum of protein–protein equilibrium constants, from nanomolar to micromolar, occurs in cells. Ypp1p may weakly associate with WT α-syn or A53T, but the short-lived complexes cannot be pulled out of solution. The A→P substitution could increase the strength of the association between α-syn and Ypp1p for many reasons. We reason that, compared with WT α-syn and A53T, A30P may adopt a different conformation when membrane bound, it may possess a different phosphorylation pattern, and it may be ubiquitinylated at a different site. These unique features could foster a strong association between A30P and Ypp1p, resulting in kinetically stable complexes that can be pulled out of solution. WT α-syn and the A53T mutant cause ER stress (; ; Cooper et al., 2006), and, above a certain threshold concentration, each protein inhibits forward vesicle transport through the ER (Cooper et al., 2006). Alleles that suppress α-syn toxicity (WT and A53T) enhance forward ER-Golgi transport, and these suppressors are , , and , and in high copy failed to suppress the toxicity of WT α-syn and A53T, thus does not enhance forward ER-Golgi traffic vis-à-vis these two α-syns. Regarding A30P, A30P has not been shown to cause ER stress or inhibit ER-Golgi traffic, and our experiments gave no indication that enhances the forward transport of A30P through the ER-Golgi (). In summary, interacts with the evolutionarily conserved genes , , , class E , and and to target A30P α-syn to the vacuole for degradation via the endocytic pathway (). It will be of interest to characterize the step or steps that Ypp1p catalyzes in the endocytic pathway, correlate Ypp1p structure with function, and determine whether the potential human orthologues mediate the trafficking of A30P to the lysosome in human cells. The yeast strains used in this study were FY23 (), S288c (Invitrogen), YPP1-GFP (Invitrogen), which contains an integrated copy of replacing the WT allele, and several strains from the Hughes collection (Open Biosystems) ().Additionally, 116 strains from the ResGen deletion collection were also used; these strains were derived from the parental strain S288c. Cells transformed with various plasmids were pregrown in synthetic sucrose (2% wt/vol) drop out media to maintain selection for plasmids. Media containing sucrose is referred to as noninducing media. α-Syn expression, as well as Ypp1p overexpression, was induced in the same drop out media with galactose (2% wt/vol) replacing sucrose (). Media containing galactose is referred to as inducing media. The various S288c deletion strains were cultured in noninducing media with selection for plasmids supplemented with 0.4 g/L G418 (Geneticin) or inducing media with selection for plasmids supplemented with 0.4 g/L G418. The Resgen deletion strains (Invitrogen) and the high copy yeast genomic library () (YEp13 ) were gifts from Dr. Kelly Tatchell (LSU Health Sciences Center, Shreveport, LA). In every experiment in this study, cells were pregrown in noninducing media (sucrose) with selection for plasmids to mid-log phase and then shifted to inducing media (galactose) with selection for plasmids and induced for 3 h. In some cases, 12-h inductions were used. Cells were grown at 30°C. was cloned from genomic DNA using the forward and reverse primers 5′-CCACTGGATCCATGCCTAACTCAAATGTTC-3′ and 5′-GAAGAAGAGCTCTTAGTAATTCGAATACCTTAG-3′, respectively. These primers contained the BamHI and SacI restriction sites. The PCR product was subcloned into the BamHI and SacI restriction sites on the 2-μm pRS326 plasmid. For all other constructs, the ORF was excised from the pRS326 plasmid at the BamHI and SacI sites and ligated into those same sites in the other plasmids. Preparation of the plasmids harboring untagged and GFP-tagged α-syns was described elsewhere (). was cloned from genomic DNA using the forward and reverse primers 5′-CATAGACGGGCCGCATGGGTCTAGCTTAAATAC-3′ and 5′-AACCTCGAGCTCACAAGTTAAAACACGGCC-3′, respectively. These primers contained the NotI and SacI restriction sites; the PCR product was subcloned into the same restriction sites on the low copy pRS313 plasmid. was cloned from genomic DNA using the forward and reverse primers 5′-CAGGTGACTAGTGAAGTAGGCCATTCACTGC-3′ and 5′-TCTTAACCGCGGTTAATCTAGAATCCAAACGGATTTTG-3′ These primers contained SpeI and SacII restriction sites; the PCR product was subcloned into the same restriction sites on the low copy pRS313 plasmid. Plasmids were propagated according to standard protocols using DH5α (). Cells transformed with pTF202 (A30P) and the high copy yeast genomic library and induced with galactose were grown to 1.5 × 10 cells/ml and then plated on SGal−Leu−Trp. A disk of sterile Whatman filter paper containing 10 μl of 8% HO was placed in the center of each plate, and plates were incubated for 3 d. Colonies growing close to the peroxide disk were selected, cultured overnight, and the plasmid DNA was isolated and then amplified in DH5α cells. Plasmids were retested for protection in the hydrogen peroxide assay. Each strand of a protective plasmid was sequenced. was discovered by this approach. Westerns were conducted as described previously (). The monoclonal antibody against human α-syn was purchased from Cell Signaling Technologies. The monoclonal antibody against Pgk1p was purchased from Invitrogen. The monoclonal antibody against myc was purchased from Sigma-Aldrich. Secondary antibodies were purchased from Sigma-Aldrich or Bio-Rad Laboratories. 20 μg of protein were loaded per well in each Western blot conducted (, , and ). The coimmunoprecipitation was performed using protein A–Sepharose beads saturated with the anti-myc antibody as described in . Fluorescent images of cells were obtained with a microscope (AX70; Olympus), including an Olympus UPlanFl 100×/1.35 NA objective with CoolSNAP HQ CCD camera (Roper Scientific). The acquisition software was IPLab v3.6 (Scanalytics). An Olympus U-MWG (510–550) filter set was used for detecting GFP (, , and ), and a U-MNG filter (530–550 nm) was used to detect FM4-64, DHR 123, and FUN1 (, , , and ; and Fig. S2). For the two-color experiment ( and ), a more restrictive Chroma JP1 (510 nm) was used to detect GFP; this filter set prevented bleed-through of GFP signal into the FM4-64 channel. All data were collected at room temperature. The ROS assay, using the DHR 123 dye, was performed as described previously (). The viability assay, using the FUN1 dye, was conducted according to . Vacuole staining, using the FM4-64 dye, was conducted according to a method adapted from . In brief, S288c cells transformed with a GFP-α-syn plasmid (pTF305-308) and with a plasmid (pTF602) were pregrown in SSuc−Leu−Ura to mid-log phase. Cells were then washed and resuspended in SGal−Leu−Ura and induced for 12 h at 30°C. At various times aliquots were removed, washed, resuspended in YPD, and incubated with 40 μM FM4-64 (Invitrogen) for 10 min at 30°C. Cells were then washed twice, resuspended in YPD, incubated an additional 30 min at 30°C, and visualized by fluorescence microscopy. Adobe Photoshop 5.5 was used to prepare all figures. Adjustments in contrast were made to Westerns blots. Adjustments in brightness and contrast were made to some differential interference contrast (DIC) images and to some merged images (DIC merged with GFP fluorescence). Specific adjustments are given in the legends. TEM samples were prepared as described previously () and examined using an electron microscope (H-7000; Hitachi), equipped with a high-resolution digital camera (Gatan, Inc.). For immunogold labeling, cells were fixed with 4% formaldehyde and 0.25% glutaraldehyde in 40 mM phosphate buffer (pH 6.7), containing 1 mM MgCl and 1 mM EGTA for 1 h at room temperature. Cells were washed with phosphate buffer, incubated in 1% sodium metaperiodate for 15 min, then in 50 mM ammonium chloride for 15 min. Cells were dehydrated with graded ethanol, embedded in LR White resin (Electron Microscopy Sciences). 60-nm-thin sections were cut and collected on nickel grids. Grids were incubated in 0.1 M glycine in PBS containing 0.1% (vol/vol) Triton X-100 for 15 min to inactivate residual aldehydes. Sections were blocked with blocking buffer (PBS containing 5% normal goat serum, 0.2% Tween 20, and 0.2% BSA) for 40 min. Grids were then incubated in the primary rabbit anti-A30P antibody, diluted 1:50 in blocking buffer for 2 h. Grids were washed then incubated in 12-nm gold–conjugated goat anti–rabbit IgG (Jackson ImmunoResearch Laboratories) diluted 1:30 in blocking buffer for 1.5 h. Labeled sections were stained with 2% uranyl acetate for 30 min and lead citrate for 30 s, examined as described above. The control was performed by substituting primary antibody with blocking buffer. A Perkin Elmer Wallac Victor 1420 multilabel counter was used to screen deletion strains for ROS accumulation. Cells transformed with the pTF302 (A30P) and pTF602 () plasmids were pregrown in noninducing media until mid-log phase and then shifted into galactose inducing media and induced for 3 h. After the second hour of induction, the DHR 123 dye was added to a final concentration of 5 μg/ml. Before adding the cells to the 96-well plates, cells were washed twice using PBS and the assay was conducted in PBS. The excitation and emission wavelengths were 485 and 535 nm, respectively. At minimum, deletion strains were analyzed in two independent experiments in triplicate. Thirty of the deletion strains were analyzed in three to four independent experiments in triplicate. ROS readings from each deletion strain were compared with the ROS readings from the S288c strain (transformed with pTF302 [A30P] and pTF602 []). Acetic acid treatment has been shown to induce apoptosis and ROS in yeast cells (), and acetic acid–treated cells were prepared and incubated with DHR 123 as described previously () as a control. Deletion strains exhibiting greater than fourfold increases in ROS signal compared with the control (S288c transformed with pTF202 and pTF602) were selected for further evaluation. One two-plasmid control consisted of the various deletion strains expressing Ypp1p (pTF602 () plus pTF300 (empty vector)). The other two-plasmid controls consisted of the various deletion strains harboring plasmids with no inserts (pTF300 and pTF604). ROS accumulation was measured after 3 h of induction. Ypp1p overexpression did not appreciably increase the ROS signal, and the two-plasmid control cells yielded a signal indicative of zero fluorescence. For the ROS (and FUN1 viability assays) ( and , and Fig. S2), P values were determined using a two-tailed test (heteroscedastic) that compared red cell counts from various +α-syn/+Ypp1 cultures to red cell counts from various +α-syn/−Ypp1 cultures. Cells were counted in two to three independent experiments. The program Excel was used for this analysis. For the ROS plate reader assay (), a paired test was used to compare the ROS signals from deletion strains to matched controls (S288c +A30P/+). To adjust for multiple comparisons, the Bonferroni correction method was used, with a family-wise error rate of 0.05 or 0.001. For example, because 116 gene deletion strains were tested, the highest accepted P value was 4.3 × 10 (= 0.05/116) for a family error rate of 0.05. The statistical software used for this analysis was SAS v9.13 (SAS Institute, Inc.). Table S1 gives exact P values (available at ). Fig. S1 shows that in high copy permitted normal growth of cells expressing A30P, but not of cells expressing WT or A53T. Fig. S2 shows that in high copy fails to protect cells from hydrogen peroxide–induced ROS. Online supplemental material is available at .

In industrialized countries, osteoarthritis affects more than one third of the adult population. Despite their clinical importance, the molecular mechanisms of joint morphogenesis are still unclear. The appendicular skeleton arises from the condensation of chondroprogenitor cells that undergo chondrocyte template formation that is subsequently replaced by bone to form the adult skeletal elements separated by cartilaginous joints. The synovial joints of the long bone elements form through segmentation of the continuous cartilaginous template with loss at the sites of the developing joints, loss of differentiated chondrocytes, and emergence of a nonchondrocytic joint-forming cell population that undergoes condensation, flattens, and develops an interzone that then cavitates to form the joint space within the articular cartilage (for review see ). There is limited information on the mechanisms that regulate the complex multistep process that leads to joint interzone formation. In fact, very few genes have been reported to be necessary and/or sufficient to initiate the joint formation process (namely , [], and [previously known as ]; ; ; ). The factors that induce the expression of these joint morphogenic molecules are undefined; furthermore, the mechanisms that determine the emergence of joint interzone cells within the chondrogenic condensates are unclear. TGF-βs elicit their signal binding to TGF-β type II receptor (TβRII) that leads to the phosphorylation of TβRI and TβRII–TβRI complex formation, which then activates the signaling cascade through R-Smad–dependent (Smad-2,-3,-4) and Smad-independent pathways. In human and mouse embryonic cartilage, TGF-βs are expressed in the endochondral template with high expression in the perichondrium (; ,; ; ). TβRI and TβRII have been reported to be expressed in the perichondrium and proliferative and differentiated chondrocytes (). Genetic manipulation of the TGF-β system genes have revealed their critical but still undefined roles in skeletogenesis (; ; ; for review see ). Targeted germline deletion of the gene in mice results in perinatal lethality, and mice present several skeletal defects, including cleft palate, skull ossification defects, shortened long bones, bifurcation of the sternum, and spina bifida occulta (). 50% of the ablated mice die early in utero before embryonic day (E) 10.5 because of a preimplantation defect; mice that are born do not display any dysmorphic phenotype but die early from diffuse inflammation (; for review see ). -null mice have cleft palate and abnormal lungs (). Variability in the skeletal phenotype in -targeted disrupted mice can be the result of differential and overlapping expression patterns of the isoforms throughout the skeletogenesis process and, therefore, compensatory effects of the other isoforms when one is ablated. Similarly, gene targeting in mice has lead to variable phenotypes from early lethality ( and ablation) to normal phenotype at birth but progressive osteoarthritis and colon adenocarcinomas in adulthood ( ablation; ; ; ; ). TβRII is the only TβR that is capable of binding all of the TGF-β isoforms and eliciting functional signaling; therefore, its ablation will allow studies of TGF-β signaling that avoid the functional redundancy of the ligands and signaling pathways. Unfortunately, mice that are germline null for exhibit early embryonic lethality that makes it impossible to evaluate the role of TGF-β signaling in skeletogenesis (). We have previously reported that in transgenic mice, overexpression of a dominant-negative (DNIIR) results in adult osteoarthritis (). However, the phenotype was only observed in a few lines, most likely because expression was inconsistent and lacked tissue-specific targeting (). Furthermore, the TGF-β cell targets and the temporal window of essential function during the endochondral process are not well defined. Conditional inactivation of in differentiated chondrocytes results in mice without any long bone defects, leading to the conclusion that TGF-β signaling is not needed in the limb endochondral process (). However, implanted TGF-β induces extra digit formation (). To circumvent the embryonic lethality of systemic ablation and to determine the role of TGF-β signaling in early limb bud development, we generated mice in which the TβRII signaling is conditionally inactivated in limb buds and in a subset of other mesenchyme tissues starting at E9.5 (). We show that in Tgfbr2PRX-1KO mice, TGF-β signaling ablation results in the following: lack of interphalangeal joint development; failure of joint interzone formation with a lack of Jagged-1 expression and aberrant survival of differentiated chondrocytes that leads to the absence of segmentation within the chondrogenic condensates; failure of joint morphogenic marker expression, including Noggin, and increased bone morphogenic protein (BMP) activity in limb bud cultures; a selective defect on the endochondral growth plate process adjacent to the interphalangeal joints at early and late chondrogenesis with an increase of prehypertrophic chondrocyte markers and a decrease of terminally differentiated chondrocyte marker expression; and midline defects and zeugopod and stylopod chondrodysplasia. Furthermore, using a TβRII reporting mouse and in situ and immunohistochemistry analyses, we have demonstrated that TβRII is highly and specifically expressed in developing joints. To conditionally inactivate the TGF-β signaling in limb buds, we crossed homozygous females with (Cre); double heterozygous males (CreTgfbr2) to generate mice (homozygous knockouts). Newborn mice showed abnormal forelimbs and hindlimbs (), which were confirmed by microcomputed tomography (micro-CT) imaging and Alizarin red/Alcian blue staining (). autopods lacked interphalangeal joint development, and, between the ossification centers of the phalanges, at the site where the joints should have been formed, there was a continuous pattern of cells with some bending, and a distal interphalangeal joint was seen sporadically (). Forelimb and hindlimb interphalangeal joints were equally affected, and studies were performed either on forelimbs or hindlimbs, but mostly on both. The forelimb and hindlimb autopods displayed smaller ossification centers of the metacarpals, carpals, metatarsals, tarsals, and phalanges as compared with controls; phalanges were bent (clinodactyly; ). zeugopods and stylopods were short with signs of chondrodysplasia (). The humerus lacked the deltoid tuberosity and, similar to the femur, had dysplastic widened, flaring, and poorly mineralized metaphyses (). The morphometric parameters of newborn mutants are summarized in . Compared with control Cre siblings, mutants are shorter, and their length is more affected than their weight, as demonstrated by the higher ponderal index. This finding indicates that in mutants, the skeletal growth is impaired by a primary defect on skeletogenesis and is not the result of a global intrauterine nutritional defect. mice had several midline defects: they lacked sternum formation, had hypoplastic incisors, and lacked the parietal and interparietal bones, whereas the frontal and squamosal bones were reduced in size (). They were capable of suckling, but they experienced massive and progressively visible intracranial bleeding (although still alive) that, at the necropsy exam, occupied most of the brain and likely was the primary cause of death. Although it is possible that a respiratory insufficiency caused by the lack of the sternum may be a concomitant cause of death, it is unlikely to be the primary cause considering the severity of the intracranial bleeding. Lack of parietal and interparietal bone development was confirmed by micro-CT analyses of living newborn mice, indicating that loss was not caused by accidental removal of the vault during the Alizarin red/Alcian blue staining procedure (). The pelvic bones of mice were smaller and poorly mineralized with signs of chondrodyplasia (). Because we observed skeletal defects in segments unexpected for the reported Prx-1–mediated Cre recombination, we decided to evaluate the Prx1-Cre expression pattern in mice by crossing females doubly homozygous for and R26R loci (-R26R) with males heterozygous for to generate -R26R mice. In the R26R mice, the ROSA26 locus is targeted by gene trapping so that Cre recombination results in expression (). We found that in -R26R whole mount embryos (E10.5), X-galactosidase staining was evident in the developing forelimbs and hindlimbs as well as in the skull and in the anterior midline region of the trunk (). In sections of E15.5 -R26R embryos, X-galactosidase staining was visualized in the skull, limbs, and in the oral, midline, and pelvic regions, which are areas where the newborn mutants showed substantial skeletal abnormalities (). Because the autopods lacked the interphalangeal joints, we decided to investigate the TβRII expression in developing joints. To this purpose, we modified bacterial artificial chromosomes (BACs) to generate a mouse reporter transgene containing both GFP and IRES-β-GEO (LacZ/Neo) reporter genes. We found that in E12.5 (, arrows) and 16.5 () whole mount and E16.5 phalangeal sections () of embryos, is highly expressed in the interphalangeal joints. We have also noted that is highly expressed in the shoulder and elbow joints () as well as in the knee and hip joints (not depicted). We have established five independent transgenic lines that demonstrate joint expression. expression was similar in hindlimb and forelimb interphalangeal joints (unpublished data). The joint expression pattern in mice was directly comparable with endogenous expression. In fact, in situ hybridization studies revealed that in E16.5 embryos, is highly expressed in the cells demarking the interphalangeal joint interzone and in the phalangeal prehypertrophic chondrocytes (, middle). Immunofluorescence studies confirmed TβRII joint expression (, top). Furthermore, there was an intense staining of phosphorylated Smad-2 in the interzone cell nuclei (, bottom). In mutants, the lack of joints was accompanied by the lack of mRNA and protein expression as well as a decrease of cell nuclei positive for phosphorylated Smad-2, indicating the effective Prx-1–mediated Cre recombination of (). Regarding the expression in the growth plate adjacent to the joints, we found that it was expressed at a much lower level than joints by cells that morphologically resemble prehypertrophic chondrocytes (). PCR amplification of genomic DNA extracted by laser capture microdissection (LCM) from paraffin sections of E16.5 joint cells and cells outlining the joint mesoderm demonstrated a specific recombination and subsequent loss of the floxed alleles ( exon 2) in joints compared with (Fig. S1, available at ). Furthermore, quantitative real-time PCR of genomic DNA extracted from and forelimb- and hindlimb-dissected digit bones and interphalangeal joints after removal of the skin and surrounding tissues showed that the efficiency of deletion of the exon 2 was 92 ± 3.0% ( = 3 mice for each group); considering the heterogeneity of the sample, efficiency is considerable. Initiation of the joint interzone is demarked by segmentation of the cartilaginous continuity across the future joint location (for review see ). In mutants at E16.5, we found a persistence of –expressing chondrocytes along the whole digit, including the potential joint site, whereas in control animals at the same age, expression was confined to the endochondral templates and absent in the fully demarked joint (). Several components of the Notch system are expressed in articular cartilage, and a Notch-1–positive population of progenitor joint cells has been recently isolated from articular cartilage. This leads to the hypothesis that Notch signaling within the articular cartilage blocks chondrocyte differentiation, maintaining clonality and proliferation of the progenitor joint-forming cells (; ). In mice, interzone develops at E12.5–13.5. We found that E13.5 mutants failed to form the interzone and lacked Jagged-1 expression, whereas in control mice, interzone cells highly expressed Jagged-1 (). It has been postulated that apoptosis may play a role in determining the fate of differentiated chondrocytes within the developing joint (for review see ). An intense positive TUNEL staining for apoptotic nuclei was observed in E13.5 control forming joints, whereas E13.5 mutants lacked cell apoptosis within the presumptive joint region (). The activation of transcription is critical for joint formation, although its regulation is unknown. Mice that are null mutants for lack joints, and heterozygous loss of function mutations are found in some of the patients with proximal symphalangism (SYM1-OMIM185800) that lack proximal and medial interphalangeal joints, whereas the distal interphalangeal joint is never affected (; ; ). Analysis of expression at E13.5 and 16.5 by in situ hybridization and immunofluorescence revealed a complete down-regulation in the joints of embryos (). is one of the earliest markers expressed in developing joints, and is mutated in the mouse, which has interphalangeal joint defects (). Furthermore, a mutation with a gain of aberrant BMP-2–like function was reported in a family with SYM1 (). It has been hypothesized that in early chondrogenesis, inhibits joint formation and induces cartilage development, whereas its role in late chondrogenesis is to maintain joint formation (). In mutants, expression was increased at E13.5, whereas it was abrogated in E16.5 (). is expressed in developing joints, and its misexpression in chicken digit rays induces ectopic joint formation, whereas the loss of in mice results in synovial chondromatosis (; ). In mutant joints, expression is down-regulated in E13.5 (not depicted) and 16.5 (). These results indicate that TGF-β signaling in developing limbs is mandatory for joint formation and to regulate , , and expressions. It is difficult to infer from the results found in mutants whether TGF-β signaling directly regulates transcription or whether TGF-β sustains the limb bud growth, ensuring an adequate environment for the joints to develop and to be expressed. Therefore, we decided to evaluate the role of TGF-β signaling in expression in limb bud micromass cultures. In cultures, expression was conditionally inactivated, and the lack of TβRII binding expression was verified by a I–TGF-β1 affinity cross-linking cell surface–binding assay (Fig. S2 A, available at ). cultures, the TβRI, TβRII, and TβRIII were identified, whereas in cultures, I–TGF-β1 binding to TβRII was greatly reduced (Fig. S2 A). The specificity of I–TGF-β1 binding was confirmed by the fact that labeled bands were displaced by cold TGF-β1 in excess (Fig. S2 A). that was negative in control cultures (Fig. S2 B). cultures, TGF-β treatment induced Noggin mRNA and protein expression as determined by quantitative RT-PCR and Western immunoblotting (WIB) analyses (). Similar results were found when wild-type micromass cultures were treated with TGF-β (7.8 ± 2.1-fold compared with untreated control [1.2 ± 0.3-fold]; P < 0.05; = 3). Noggin binds to BMPs, preventing BMP receptor activation and, therefore, signaling; the canonical BMP signal is through the phosphorylation cascade of Smad-1, -5, and -8 that complex and induce transcription. cultures, TGF-β decreased BMP activity, as indicated by a decrease of phosphorylated Smad-1, -5, and -8 (). cultures resulted in the abrogation of TGF-β effects on Noggin expression and Smad-1, -5, and -8 phosphorylation (). compared with cultures, possibly as a result of the unresponsiveness of cells to endogenous TGF-β. It has been hypothesized that the developing joints act as signaling centers to control the adjacent endochondral template development (for review see ). Because the mutant lacks the interphalangeal joints, it represents an ideal model to test this hypothesis. Therefore, we performed a systematic evaluation of cartilage marker expressions in growth plates adjacent to the presumptive interphalangeal joints at E13.5, E16.5, and postnatal day (P) 0 in mutants and control siblings (). We found that growth plates presented several remarkable and selective abnormalities; expression is consistently decreased at E13.5, E16.5, and P0 compared with Tgfbr2 controls, indicating a dramatic delay in chondrocyte hypertrophy in the mutants (). Conversely, in growth plates, () was increased and more widely expressed from the proliferative zone to the canonical prehypertrophic chondrocyte zone compared with Tgfbr2 controls; this finding was consistent at E13.5, E16.5, and P0 (). In mutants at E16.5 and P0, parathyroid hormone–related protein () expression is increased and more diffuse in the prehypertrophic/upper proliferative zone and in the perichondrium; at E13.5, expression is similar to the control (). and expressions are similar to the control in , but –expressing cells at P0 display a less organized columnar distribution than controls (). This disorganization was also observed in the hematoxylin and eosin staining that also showed that hypertrophic chondrocytes are larger but show a decreased expression of (). italic #text To generate mutants, female homozygous mice were mated with and heterozygous males (; ). As previously reported, we have generated the mouse by flanking with loxP sites the exon 2 of TβRII that transcribes for the TGF-β–binding domain (). In the mouse (a gift from C. Tabin, Harvard Medical School, Boston, MA), a limb enhancer drives Cre recombinase expression in limb buds and in calvaria mesenchyme beginning at E9.5 (). Genotyping was performed using PCR primers for and (). Because a penetrant germline recombination has been reported when crossing females with mice carrying floxed genes, only Cre males were used (). CreTgfbr2 males (Swiss-Webster background) were backcrossed propagated to C57BL/6 females, and experiments were performed in mice that were in the C57BL/6 strain for at least eight generations. To generate the mouse, the mouse BAC clone RP24-317C21 containing was obtained from the Children's Hospital Oakland Research Institute. As schematically presented in Fig. S3 (available at ), a GFP-IRES-β-GEO cassette was inserted into at the endogenous translational start site using the homologous recombination technique of , , and as previously reported (; ) as follows: the plasmid pIBGFTet was generated by ligating the IRES-β-GEO-SV40pA cassette from pGT1.8 and an FRT-flanked tetracycline resistance cassette into a modified pBluescript II SK+ backbone (). An eGFP open reading frame (CLONTECH Laboratories, Inc.) was then inserted upstream of IRES-β-GEO-SV40pA to create pGFPIBGFTet. The recombination cassette was constructed by subcloning 50-bp 5′ and 3′ recombination arms (both containing part of the exon 1) into pGFPIBGFTet such that the recombination arms flanked the GFP-IRES-β-GEO-FRT-Tet-FRT cassette. The forward strand (relative to ) sequences of the 50-bp homology arms were as follows: for the 5′ arm, CGGTTCGTGGCGCACCAGGGGCCGGTCTATGACGAGCGACGGGGGCTGCC; and for the 3′ arm, ATGGGTCGGGGGCTGCTCCGGGGCCTGTGGCCGCTGCATATCGTCCTGTG. Both recombination arms were created by annealing PAGE-purified oligonucleotides designed to allow direct ligation to pGFPIBGFTet. The final cassette with recombination arms was digested from the vector, gel purified, and recombined with BAC as described previously (). Successful recombinants were selected by tetracycline resistance. The tetracycline resistance gene was then removed by FLPe recombinase excision (; ). The correctly modified BAC was verified by conventional and pulsed-field gel analysis of restriction digests to confirm expected banding patterns as well as direct BAC sequencing. -BAC DNA was purified according to established techniques and was used for pronuclear injection of C57BL/6J × DBA/2J F1 hybrid embryos (). Injections and oviduct transfers were performed by the Vanderbilt Transgenic Core Facility using standard techniques in accordance with protocols approved by the Vanderbilt University Institutional Animal Care and Use Committee. All BACs were injected as uncut circular DNAs. To generate the - mouse, the and (obtained from P. Soriano, Fred Hutchinson Cancer Research Center, Seattle, WA; ) were first crossed to obtain the - female mice that were then crossed with heterozygous males. For timed pregnancies, noon of the day when evidence of a vaginal plug was found was considered E0.5. Alizarin red/Alcian blue staining was performed as previously reported (). Images were taken using a stereo microscope (SZX16; Olympus) equipped with a digital camera (DP71; Olympus) and imported into Photoshop (Adobe). Living animal micro-CT imaging (Imtek MicroCAT-II-CT) was performed setting micro-CT slices at 40 μm that were then reconstructed in 3D arrays using the same thresholds. In situ hybridization studies were performed as previously reported (). Digoxigenin-UTP-riboprobes were synthesized (DIG-RNA-Labeling kit; Roche) from plasmids with insertion of (provided by R. Harland, University of California, Berkeley, Berkeley, CA), (provided by S.K. Dey, Vanderbilt University, Nashville, TN), mouse and (provided by D. Kingsley, Stanford University, Palo Alto, CA), mouse (provided by A. McMahon, Harvard University, Cambridge, MA), and (provided by H. Kronenberg, Harvard University; ; ; ). was made using a mouse cDNA clone (IMAGE clone #30435371; GenBank/EMBL/DDBJ accession no. ); the plasmid was linearized with XmnI and riboprobe synthesized with T7 polymerase. and probes were also made. The primers for were forward (GACATGTAAAGGAAGGTAACGATTG) and reverse (AGG CTAAGGGACACTCTTGAACTA); the primers for were forward (GCCAGGTCTCAATGGTCCTA) and reverse (GATCCAGGTAGCCTTTGCTG). PCR products were cloned into pGEMT-Easy and linearized with NcoI, and riboprobes were synthesized with T7 polymerase. Whole mount embryo X-galactosidase staining was performed as previously reported (). Images were taken using an inverted microscope (1X71; Olympus) equipped with a digital camera (DP71; Olympus) and were imported into Photoshop (Adobe), where they were formatted without using any imaging enhancement. For cryosectioning, whole mount stained embryos were cryoembedded in optimal cutting temperature compound (Sakura). 50-μm sections were warm adhered on Superfrost-Plus slides (Fisher Scientific), washed thoroughly, and mounted using Aqua-Polymount (Polysciences). Section images were taken using the IX71 microscope with DP71 digital camera. Sections subjected to immunofluorescence were imaged using a microscope (Axiophot; Carl Zeiss MicroImaging, Inc.) with a camera (Micromax; Princeton Instruments), and images were imported into Photoshop for formatting. Limb bud mesenchymal cells from E11.5 embryos were isolated and micromass cultured as described previously (). To conditionally inactivate TβRII, micromasses from or embryos were infected either with HR-MMPCreGFP or MMP-GFP retroviral vectors as previously reported to generate CreTgfbr2 micromasses or MMPTgfbr2 micromasses, respectively (; ). Infections were performed 1, 24, and 48 h after seeding. I–TGF-β1–TßR affinity cross-linking was performed as previously reported (). Cell lysates and total RNA were obtained as previously reported from CreTgfbr2 micromasses or MMPTgfbr2 micromasses cultured for 36 h and were treated with or without 20 ng/ml TGF-β (). Treatment was repeated after 24 h, and cells were harvested 12 h later (total TGF-β treatment time was 36 h); WIB analysis was performed as previously reported (). Noggin and phospho-Smad-1/-5/-8 polyclonal antibodies were obtained from Cell Signaling, and anti–β-actin antibody was purchased from Sigma-Aldrich. WIB images were semiautomatically analyzed using a custom-built densitometric image analysis code in MATLAB (Mathworks). Quantitative RT-PCR was performed as previously described (). PCR primers for amplification were 5′-AAGGAGAAGGATCTGAACGAGACG-3′ and 5′-TCGGAGAACTCCAGCCCTTTGAT-3′. LCM was performed as previously described (). In brief, E16.5 autopod paraffin sections (5 μm on uncharged slides) were immediately subjected to LCM on an LCM system (PixCell Iie; Arcturus). The captured cells were subsequently extracted for DNA amplification to determine the recombination of Tgfbr2 exon 2 by PRC as previously described (). Genomic DNA was also obtained from E17.5 Tgfbr2 and Tgfbr2 forelimb- and hindlimb-dissected digit bones and interphalangeal joints after removal of the skin and surrounding tissues and was subjected to quantitative real-time PCR using previously reported primers (; ). Data are presented as mean ± SD and are analyzed using an unpaired test or one-way analysis of variance (Sigmastat Software; Sigma-Aldrich). Statistical significance was set at P < 0.05. Fig. S1 shows recombination of the TβRII exon 2 in the joints by LCM followed by PCR. Fig. S2 shows the recombination of Tgfr2 in CreTgfbr2 and in CreTgfbr2-R26R micromass cultures. Fig. S3 shows a schematic diagram of the β− reporter construct. Online supplemental material is available at .

Tissue scarring, characterized by cell activation, excessive deposition of ECM, and extravascular fibrin deposition, is considered a limiting factor for tissue repair. Fibrin, the major substrate of the serine protease plasmin, is a provisional matrix deposited after vascular injury (). The two plasminogen activators (PAs), namely tissue plasminogen activator (tPA) and urokinase plasminogen activator (uPA) and their inhibitors, such as plasminogen activator inhibitor-1 (PAI-1), are key modulators of scar resolution by spatially and temporally regulating the conversion of plasminogen to plasmin resulting in fibrin degradation and ECM remodeling (). In the peripheral nervous system, previous work by us and others showed that inhibition of fibrinolysis in mice deficient in plasminogen or tPA exacerbated axonal damage () and impaired functional recovery after nerve injury (). In accordance, mice deficient for fibrinogen showed increased regenerative capacity (). Studies of fibrin deposition in human diseases, in combination with experiments from mice deficient in plasminogen and PAs, have provided information about a wide range of physiological and pathological conditions that are exacerbated by defective fibrin degradation, such as wound healing, metastasis, atherosclerosis, lung ischemia, rheumatoid arthritis, muscle regeneration, and multiple sclerosis (MS) (; ). However, the molecular mechanisms that regulate proteolytic activity remain unclear. In our current work, we focus on the mechanisms that regulate fibrinolysis after injury. Our previous studies demonstrated a correlation between fibrin deposition and expression of p75 neurotrophin receptor (p75) after nerve injury (). Up-regulation of p75 is observed in MS (), stroke (), and spinal cord () and sciatic nerve injury (), all of which are associated with fibrin deposition. p75 is also expressed in non-neuronal tissues () and is up-regulated in non-nervous system diseases associated with defects in fibrin degradation, such as atherosclerosis (), melanoma formation (), lung inflammation (), and liver disease (). p75 has been primarily characterized as a modulator of cell death () and differentiation () in non-neuronal tissues. The expression of p75 by cell types such as smooth muscle cells and hepatic stellate cells, which actively participate in tissue repair by migration, and secretion of ECM and extracellular proteases, raises the possibility for a functional role of p75 in disease pathogenesis that extends beyond apoptosis and differentiation. We find that p75 is involved in the regulation of proteolytic activity and fibrin degradation. Mice deficient for p75 () show increased proteolytic activity and decreased fibrin deposition in two disease models: sciatic nerve injury and lung fibrosis. p75 regulates proteolytic activity by simultaneously down-regulating tPA and up-regulating PAI-1 via a novel cAMP/PKA pathway. p75 decreases cAMP via interaction with the cAMP-specific phosphodiesterase (PDE) isoform PDE4A4/5. This is of particular note, as selective PDE4 inhibitors have an anti-inflammatory action and have potential therapeutic utility in inflammatory lung disease, as well as in a wide range of neurologic diseases such as depression, spinal cord injury, MS, and stroke (; ; ). Overall, the regulation of plasminogen activation by p75 identifies a novel pathogenic mechanism whereby p75 interacts with PDE4A4/5 to degrade cAMP and thus perpetuates scar formation that could possibly render the environment hostile for tissue repair. mice after injury. In wt mice, there is a dramatic increase of fibrin deposition () and p75 expression () after injury, when compared with uninjured nerves (). mice show reduced fibrin deposition after injury (). mice have decreased fibrin by threefold 3 d and fourfold 8 d after injury (). mice have significantly decreased fibrin (, P < 0.003). These results suggest that loss of p75 decreases the levels of fibrin in the sciatic nerve after injury. mice (unpublished data), suggesting the decrease in fibrin deposition is not the result of hypofibrinogenemia. mice reflects an up-regulation of the proteolytic activity. mice have increased proteolytic activity () when compared with wt mice () that is statistically significant (, P < 0.05). Uninjured nerves exhibit minimal proteolytic activity (), as expected (). sciatic nerves do not show lysis of fibrin in the absence of plasminogen (), suggesting that the proteolytic activity is plasminogen dependent. The tPA/plasmin system regulates fibrin clearance after nerve injury (). mice (). p75 is strongly activated by withdrawal of axons () and its expression correlates with proliferating, non-myelin producing Schwann cells (SCs) (). After sciatic nerve injury both p75 (, red) and tPA (, green) increase when compared with uninjured controls (), but show little colocalization (), suggesting that p75-reexpressing SCs do not express tPA. Expression of tPA (, red) and p75 (, red) in SCs is confirmed using double immunofluorescence with the SC marker S100 (; green). mice was due to tPA, we crossed mice with mice and generated double-knockout mice. mice show a decrease in fibrin deposition () and an increase in proteolytic activity (), compared with wt control mice (, respectively). mice show increased fibrin deposition () when compared with mice () and no evidence of proteolytic activity (). mice also show no evidence of proteolytic activity after sciatic nerve crush injury (), as described previously (). Quantification of proteolytic activity is shown in . mice ( mice, ) are in accordance with the pharmacologic inhibition of tPA activity in the sciatic nerve using tPASTOP (). mice is due to up-regulation of tPA. mice and cultured them on a three-dimensional (3D) fibrin gel. Wt SCs, which express high levels of p75, form a monolayer on the fibrin gel (). SCs degrade the fibrin gel () and show a 2.7-fold increase of fibrin degradation (). SCs show a sixfold increase in tPA levels, when compared with wt SCs (; P < 0.01). These results suggest that p75 down-regulates tPA activity and blocks fibrin degradation in SCs in vitro. mice, we used stable and transient transfections of p75 as well as siRNA against to test the properties of p75 in heterologous systems. To examine whether p75 could inhibit fibrin degradation, we first used NIH3T3 fibroblasts stably transfected with p75 that exhibit high levels of p75 (10 receptors/cell) (). NIH3T3 cells on a 3D fibrin gel degrade fibrin (), whereas NIH3T3p75 cells do not (). Expression of p75 inhibits fibrin degradation by 12-fold (; P < 0.001). NIH3T3 cells form lytic areas (), whereas NIH3T3p75 cells grow uniformly on fibrin (). NIH3T3 cells fully degrade the plasmin substrate casein () but NIH3T3p75 cells do not degrade casein (), suggesting impaired proteolysis in NIH3T3p75 cells. Aprotinin, a general inhibitor of serine proteases, completely inhibits fibrin degradation by NIH3T3 cells (not depicted). In fibroblasts both tPA and uPA are involved in activation of plasminogen and fibrin degradation. tPA activity is significantly decreased in the NIH3T3p75 cells (). In contrast, expression of p75 has no effect on uPA activity (). is a transcriptionally regulated immediate-early gene (). Indeed, expression of p75 down-regulates tPA transcripts (). In addition, mRNA of PAI-1 is also up-regulated in NIH3T3p75 cells (). Real-time quantitative PCR shows a 10.1-fold decrease in tPA mRNA, a fourfold increase in PAI-1 mRNA, and a twofold decrease in uPA mRNA in NIH3T3p75 cells. Upon expression of p75, the decrease of uPA RNA does not affect uPA activity (). In contrast, the decrease of tPA RNA in NIH3T3p75 cells results in a corresponding decrease in tPA activity (; P < 0.01). mice show a fourfold increase of RNA when compared with wt (). mice show an increase in tPA RNA in primary cerebellar granule neurons (CGNs) (Fig. S1 c, available at ), and increased proteolytic activity in the cerebellum (Fig. S1, a and b). Overall, these data suggest that expression of p75 inhibits the tPA/plasmin system both in vivo in the cerebellum and after sciatic nerve injury, as well as in vitro in primary neurons, SCs, as well as fibroblasts. Transcriptional regulation of tPA depends on the cAMP/PKA pathway (). Indeed, elevation of cAMP, using dibutyryl-cAMP (db-cAMP), overcomes the inhibitory effect of p75 (). Moreover, cAMP elevation, elicited using the general PDE inhibitor IBMX, elevates tPA activity in NIH3T3p75 to the levels seen in NIH3T3 cells (). IBMX does not affect basal levels of tPA in NIH3T3 cells (). These data suggest that PDE activity is required for the p75-induced tPA decrease. PKA activity is decreased in NIH3T3p75 cells (, lanes 3 and 4) compared with NIH3T3 cells (, lanes 1 and 2), suggesting that p75 expression reduces PKA activity. KT5720, a specific PKA inhibitor, decreases tPA activity in NIH3T3 cells (). Because the cAMP/PKA pathway enhances tPA transcription and suppresses PAI-1 secretion (), we tested whether the cAMP/PKA pathway influences the p75 regulation of tPA and PAI-1. Forskolin-induced cAMP elevation increases, whereas KT5720-induced PKA inhibition decreases tPA RNA in NIH3T3 cells (). Forskolin treatment of NIH3T3p75 cells also increases both tPA RNA () and activity (not depicted), whereas forskolin decreases PAI-1 RNA in both NIH3T3 and NIH3T3p75 cells (). SCs (). Brain-derived neurotrophic factor (BDNF)/TrkB signaling has been shown to regulate tPA in primary cortical neurons (). In contrast to cortical neurons, SCs are known to express minute levels of TrkB but high levels of p75 (). We show here that treatment of SCs with either BDNF or nerve growth factor (NGF) has no effect on tPA (). Similar results are obtained after treatment of SCs with pro-NGF, the high-affinity ligand of p75 () (unpublished data). In addition, in NIH3T3 and NIH3T3p75 cells, which do not express Trk receptors, the p75-mediated suppression of tPA activity occurs independent of neurotrophins or serum (unpublished data). In accordance, in NIH3T3 cells transient expression of the intracellular domain (ICD) of p75 decreases tPA similar to the full-length (FL) p75 (). These data suggest that neurotrophin/p75 signaling is not involved in the regulation of tPA in SCs and fibroblasts and that regulation of tPA by p75 is independent of neurotrophins. Because the effects of p75 were overcome by elevating cAMP, we examined whether p75 reduced cAMP levels. Indeed, cAMP is decreased 7.8-fold in NIH3T3p75 cells (; P < 0.0001). Transient expression of p75 in NIH3T3 cells decreases levels of cAMP (; P < 0.0005).Furthermore, siRNA knockdown against leads to increased cAMP levels in both NIH3T3p75 cells (; P < 0.02) and primary SCs (; P < 0.03). NIH3T3 cells transiently transfected with p75 express fivefold less p75 than the stably transfected NIH3T3p75 cells (unpublished data). Differences in expression between stably and transiently transfected cells may account for the differences in the fold-decrease of cAMP and tPA between these two systems. mice (). In neurons BDNF elevates cAMP exclusively via TrkB (). In NIH3T3p75 cells, which do not express TrkB, stimulation with NGF or BDNF does not affect the p75-mediated suppression of cAMP (Fig. S2). Similarly, inhibition of neurotrophins by Fc-p75 or BDNF by Fc-TrkB does not alter cAMP levels in NIH3T3p75 cells (Fig. S2, available at ). In accordance, transient expression of the ICD of p75 decreases cAMP similar to the FL p75 in NIH3T3 cells (). Overall, these data suggest a neurotrophin-independent PDE4/cAMP pathway downstream of p75, which consequently leads to decreases in extracellular proteolysis. Down-regulation of cAMP can be mediated either by inhibition of cAMP synthesis via the action of G, a G protein that inhibits adenylyl cyclase or via the action of PDEs. Treatment of cells with pertussis toxin (PTX) that blocks interactions between the G and G protein coupled receptors, does not rescue the p75-mediated down-regulation of cAMP (; P > 0.5). In contrast, the PDE inhibitor IBMX resulted in significant increase of cAMP in the NIH3T3p75 cells when compared with control NIH3T3p75 cells (; P < 0.000001). Use of specific chemical inhibitors for PDE isoforms shows that only rolipram, a specific inhibitor of PDE4, significantly increases cAMP levels in NIH3T3p75 cells (; P < 0.000001) to the levels of NIH3T3 cells (; P = 0.051), suggesting that the p75-induced cAMP decrease is mediated via PDE4. Recruitment of PDE4 to subcellular structures such as the plasma membrane concentrates the activity of PDEs and reduces PKA activity by enhancing degradation of cAMP (; ). We therefore examined whether p75 regulates cAMP via recruitment of PDE4. In NIH3T3p75 cells, p75 coimmunoprecipitates (co-IPs) with endogenous PDE4A (). No association is observed with the other three PDE4 sub-families, namely PDE4B, PDE4C, or PDE4D (unpublished data), suggesting that the effect was PDE4A specific. Based on the molecular weight of PDE4A at 109 kD, we determined that p75 co-IPs with the PDE4A5 isoform. Endogenous co-IP in CGNs (Fig. S3 a, available at ) and in injured sciatic nerve (Fig S3 b) shows that p75 and PDE4A5 interact at endogenous expression levels. Analysis of lysates shows that the levels of PDE4A are similar in NIH3T3 and NIH3T3p75 cells (Fig. S3 c). These results show that p75 forms a complex with PDE4A5. A functional consequence of the p75–PDE4A5 interaction would be recruitment of PDE4A5 to the membrane resulting in decreased membrane-associated cAMP/PKA signaling. To investigate whether p75 reduces membrane-associated PKA activity, we modified the genetically encoded A-kinase activity reporter, AKAR2 () and generated pm-AKAR2.2, a membrane-targeted fluorescent reporter of PKA activity that generates a change in fluorescence resonance energy transfer (FRET) when it is phosphorylated by PKA in living cells (Fig. S4 a). As expected, NIH3T3 cells show a dramatic emission ratio change for the pm-AKAR2.2 in response to forskolin (). In contrast, NIH3T3p75 cells show an attenuated response, revealing reduced PKA activity at the plasma membrane (). Transient transfection of p75 confirmed the results observed in the stable NIH3T3p75 cells using the latest generation of plasma membrane–specific PKA biosensor AKAR3 () (Fig. S4 b, available at ). As expected, increased cAMP degradation at the plasma membrane results in decreased intracellular cAMP (Fig. S4 c; ). Overall, our results showing reduced membrane-associated PKA activity upon expression of p75 suggest that p75 targets cAMP degradation to the membrane via its interaction with PDE4A5. To verify the specificity of p75–PDE4A5 association, a series of mapping studies were conducted using deletion mutants. PDE4A5 interacts with FL p75, as well as deletions Δ3, Δ62, Δ83, but not a deletion missing the distal 151 amino acids, Δ151 (), suggesting that the interaction between p75 and PDE4A5 occurs in the juxtamembrane region of p75, requiring sequences between residues 275 and 343. To explain the specificity of the interaction of p75 with a single PDE4 isoform, we reasoned that p75 would interact with a unique region of PDE4A5 that is not present in other PDE4s. Although the PDE4 isoforms are highly homologous, PDE4A5 contains a unique C-terminal region with a yet unknown biological function (). Co-IP experiments in HEK293 cells using the PDE4A4CT mutant that is missing the C-terminal region (aa 721–886) abolishes the interaction with p75 (). To examine whether p75 could interact with PDE4A5 in a direct manner, we performed in vitro pull-down assays using recombinant proteins. A GST fusion protein of p75 encoding the entire ICD interacts with both recombinant PDE4A5 and its human homologue PDE4A4 (). In contrast, p75 ICD does not interact with recombinant PDE4D3 (). These results are in accordance with both the endogenous co-IPs in cells (; Fig. S3) and the PDE4A4 mutagenesis data () because similar to PDE4A4CT, PDE4D3 does not contain the unique C-terminal domain of PDE4A4/5. We have used peptide array technology to define sites of direct interaction in other PDE4s (). Screening a peptide array library of overlapping 25-mer peptides that scanned the sequence of PDE4A4 with GST-ICD p75 identified interactions with the LR1 domain, whose sequence is unique to the PDE4A subfamily (peptides 40 and 41, aa 191–220), and also to a sequence within the catalytic domain (peptides 135 and 136, aa 671–700). However, the strongest interaction was observed with sequences within the C-terminal region of PDE4A4 (peptides 172 and 173, aa 856–885). Alanine scanning mutagenesis shows that substitution of C862 abolishes the interaction of p75 with the 173 peptide that is unique to PDE4A (). The p75-interacting sequences within the LR1 and C-terminal domains are highly conserved between the human PDE4A4 and the rodent PDE4A5. Indeed, peptide array screening for PDE4A5 reveals direct interaction with p75 similar to that seen for PDE4A4 (unpublished data). Overall, these results suggest that the interaction of p75 with PDE4A4/5 is direct and that sequences within the juxtamembrane region of p75 and the unique C-terminal region of PDE4A4/5 are primarily required for the interaction (; Fig. S3). Because expression of p75 inhibits fibrinolysis in fibroblasts, we hypothesized that the role of p75 as a modulator of fibrinolysis extends to tissues outside of the nervous system that express p75 after injury or disease. mice in a model of lipopolysaccharide (LPS)-induced lung fibrosis (). LPS-treated wt mice showed widespread extravascular fibrin deposition () and decreased proteolytic activity after LPS treatment (), when compared with saline-treated wt mice (). mice show a 2.58-fold decrease of fibrin immunoreactivity () and increased proteolytic activity (). Decreased proteolytic activity in the lung after injury depends on the up-regulation of PAI-1 (). Loss of PAI-1 protects from pulmonary fibrosis in LPS-induced airway disease, hyperoxia, and bleomycin-induced fibrosis (). Because p75 increases PAI-1 ( and ), we examined whether p75 regulates expression of PAI-1 in vivo. PAI-1 is up-regulated in LPS-treated wt mice () when compared with saline-treated wt mice (). mice show similar immunoreactivity for PAI-1 () as saline-treated wt mice (), suggesting that p75 induces up-regulation of PAI-1 after injury in the lung. mice (). mice, rolipram reduces fibrin deposition in the lung (Fig. S5, a and b; available at ) and sciatic nerve (Fig. S5, d–f), and decreases PAI-1 in the lung (Fig. S5 c), suggesting the involvement of PDE4 in p75-mediated inhibition of fibrinolysis in vivo. Collectively, our data show that p75 increases fibrin deposition via a PDE4-mediated inhibition of plasminogen activation in both LPS-induced lung fibrosis and sciatic nerve crush injury. These data suggest a role for p75/PDE4 signaling as a general regulator of plasminogen activation and fibrinolysis at sites of injury. Our study shows a novel direct interaction of p75 with PDE4A4/5, a specific PDE4 isoform, which results in the regulation of cAMP, a major intracellular signaling pathway, and mediates a major biological function of extracellular proteolysis and fibrinolysis (). p75 is expressed in a wide range of tissue injury models, where repair depends upon both cell differentiation and ECM remodeling. For example, we recently showed that in the absence of plasminogen the effects of p75 in tissue repair are protective due to its beneficial effects in cell differentiation (). Similarly, in the flow-restricted carotid artery model of vascular injury that depends on uPA and not on tPA-mediated fibrinolysis (), p75 is protective due to the induction of smooth muscle cell apoptosis (). In the sciatic nerve p75 appears to have a dual role by sustaining fibrin deposition (our study), and also promoting myelination (; ). mice after peripheral nerve injury would reveal the contribution of p75-mediated ECM remodeling and remyelination to the regeneration process. Overall, the biological role of p75 after tissue injury would probably depend on the relative contributions of its role as a regulator of cell death and differentiation and its role as an inhibitor of fibrinolysis. We identify regulation of cAMP as a novel signaling mechanism downstream of p75. Previous studies showed that β-adrenergic receptors target degradation of cAMP to the membrane via recruitment of multiple PDE4 isoforms, such as PDE4B1, PDE4B2, and PDE4Ds (). Our finding of interaction between p75 and PDE4A4/5 represents the first example of recruitment of a single PDE4 isoform to a transmembrane receptor. While interaction of β-adrenergic receptors with PDE4s is mediated via β-arrestin, our study suggests that the interaction of p75 with PDE4A4/5 could be potentially mediated by direct binding to PDE4A domains, such as the C-terminal domain that is unique to this sub-family. In addition, in co-IP experiments we do not detect an interaction between p75 and β-arrestin (unpublished data). It is possible that the unique C-terminal domain of PDE4A could regulate isoform-specific PDE4 recruitment to subcellular locations. Biological roles have been described for PDE4D in ischemic stroke () and heart failure () and for PDE4B in schizophrenia (). Our study identifies a biological function for PDE4A4/5 as a molecular mediator of p75/cAMP/PKA signaling involved in the regulation of tPA and fibrinolysis. In spinal cord injury in rodents, elevation of cAMP via the PDE4 inhibitor, rolipram, promotes axonal regeneration and functional recovery (). In the sciatic nerve, reduction of cAMP after injury is attributed primarily to up-regulation of PDE4 by SCs (). Based on our findings, it is possible that reexpression of p75 after injury could contribute to the activation of PDE4 and down-regulation of cAMP. BDNF, but not NGF, increases cAMP in neurons via TrkB (). Moreover, BDNF/TrkB signaling overcomes the inhibition of nerve regeneration by myelin proteins via inhibition of PDE4 (). We provide the first evidence for p75 in the regulation of cAMP by using genetic depletion, siRNA knockdown or up-regulation of the p75. Our results suggest that p75 might exert the opposite function as Trk receptors by recruiting PDE4A4/5 and decreasing cAMP. Interestingly, PDE4A has been detected as the predominant PDE4 isoform at the corticospinal tract (). Because p75 can act as a coreceptor for NogoR, a mediator of the inhibition of nerve regeneration, PDE4A interaction with p75 could play an inhibitory role in nerve regeneration by competing with neurotrophin signaling via Trk receptors. It is possible that the increased expression of p75 by neurons, glia, and brain endothelial cells could regulate the temporal and spatial pattern of tPA expression during brain injury or inflammation. p75 might also be upstream of other non-fibrinolytic functions associated with tPA, such as neurodegeneration, synaptic plasticity, and long-term potentiation (). Given the dependence of p75 functions on the availability of ligands and coreceptors (; ), further analysis will determine the role of p75 in extracellular proteolysis and ECM remodeling in different cellular systems. We show that expression of p75 can inhibit tPA in the absence of neurotrophin ligands. Constitutive expression of p75 may signal in a neurotrophin-independent manner to induce neuronal apoptosis (), activation of Akt (), and RhoGTPase (). The regulation of cAMP identified here is an effect of expression of p75 that does not appear to depend on neurotrophin signaling. The cellular distribution of PDE4A4/5 would determine the involvement of p75 in the regulation of cAMP. It is possible that non-neurotrophin ligands that bind directly to p75, such as β-amyloid (), as well as the myelin/NogoR p75–dependent inhibitors of nerve regeneration (), are able to regulate both cAMP and plasminogen activation by p75. Because cAMP analogues decrease expression of p75 (), it is possible that p75 by decreasing cAMP contributes to the positive regulation of its expression. Because PKA phosphorylates p75 and regulates its translocation to lipid rafts (), p75 via regulation of PKA might regulate its own subcellular localization. Given the multiple genes regulated by cAMP and PKA, other cellular functions may be regulated by p75/cAMP signaling. NGF/p75 signaling has been suggested to enhance local neurogenic inflammation to exacerbate pulmonary disease (). Our study suggests an additional pathway for p75 as a regulator of expression of PAI-1 and a mediator of fibrosis. p75 in the lung is detected mainly in basal epithelial cells of bronchioles (unpublished data). Similar to p75, PAI-1 is expressed by bronchial epithelial cells () and its expression results in an antifibrinolytic environment within the airway wall. Fibrin regulates both inflammation and airway remodeling (; ). It is therefore possible that p75-mediated regulation of PAI-1 via PDE4 could influence inflammatory and tissue repair processes in pulmonary disease. Although in chronic obstructive pulmonary disease the PDE4A4 isoform is specifically up-regulated () and considered a pharmacologic target (), all available inhibitors target the common catalytic domain of all PDE4 isoforms resulting in unwanted side-effects. Our study suggests that the p75/PDE4A4 interaction could be a potential therapeutic target that will achieve the specific inhibition of a single PDE4 isoform and may have therapeutic potential. Collectively, we have identified a novel cAMP-dependent signaling pathway initiated by p75 that specifically regulates plasminogen activity and scar formation after sciatic nerve and lung injury. Though p75 is responsible for a variety of cell survival and death decisions (), our data has revealed an unrecognized property of this receptor to regulate the degradation of cAMP. This property provides a potential mechanism to account for how p75 acts at sites of injury to promote ECM remodeling. The impact of high levels of p75 expression upon inhibition of extracellular proteolysis indicates that the detrimental effects of p75 extend beyond cell growth and axon inhibition. Finally, the dramatic inhibitory effect of p75 signaling on plasminogen activation suggests that the p75/PDE4A4 interaction represents a novel target for therapeutic intervention in both neuronal and nonneuronal tissues. mice () were in C57Bl/6 background and purchased from The Jackson Laboratory. mice were generated by crossing mice with mice. C57Bl/6J mice were used as controls. Sciatic nerve crush was performed as described previously (). Lung fibrosis was induced as described previously (). For the rolipram treatments, mice were administered 5 mg/kg rolipram (Calbiochem) before the LPS injection as described previously (). Mice were killed 4.5 h after LPS or saline administration. For rolipram treatment after sciatic nerve injury, mice were injected with rolipram (1 mg/kg) once daily for 8 d until tissue was harvested and processed for immunostaining. Immunohistochemistry was performed as described in . Primary antibodies were sheep anti–human fibrin(ogen) (1:200; US Biologicals), rabbit anti–human tPA (1/300; Molecular Innovations), rabbit anti-p75 clone 9651, (1:1,000), goat anti-p75 (1/200; Santa Cruz Biotechnology, Inc.), rabbit anti–mouse PAI-1 (1:500; a gift from David Loskutoff, Scripps Research Institute, La Jolla, CA), and mouse anti-S100 (1:200; Neomarkers). For immunofluorescence, secondary antibodies were anti–rabbit FITC and anti–goat Cy3 (1:200; Jackson Immunochemicals). Images were acquired with an Axioplan II epifluorescence microscope (Carl Zeiss MicroImaging, Inc.) using dry Plan-Neofluar lenses using 10× 0.3 NA, 20× 0.5 NA, or 40× 0.75 NA objectives equipped with Axiocam HRc digital camera and the Axiovision image analysis system. Immunoblot was performed as described previously (). Antibodies used were rabbit anti-p75 clones 9992 and 9651 (1:5,000), mouse anti-fibrin (1:500; Accurate Chemical & Scientific Corp.), rabbit anti-myosin (1:1,000; Sigma-Aldrich), rabbit anti-GAPDH (1:5,000; Abcam) and rabbit anti-PAI-1 (1:5,000; a gift of David Loskutoff). Quantification was performed on the Scion NIH Imaging Software. Fibrin precipitation and quantification from lung tissues was performed exactly as described previously (). Co-IP was performed as described previously (). Cell lysates were prepared in 1% NP-40, 200 mM NaCl, 1 mM EDTA, and 20 mM Tris-HCl, pH 8.0. IP was performed with an anti-p75 antibody and immunoblot with anti-PDE4A, PDE4B, PDE4C, and PDE4D (Fabgennix). The co-IP buffer using NP-40 has been previously used to examine interactions of p75 with other intracellular proteins, such as TRAF-6 () and PKA (). For mapping experiments, PDE4A5 cDNA was cotransfected with HA-tagged p75 deletion constructs into HEK293 cells. IP was performed with an anti-HA antibody (Cell Signaling). Cell lysates were probed with an anti-PDE4A or an anti-p75 antibody. For co-IP experiments using recombinant proteins, equimolar amounts (2 μM) of purified recombinant MBP-PDE4A5 (), MPB-PDE4A4 (), MBP-PDE4D3 (), and GST-p75-ICD () were mixed in binding buffer (50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 2 mM MgCl, 1 mM DTT, 0.5% Triton X-100, and 0.5% BSA) and incubated for 1 h at 4°C. Washed glutathione-Sepharose beads were added according to the manufacturer's instructions for an additional hour. Beads were sedimented by centrifugation (10,000 for 1 min) and washed three times. Proteins associated with the beads were eluted by boiling in loading buffer and separated by SDS-PAGE. RT-PCR was performed as described previously (). Primers for , and genes were used as described previously (). Real-time PCR was performed using the Opticon DNA Engine 2 (MJ Research) and the Quantitect SYBR Green PCR kit (QIAGEN). Results were analyzed with Opticon 2 software using the comparative Ct method as described previously (). Data were expressed as ΔΔCt for the gene normalized against . Quantification of tPA and uPA activity in SC and fibroblast in lysates and supernatants was performed according to the directions of the activity assay kits from American Diagnostica and Chemicon, respectively. To elevate cAMP cells were treated either with 2 mM db-cAMP (Sigma-Aldrich) or with 10 μM forskolin (Sigma-Aldrich) for 16 h. To block PKA activity, cells were treated with 200 nM KT5720 (Calbiochem). Induction with neurotrophins was performed using 100 ng/ml NGF and 50 ng/ml BDNF for 16 h before tPA assay. Coating with fibrin was prepared as described previously (). To quantitate fibrin degradation, the supernatant was aspirated and the remaining gel was weighed using an analytical balance. Decrease of gel weight corresponded to increased fibrin gel degradation. Murine SCs were isolated as described previously (). NIH3T3 or HEK293 cells were cotransfected either with p75 FL, ICD or deletion constructs, and PDE4A5 cDNAs using Lipofectamine 2000 (Invitrogen) as described in the Results section. CGNs were isolated from P10 animals (). CGNs were lysed immediately for co-IP, without plating. siRNA directed against p75 (Dharmacon) was transfected into SCs and NIH3T3p75 cells using Dharmafect (Dharmacon). 10 fibroblasts or 500,000 SCs were lysed in 0.1 N HCl solution and cAMP was measured using a competitive binding cAMP ELISA (Assay Designs). Cells were treated with 100 ng/ml PTX for 16 h. For inhibition of PDE activity, cells were treated for 16 h with 500 μM isobutyl methylxanthine (IBMX; Calbiochem), 18.7 μM 8-methoxymethyl-3-isobutyl-1-methylxanthine (PDE1 inhibitor; Calbiochem), 80 μM -9-(2-Hydroxy-3-nonyl)adenine (PDE2 inhibitor; Calbiochem), 100 nM trequinsin (PDE3 inhibitor; Calbiochem), and 10 μM rolipram (PDE4 inhibitor; Calbiochem). Cells were treated with forskolin in the presence of the inhibitors for 1 h. Because these inhibitors specifically inhibit a PDE isoform and have no effect on the other PDE isoenzymes (), they are extensively used for the identification of the specific PDE isoform that is involved in different cellular functions. Induction with neurotrophins was performed using 100 ng/ml NGF or 50 ng/ml BDNF, 750 ng/ml of FcTrkB, or 1.35 ug/ml of Fcp75 for 1 h before cAMP assay. For the qualitative and quantitative PKA assay (Promega), cells were treated with 10 μM forskolin for 30 min, lysed in 1% NP-40 buffer with 150 mM NaCl, 50 mM Tris, and 1 mM EGTA, and protein concentration was determined using the Bradford Assay (Bio- Rad Laboratories). 1 μg was loaded into the PKA assay reaction mix according to the manufacturer's protocol (Promega). In situ zymographies were performed as described previously (). Quantification of in situ zymographies was performed by measuring the area of the lytic zone surrounding each nerve, and dividing that value by the area of the nerve. Images were collected after 8 h of incubation for the sciatic nerve and 4 h of incubation for the lung. For cell zymographies, cultures were washed four times with PBS/BSA and overlaid with 200 μl of Dulbecco's minimum essential medium containing 1% LMP agarose, 2.5% boiled nonfat milk, and 25 μg/ml human plasminogen. The overlay was allowed to harden, and plates were incubated in a cell culture incubator at 37°C. Pictures of lytic zones were taken using an inverted microscope under dark field (Carl Zeiss MicroImaging, Inc.). For the construction of pm-AKAR2.2 we used the previously described cytoplasmic PKA sensor, AKAR2 (). pm-AKAR2.2 consists of a cDNA containing a FRET pair, monomeric enhanced cyan fluorescent protein (ECFP), and monomeric citrine (an optimized version of YFP), fused to forkhead associated domain 1 (FHA1) (Rad53p 22–162), and the PKA substrate sequence LRRATLVD via linkers. A206K mutations were incorporated to ECFP and Citrine by the QuikChange method (Stratagene). The C-terminal sequence from K-Ras KKKKKKSKTKCVIM was added to target the construct to the plasma membrane. For expression in mammalian cells, the chimaeric proteins were subcloned into a modified pcDNA3 vector (Invitrogen) behind a Kozak sequence as described previously (). For the generation of the PDE4A4CT, PDE4A4 was subcloned into p3XFLAG-CMV-14 using plasmid pde46 (GenBank/EMBL/DDBJ accession no. ) as template from Met-1 to Iso-721 (). A forward (5′) primer containing a HindIII restriction site immediately 5′ to the initiating Met-1 (ATG) of PDE4A4 and a reverse primer designed to the DNA sequence ending at Iso-721 (ATA) with BamHI restriction site immediately 3′ to Iso-721 was used to amplify Met-1 to Iso-721. The C terminus was removed simply by amplifying from Iso-721 instead of the final codon at the end of the full-length PDE4A4B. The C-terminally truncated PDE4A4B was cloned in-frame with three FLAG (Asp-Tyr-Lys-Xaa-Xaa-Asp) epitopes (Asp-726, Asp-733 & Asp-740) after the BamHI restriction site, therefore at the C terminus of the now-truncated PDE4A4. The stop codon (TAG) after the FLAG epitopes is located immediately after Lys-747. This strategy generates a C-terminal truncate of PDE4A4 from 1–721. NIH3T3 cells and NIH3T3p75 cells were transiently transfected with pm-AKAR2.2, AKAR3, or pm-AKAR3 () and imaged within 24 h of transfection. Cells were rinsed once with HBSS (Cellgro) before imaging in HBSS in the dark at room temperature. An Axiovert microscope (Carl Zeiss Microimaging, Inc.) with a MicroMax digital camera (Roper-Princeton Instruments) and MetaFluor software (Universal Imaging Corp.) was used to acquire all images. Optical filters were obtained from Chroma Technologies. CFP and FRET images were taken at 15-s intervals. Dual emission ratio imaging used a 420/20-nm excitation filter, a 450-nm dichroic mirror and a 475/40-nm or 535/25-nm emission filter for CFP and FRET, respectively. Excitation and emission filters were switched in filter wheels (Lambda 10–2; Sutter Instrument Co.). Peptide libraries were synthesized by automatic SPOT synthesis (). Synthetic overlapping peptides (25 amino acids in length) were spotted on Whatman 50 cellulose membranes according to standard protocols by using Fmoc-chemistry with the AutoSpot Robot ASS 222 (Intavis Bioanalytical Instruments AG). Membranes were overlaid with 10 μg/ml recombinant GST-p75 ICD. Bound recombinant GST-p75 ICD () was detected using rabbit anti-GST (1:2,000; GE Healthcare) followed by secondary anti–rabbit horseradish peroxidase antibody (1:2,500; Dianova). Alanine scanning was performed as described previously (). Statistical significance was calculated using JMP2 Software by unpaired test for isolated pairs or by analysis of variance (one-way ANOVA) for multiple comparisons. Data are shown as the mean ± SEM. Fig. S1 shows that genetic loss of p75 increases tPA mRNA levels and proteolytic activity in the cerebellum. Fig. S2 demonstrates that treatment with neurotrophins has no effect on cAMP levels in NIH3T3 cells. Fig. S3 shows endogenous coimmunoprecipitations of PDE4A5 and p75 from injured sciatic nerve and from primary CGNs. Fig. S4 shows results from transient transfections of NIH3T3 cells with p75 and the PKA activity reporters, AKAR3 and pm-AKAR3. Fig. S5 shows that inhibition of PDE4s with rolipram decreases fibrin deposition both in LPS-induced lung fibrosis and sciatic nerve crush injury. The online version of this article is available at .

Shigellosis, which is characterized by diarrhea and dysentery, is a worldwide human health problem caused by the group of bacteria (). These bacteria produce a toxin, Shiga toxin, that is comprised of two components referred to as A and B, in which the A subunit is the enzymatically active toxin responsible for inactivation of the 28s ribosomal RNA, and the B subunit is required for binding and entry into target cells after transport across the polarized epithelial cells of the gut from their apical surface (for reviews see , ; ; ; ). How Shiga toxin is able to enter and intoxicate cells is of great medical interest; therefore, this has been studied extensively (; , ). The Shiga toxin B subunit (STxB) binds to Gb3, a neutral glycosphingolipid at the surface of target cells in the gut vasculature, as well as in the kidney and brain and is then internalized by a combination of clathrin-mediated and clathrin-independent endocytic pathways (; ; for review see ). Once internalized, Shiga toxin is then transported directly from early and recycling endosomes to the trans-Golgi network (; ). This pathway is dependent on the function of clathrin and dynamin (), the small GTP-binding protein Rab6 and its effector proteins, the lipid phosphatase OCRL1, a defined set of SNAREs (; ; ; ; ), and a second GTP-binding protein Arl1 and its effector golgin-97 (). From the TGN, the toxin is then transported by a retrograde pathway dependent on Rab6 through the Golgi apparatus to the ER (; ), where the toxin A subunit enters the cytoplasm most likely by retrotranslocation (for review see ). Despite the multistep nature of the Shiga toxin uptake pathway, only a single Rab GTP-binding protein, Rab6, has been shown to play any role in its transport (; ; ). Because each membrane trafficking step is thought to involve a discrete set of Rabs (), it seems likely that additional Rabs are involved in Shiga toxin uptake. To further define the pathway of Shiga toxin transport between the cell surface and the Golgi apparatus, we have focused on the identification of Rab GTPase-activating proteins (GAPs) that interfere with this step. Rab GAPs are characterized by a conserved catalytic domain, the TBC (Tre2/Bub2/Cdc16) domain (; ; ), that promotes GTP hydrolysis by a dual arginine/glutamine finger catalytic mechanism related to the arginine finger mechanism of Ras GAPs (, 1999; ; ). As we have shown previously in the case of Rab5 (), Rab GAPs can be used to specifically inactivate the endogenous pool of a Rab and, thus, interfere with the process this Rab is involved in. Nonspecific effects of Rab GAP expression can easily be discriminated from the specific effects of Rab inactivation by use of an inactive point mutant in which the catalytic arginine residue is replaced by alanine (). Because the human genome encodes >60 Rabs and at least 39 TBC domain–containing proteins, the identification of specific Rab–Rab GAP pairs is not a trivial task (; ). In the present study, we have screened for human Rab GAPs that specifically inhibit the transport of Shiga toxin to the Golgi apparatus and do not have any effects of the uptake of EGF and its receptor. By combining this functional assay with biochemical analysis of GTP hydrolysis, we are able to identify discrete pairs of Rabs and Rab GAPs acting in the Shiga toxin and EGF transport pathways. To investigate the requirements for specific Rab GAPs in the trafficking of Shiga toxin to the Golgi but not the endocytic uptake of growth factors such as EGF, it was necessary to establish conditions for the uptake of these two ligands. Fluorescently labeled STxB and EGF were bound to the cell surface on ice, and, after washing to remove excess toxin, cells were warmed, and uptake of the two ligands was followed by immunofluorescence microscopy (). Initially, both ligands were present on the cell surface (0 min) and became internalized into small punctate structures at 10–20 min (). After this, the two markers separated, and Shiga toxin accumulated in a perinuclear compartment from 30 to 60 min that was previously shown to correspond to the trans-Golgi network and Golgi stacks (; ), whereas, as expected, EGF remained in punctate endosomal structures (). Interestingly, although both ligands have been reported to be taken up by clathrin-dependent pathways (; ), the kinetics of uptake are clearly different, indicating that there may be some differences in the mechanistic details such as the requirement for Rab GTPases. Therefore, for further experiments, it was decided to take a 60-min uptake of Shiga toxin to measure the efficiency of its transport to the Golgi and a 30-min uptake of EGF to measure its uptake into early endosomes. The uptake of Shiga toxin has previously been shown to be Rab6 dependent (; ), whereas that of EGF should be Rab6 independent. To verify this, the uptake of these two ligands was followed in cells depleted of Rab6 (). Immunofluorescence analysis showed that Rab6 was efficiently depleted after 72 h of treatment with specific siRNA duplexes (). Importantly, the Rab6 effector protein Bicaudal-D1 was redistributed from the Golgi to the cytoplasm (), indicating that Rab6 function was impaired. Compared with control cells, Rab6 depletion resulted in a decrease in the amount of Shiga toxin reaching the Golgi apparatus after 60 min of uptake (). Under these conditions, the Golgi marker TGN46 was unaltered, suggesting that this decrease is not caused by disruption of the Golgi apparatus. In contrast, no obvious differences could be seen in the uptake of EGF when comparing control and Rab6-depleted cells (). Expression of a dominant-active mutant of Rab5 resulted in the formation of enlarged early endocytic structures coated with the Rab5 effector EEA1 as expected (). These structures accumulated both EGF and Shiga toxin, suggesting that these ligands traffic via a Rab5-dependent pathway through early endosomes (). Therefore, EGF takes a Rab6-independent trafficking route, whereas Shiga toxin transport to the Golgi is Rab6 dependent. Based on these findings, a Rab6 GAP would be expected to block Shiga toxin but not EGF uptake, whereas a Rab5 GAP might block both. Because specific GAPs for both Rab5 and Rab6 have previously been identified (; ; ), these predictions were tested. It has previously been reported that TBC1D11/GAPCenA is the GAP for Rab6 (), and, thus, this was tested for its ability to block the trafficking of Shiga toxin to the Golgi apparatus. Surprisingly, the expression of TBC1D11/GAPCenA had no obvious effect on the transport of Shiga toxin to the Golgi apparatus () or on the ability of Rab6 or the Rab6-dependent effector Bicaudal-D1 (; ) to target to Golgi membranes (). Therefore, TBC1D11/GAPCenA was further investigated to find out whether it has the biochemical properties expected of a Rab6 GAP. Confirming previous findings (), a C-terminal fragment of TBC1D11/GAPCenA was identified when Rab6 was screened against a yeast two-hybrid cDNA library (unpublished data). However, this fragment corresponding to a predicted coiled-coil region adjacent to the GAP domain does not show a specific interaction with Rab6 and binds to human Rabs of many different subfamilies when tested against a representative collection of human Rabs (). When full-length TBC1D11/GAPCenA was tested, it showed a strong interaction with Rab4, a weaker interaction with Rab11, and only a very weak interaction with Rab6 (). A more detailed analysis in which multiple deletion constructs of TBC1D11/GAPCenA were tested revealed that regions N and C terminal to the core TBC domain contribute to the recognition of specific Rabs (Fig. S1, available at ). This suggested that Rab6 might not be the target for TBC1D11/GAPCenA after all, and biochemical assays were performed to test this. In agreement with the yeast two-hybrid data, TBC1D11/GAPCenA had strong GAP activity toward Rab4 and a lesser activity toward Rab11 (). Only very weak activity was detected toward Rab2 and 14 (), which fall into the same Rab subfamily as Rab4 and 11, showing that TBC1D11/GAPCenA can discriminate between closely related Rabs. No activity could be detected toward Rab6 or any of the other Rabs tested (), although all were loaded with GTP to the same extent. It was previously suggested that Rab6 has to be prenylated to be an efficient substrate for TBC1D11/GAPCenA, and this was tested. However, no GAP activity could be seen when prenylated Rab6 purified from insect cells was used (, Rab6). It should also be noted that Rab4 was prepared in bacteria and was therefore not prenylated but was still an excellent substrate for TBC1D11/GAPCenA. These data show that TBC1D11/GAPCenA is a GAP for Rab4 but not Rab6 and that it does not alter the trafficking of Shiga toxin to the Golgi apparatus. An unbiased approach was taken to identify which Rab GAPs act on the Shiga toxin transport pathway. Cells expressing the various predicted human Rab GAPs were tested blindly for their ability to transport Shiga toxin to the Golgi apparatus and EGF to early endosomes (unpublished data). Candidate positives from this first round of screening were then retested, comparing the effects of the wild-type GAP to that of a catalytically inactive point mutant. In this screen, 6 of 39 predicted GAPs showed catalytic activity–dependent effects on Shiga toxin trafficking to the Golgi apparatus ( and ). In contrast, in a parallel screen, only RabGAP-5 was able to block the uptake of EGF to early endosomes ( and ). To be sure that the lack of a Golgi signal for Shiga toxin after 60 min of uptake was not simply a result of disruption of the Golgi, the integrity of this organelle was tested using markers for the cis- and trans-Golgi compartments (Fig. S2, available at ). This resulted in the elimination of GAPs such as TBC1D22A and 22B in which the Golgi was fragmented, and Shiga toxin was found to localize to these fragments (Fig. S2 and not depicted). In the case of TBC1D14, although the Golgi was fragmented, Shiga toxin did not localize to these Golgi fragments, indicating there was a block in transport. However, these effects were not caused by Rab inactivation because the inactive point mutant gave a similar phenotype (; and Fig. S2). Interestingly, there was no effect on the trafficking of EGF, suggesting that TBC1D14 is not disrupting early endosome function. In contrast, RN-tre, although fragmenting the Golgi, caused a block of Shiga toxin in more peripheral structures discrete from the Golgi dependent on its catalytic activity; therefore, this was counted as a positive ( and S2, and not depicted) . The ability of candidate positives to protect cells against the cytotoxic effects of complete Shiga-like toxin 1 was then tested (). This assay measures protein synthesis, which is inhibited if Shiga toxin can reach the ER and enter the cytoplasm and, therefore, is a more stringent measure of Shiga toxin trafficking than fluorescent assays (). This approach showed that EVI5, RN-tre, and TBC1D17 protect against Shiga-like toxin 1 between 1.87- and 3.35-fold, whereas RabGAP-5 with a fold protection of 1.07 could not (). The TBC1D10 family of GAPs showed variable extents of protection, and, therefore, it is unclear whether these represent specific positives with this assay. Interestingly, none of the positive GAPs were able to cause the release of the Rab6 effector Bicaudal-D1 from Golgi membranes (Fig. S3, available at ), even those that showed some effects on Golgi morphology (Fig. S2). This suggests that they do not act on Rab6. The trafficking pathway of Shiga toxin from the cell surface to the Golgi is therefore defined by the following Rab GAPs: EVI-5, RN-tre, the TBC1D10 family, and TBC1D17. Although RN-tre was initially suggested to act on Rab5 and, thus, the EGF uptake pathway (), more recently, another GAP, RabGAP-5, has also been proposed to fulfill this function (). To clarify this point, the effects of RN-tre and RabGAP-5 on the uptake of EGF and Shiga toxin were directly compared (). RabGAP-5 was able to block the uptake of EGF () and displace the Rab5-dependent effector molecule EEA1 from endosomes () but had no effect on the uptake of Shiga toxin (). RN-tre or a catalytically inactivate point mutant of RabGAP-5 did not show these effects on EGF uptake (). In contrast, RN-tre was able to block the transport of Shiga toxin to the Golgi apparatus () and caused partial fragmentation of the Golgi apparatus ( and S2), which is consistent with previous observations that its target, Rab43, is localized to this organelle (). Note that Rab43 was referred to as Rab41 in a previous study (), which is consistent with the naming in mice; however, the gene nomenclature has now been clarified so that this gene is called Rab43 in both the mouse and human. RabGAP-5 or a catalytically inactive point mutant of RN-tre did not show these effects (). In some cells expressing very high levels of RN-tre, the uptake of EGF was blocked, but further investigation revealed that these cells lack detectable EGF receptor on the cell surface (unpublished data). Therefore, the apparent uptake block is caused by a lack of EGF binding and not by a defect in endocytosis of the receptor–ligand complex. One explanation for the reduced level of EGF receptor at the cell surface could be the disrupted Golgi apparatus seen in RN-tre–expressing cells ( and S2). Thus, any effects of RN-tre on EGF receptor uptake would appear to be indirect. To identify the target Rabs of the positive GAPs identified thus far, they were tested for their ability to accelerate GTP hydrolysis by a specific Rab in biochemical assays with a wide set of human Rabs (). These data clearly show that of the Rabs tested, the Rab5a-c subfamily are substrates for RabGAP-5, whereas Rn-tre acts on Rab43. EVI5 showed strong and specific activity toward Rab35 (>1,200 pmol/h of hydrolyzed GTP), but Rab35 was found to be weakly (sixfold less) activated with several of the GAPs tested here, suggesting that this may be a false positive hit. The substrates of TBC1D10B and TBC1D17 cannot be identified with certainty from the data obtained. Careful analysis of the pattern of GTP hydrolysis suggests that TBC1D10B might act on Rab22a/Rab31, which fall into the same branch of the Rab5 subfamily, whereas TBC1D17 might be a GAP for Rab21 (). If the effects of RabGAP-5 on EGF uptake are mediated through Rab5 and those of RN-tre on Shiga toxin transport are mediated through Rab43, cells depleted of these Rabs should display a similar defect to that seen on overexpressing the respective GAP because both conditions reduce the pool of active Rab. In line with this prediction, EEA1 fails to localize to endosomes, and there is a strong reduction in EGF uptake in cells depleted of Rab5 (). These cells are still able to transport Shiga toxin to the Golgi apparatus similar to the control cells (). In contrast, the Golgi is disorganized, and there is a strong reduction in the transport of Shiga toxin to the Golgi in cells depleted of Rab43 (). However, these cells are still able take up EGF similar to the control cells (). Consistent with a role for RN-tre and Rab43 in the transport of Shiga toxin to the Golgi apparatus, Rab43 localized to this organelle and showed a staining pattern most similar to the trans-Golgi marker TGN46 (). Rab43 showed no overlap with the early endosome marker EEA1 (), suggesting that it does not function at early endosomes. Finally, although the siRNA duplexes used were efficient in targeting the appropriate Rab (Fig. S5 A, available at ), the depletion of Rab21, 22a, 31, and 35 had no clear effects on either Shiga toxin or EGF uptake (unpublished data), and it is therefore unclear whether these Rabs are the targets of the TBC1D10 family of GAPs, TBC1D17, and EVI5. Thus, it will require further investigation to validate the targets of these GAPs. In this study, we have presented evidence that different sets of Rab GAPs and their target Rabs control specific endocytic trafficking pathways. These results suggest that although the trafficking of Shiga toxin from the cell surface to the Golgi is a multistep process dependent on at least six different Rabs and Rab GAPs, EGF trafficking depends only on the action of Rab5 and RabGAP-5. There has been some controversy in the literature concerning the relative importance of RN-tre and RabGAP-5 as regulators for Rab5 and their involvement in the uptake of endocytosed ligands (; ). However, the results presented here clearly show that although both of these GAPs are involved in the trafficking of endocytosed ligands, they act on discrete trafficking pathways. One explanation for the differential sensitivity of EGF receptor trafficking to RN-tre could relate to the conditions and cell lines used. With low doses of EGF, the EGF receptor is almost entirely transported via the clathrin-dependent pathway, whereas at higher or saturating doses, it spills over into the clathrin- independent pathway (). Thus, with saturating amounts of EGF, RN-tre would be able to prevent transport of the pool of EGF receptor entering via the clathrin-independent pathway. However, our data suggest that this would be a Rab5-independent and Rab43-dependent pathway. None of the GAPs capable of blocking Shiga toxin transport was able to cause the release of Bicaudal-D1 from the Golgi to the cytoplasm (Fig. S3), and this indicates that they do not act on Rab6. Why the Rab6 GAP was not found in the screen is unclear, but because the depletion of Rab6 reduces Shiga toxin transport to the Golgi (; ), this has to be viewed as a false negative. There is evidence that some TBC domain proteins form heterodimeric complexes () and that this is necessary for their activity. If this is the case for the Rab6 GAP, it could have been missed in the current screen. Some of the results presented here suggest that there are problems with the use of dominant-active GTP-locked Rabs. Although the expression of dominant-active Rab5 results in the trapping of Shiga toxin in an early endocytic compartment, indicating its transport is Rab5 dependent, this interpretation is not supported by the use of specific Rab GAPs. Although RabGAP-5 does block EGF and transferrin trafficking (), it has no effect on the transport of Shiga toxin to the Golgi (). This suggests that although Shiga toxin trafficking is normally Rab5 independent, when dominant-active Rab5 is expressed, it becomes trapped in the enlarged early endocytic compartment created under these conditions (). One explanation for this could be that Shiga toxin, although normally trafficking in a Rab5-independent fashion, does pass through a Rab5-positive endocytic compartment, and this can also be perturbed by the overexpression of active mutant forms of Rab5. In contrast, EGF trafficking is blocked by both RabGAP-5 and Rab5 ( and ) and, therefore, does follow a Rab5-dependent pathway through early endosomes. Thus, dominant-active Rab5 appears to create a situation in which the function of the early endocytic pathway is perturbed such that molecules that normally traffic in a Rab5-independent manner are also affected. Dominant-active Rabs should therefore be used with some caution, and, as shown here, specific Rab GAPs provide valuable and specific tools to manipulate the activity of endogenous Rabs. It has been suggested that Rab GAPs are not particularly specific toward their target Rab (; ). However, this is somewhat counterintuitive because Rabs are argued to be important components acting at specific membrane trafficking steps and for the specification of organelle identity (; ), and one would therefore expect their regulators to be equally specific. We believe that Rab GAPs are likely to be specific toward particular target Rabs or Rab subfamilies and have several arguments to support this statement. We show that GAPs such as TBC1D11/GAPCenA, RabGAP-5, and RN-tre are highly active toward specific Rabs in biochemical assays. In addition, we can demonstrate that Rab GAPs have highly specific effects on discrete membrane trafficking pathways, which is consistent with the idea that they act on specific Rabs in vivo. A possible explanation for this discrepancy comes from our investigation of TBC1D11/GAPCenA, which was previously suggested to act on Rab6 (). In our hands, TBC1D11/GAPCenA acts on Rab4 but not Rab6 and does not have the effects on Rab6 effectors or Shiga toxin transport that is expected for a Rab6 GAP (). Investigation of the interaction between TBC1D11/GAPCenA and Rabs shows that regions outside of the minimal predicted catalytic domain are required for specific Rab binding ( and S1). Strikingly, deletion of these regions relaxes the interaction specificity of the protein. Previous studies have used truncated GAP domains rather than full-length proteins (; ), and, in light of our findings, this may be problematic because these proteins may have a relaxed specificity and, thus, may be able to accelerate GTP hydrolysis on a broader spectrum of Rabs. An analogous situation has been described for the GAP1 family of bifunctional Ras and Rap GAPs (). In this case, regions outside of the core catalytic domain are required for the recognition of Rap (). Further investigation of Rab GAP specificity is needed before any general conclusions can be made, but it is clearly important to study the full-length proteins and not only the truncated fragments corresponding to predicted domains. There are >60 Rab family GTPases encoded by the human genome and at least 39 TBC domain–containing Rab GAPs. However, little is known about the function of the majority of these proteins. This is partly the result of the enormous complexity of membrane trafficking systems and the lack of suitable cell biological models for Rabs with functions in specific tissues. As a first step in unravelling this complex network, we have screened for Rab GAPs that can influence the trafficking of Shiga toxin to the Golgi or of EGF to early endosomes. Using biochemical assays, we have then identified target Rabs for several of these GAPs. This approach should be useful for the study of other trafficking events, for trafficking between the ER and Golgi apparatus, or for a wide variety of different regulated secretory events. The antibody reagents used were as follows: mouse anti-GM130 and EEA1 (Becton Dickinson), sheep anti-TGN46 (Serotec), rabbit anti-Rab6 (Santa Cruz Biotechnology, Inc.), and Bicaudal-D1 (). Donkey secondary antibodies conjugated to HRP, AMCA, CY2, and CY3 were obtained from Jackson ImmunoResearch Laboratories. The bacterial expression construct for STxB was provided by Y. Misumi (Fukuoka University School of Medicine, Fukuoka, Japan). Human Rab GAPs were identified by searching the GenBank/EMBL/DDBJ database using the TBC domain signature motifs defined by . The human Rab GAPs (EVI-5, RN-tre [USP6NL], RUTBC1, RUTBC2, RabGAP-5 [RUTBC3], TBC1D1, TBC1D2, TBC1D3B, TBC1D4, TBC1D5, TBC1D6, TBC1D7, TBC1D8, TBC1D10A, TBC1D10B, TBC1D10C, TBC1D11 [GAPCenA], TBC1D12, TBC1D13, TBC1D14, TBC1D15, TBC1D16, TBC1D17, TBC1D18, TBC1D19, TBC1D20, TBC1D21, TBC1D22A, TBC1D22B, USP6, AK074305, KIAA1055, KIAA0676, KIAA0882, NP_060222, NP_060779, EVI5-like, KIAA0984, and KIAA1171) were then amplified from either human testis cDNA (Becton Dickinson) or HeLa cDNA using the pfu polymerase (Stratagene) and were cloned in pCRII-TOPO (Invitrogen). Point mutations were introduced using the QuikChange method (Stratagene). Constructs were confirmed by DNA sequencing (Medigenomix; Max Planck Institute biochemistry sequencing core facility). Mammalian expression constructs were made in pcDNA3.1+ (Invitrogen) modified to encode a myc epitope tag and the pEGFP-C2 vector (CLONTECH Laboratories, Inc.). For yeast two-hybrid assays, Rabs were inserted into the bait vector pFBT9, pGBT9 (CLONTECH Laboratories, Inc.) was modified to carry kanamycin resistance, and Rab GAPs and mutants thereof were inserted into the prey vector pACT2 (CLONTECH Laboratories, Inc.). Yeast two-hybrid assays were performed according to the yeast protocol handbook (CLONTECH Laboratories, Inc.) as described previously (). Bacterial expression was performed using the T7 polymerase hexahistidine-GST expression vector pFAT2 for Rabs and the maltose-binding protein expression vector pMalC2 (New England Biolabs, Inc.) for Rab GAPs and the BL21(DE3) and JM109 strains, respectively. Fusion proteins were purified over nickel-nitrilotriacetic acid agarose (QIAGEN) or amylose resin (New England Biolabs, Inc.). Proteins were dialysed overnight against 50 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 2 mM DTT, and aliquots were frozen in liquid nitrogen for storage at −80°C. For Rab-loading reactions, 10 μl of assay buffer, 73 μl HO, 10 μl 10 mM EDTA, pH 8.0, 5 μl of 1 mM GTP, 2 μl γ-[P]GTP (10 mCi/ml; 5,000 Ci/mmol; GE Healthcare), and 100 pmol Rab protein were mixed on ice. After 15 min of incubation at 30°C, loaded Rabs were stored on ice. GTP binding was measured using a nitrocellulose filter-binding assay (). GAP reactions were started by the addition of 0.5 pmol Rab GAP as specified in the figures. A 2.5-μl aliquot of the assay mix was scintillation counted to measure the specific activity in counts per minute/picomole GTP. Reactions were then incubated at 30°C, taking 5-μl samples in duplicate at suitable time points as indicated in the figures. The 5-μl aliquots were immediately added to 795 μl of ice-cold 5% [wt/vol] activated charcoal slurry in 50 mM NaHPO, left for 1 h on ice, and centrifuged at 16,100 in a benchtop microfuge (5417R; Eppendorf) to pellet the charcoal. A 400-μl aliquot of the supernatant was scintillation counted, and the amount of GTP hydrolyzed was calculated from the specific activity of the reaction mixture. HeLa cells were cultured at 37°C and 5% CO in growth medium (DME containing 10% FCS). HeLa cells plated on glass coverslips in a six-well plate at a density of 70,000 cells/well were used for plasmid transfection and at 25,000 cells/well for RNAi using conditions that were described previously (). Rabs were targeted using siRNA duplexes obtained from Dharmacon. The sequences are listed in Fig. S5 B. For Western blotting, cells from three wells of a six-well plate were washed in 2 ml PBS and lysed in 70–80 μl of 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, and 0.1% [wt/vol] Triton X-100. For each lane of a minigel, 10 μg of the protein lysate was loaded. Cytotoxicity was defined as a decrease in the ability of cells to incorporate [S]methionine into acid-precipitable material after Shiga-like toxin 1 treatment using an established method (). HeLa cells were transiently transfected with either wild-type or RA mutant Rab GAPs for 9 h using Fugene-6 and the manufacturer's protocol (Roche Diagnostics) and were replated in 96-well plates at a density of 1.5 × 10 cells/well and left for another 15 h. After washing with PBS, cells were incubated for 1 h with 100 μl DME/FCS containing serial twofold dilutions from 0.05–50 ng/ml Shiga-like toxin 1. Subsequently, cells were washed with PBS and incubated in PBS containing 50 μCi/ml [S]methionine for 30 min. Labeled proteins were precipitated with three washes in 5% [wt/vol] trichloroacetic acid, the wells were washed twice with PBS, 50 μl of scintillation fluid was added, and the amount of radiolabel incorporated was then determined in a counter (MicroBeta 1450 TriLux; PerkinElmer). For each condition, the IC for Shiga-like toxin 1 was calculated from the toxin titration performed in triplicate. EGF coupled to 200 μg/ml AlexaFluor488 or -555 (40× stock; Invitrogen) were stored as stock solutions in PBS at −20°C. For uptake assays, HeLa cells plated on glass coverslips at a density of 70,000 cells/well of a six-well plate were washed three times with serum-free growth medium 36 h after plating and were incubated in serum-free growth medium for 15–16 h at 37°C and 5% CO. Coverslips were then washed three times in ice-cold PBS and placed on 40-μl drops of uptake medium (DME, 2% [wt/vol] BSA, and 20 mM Hepes-NaOH, pH 7.5) and 5 μg/ml EGF on an ice-cold metal plate covered in Parafilm (Pechiney Plastic Packaging). After 30 min of incubation, the coverslips were washed three times in ice-cold PBS to remove excess ligand. One coverslip was fixed to give the total bound ligand, whereas the remaining coverslips were transferred to a six-well plate containing prewarmed growth medium and were incubated at 37°C and 5% CO. At the time points indicated in the figures, coverslips were fixed and processed for immunofluorescence microscopy. Recombinant STxB was prepared as described previously (; ) and labeled on amine residues for 5 min at room temperature with an -hydroxysuccinimidyl ester of Cy3 according to the manufacturer's instructions (GE Healthcare). These conditions lead to a stoichiometry of five Cy3 dye molecules per pentamer of STxB. The cell line and conditions used for uptake assays were the same as those described for EGF except that 0.7 μg/ml Cy3-STxB were used. For combined EGF and STxB assays, both proteins were mixed and bound simultaneously to the cell surface, and the standard procedure was followed. Cells to be imaged were fixed for 20 min in 3% [wt/vol] PFA, quenched for 10 min with 50 mM ammonium chloride, and permeabilized with 0.1% [vol/vol] Triton X-100 for 5 min to allow labeling of internal cell structures. For cell surface labeling, cells were not permeabilized. All solutions were made in PBS, and antibody staining was performed for 60 min using a 1,000-fold dilution of antiserum or purified antibody at a final concentration of 1 μg/ml. Coverslips were mounted in 10% [wt/vol] Moviol 4-88, 1 μg/ml DAPI, and 25% [wt/vol] glycerol in PBS. Images were collected at a room temperature of 22°C using a microscope (Axioskop-2; Carl Zeiss MicroImaging, Inc.) with a 63× plan Apochromat oil immersion objective (Carl Zeiss MicroImaging, Inc.) of NA 1.4, standard filter sets (Carl Zeiss MicroImaging, Inc.), a 1,300 × 1,030-pixel cooled CCD camera (CCD-1300-Y; Princeton Instruments), and Metavue software (Visitron Systems). Images were cropped in Photoshop 7.0 (Adobe) or CS2 software (Adobe) without contrast or other adjustments, sized, and placed using Illustrator 11.0 (Adobe) or CS2. Fig. S1 shows that the interaction specificity of TBC1D11/GAPCenA with human Rab GTPases involves regions outside of the TBC domain. Fig. S2 shows the effect of Rab GAPs blocking Shiga toxin uptake on Golgi morphology. Fig. S3 shows the effect of Rab GAPs blocking Shiga toxin uptake on the Rab6 effector Bicaudal-D1. Fig. S4 shows that RabGAP-5 interacts with Rab5A-C but not other Rabs. Fig. S5 shows the specific depletion of target Rabs using RNAi. Online supplemental material is available at .

Mitochondria have been recently established as both physiological targets and relay points in intracellular Ca signaling, contributing to a spectrum of cellular events ranging from oxidative ATP generation (; ; ) to apoptotic cell death (; ). This versatile function of mitochondria depends on the generation of mitochondrial matrix [Ca] ([Ca]) signals. The [Ca] signal results from activation of the uniporter-mediated Ca uptake that shows a relatively low Ca affinity (; ). The mitochondrial Ca uptake of the IP receptor (IP3R)–mediated Ca release is facilitated locally by the high cytoplasmic [Ca] ([Ca]) microdomains around the IP3Rs at focal contact areas between the ER and mitochondria (). Notably, mitochondria exhibit structural and functional diversity (), and subsets of mitochondria may interact locally with other organelles (; ). The local [Ca] control between IP3Rs and mitochondria seems to occur at stable sites between the ER and mitochondria () and displays a “quasisynaptic” organization (). Mitochondria-associated ER membranes are also involved in multiple mechanisms of joint operation between the two organelles, in the synthesis of the mitochondrial cytochrome oxidase () and phospho- and glycosphingolipids (). The existence of physical links between ER and mitochondria have been suggested based on cosedimentation of ER particles with mitochondria and electron microscopic observations of close associations between mitochondria and ER vesicles (; ; ). Recently, several mitochondria or ER bound proteins have been shown to be important for maintaining the spatial relationship between ER and mitochondria and, hence, have also been implicated as possible linking elements: DLP-1/DRP1-1 (; ), tumor autocrine motility factor receptor (), and PACS-2 and BAP31 (). IP3Rs have also been postulated to interact with the Voltage-dependent anion-selective channel to form an ER–mitochondria Ca tunnel (). Heterogeneity in the distance between the interfacing ER and outer mitochondrial membranes (OMMs; ) also indicates that the contact formation may depend on several factors and raises the intriguing possibility that the ER–mitochondria distance may be controlled to affect ER and mitochondria function. However, despite the attention paid to the structural basis of the ER–mitochondria communication, the fundamental question of whether direct physical linkage between ER and mitochondria is required for the local [Ca] coupling remains to be elucidated. Here, we visualize the ER–mitochondria tethers and show that the local Ca coupling can be weakened and strengthened by demolition and enforcement of the interorganellar protein linkage, respectively. Furthermore, our data reveal a novel regulatory role of the ER–mitochondria gap width in Ca signaling and in cell survival. To directly visualize the structures responsible for the physical association of the ER with mitochondria, we used electron tomography (ET), which can reveal fine structural details missed in conventional micrographs because of overlapping densities (). Tomographic analysis of isolated rat-liver mitochondria (conventionally fixed, plastic-embedded or unfixed, frozen-hydrated) show narrow particles connecting the OMM to putative ER vesicles ( and Fig. S1 A, available at ). These “tethers” tend to occur in clusters of six or more, spaced 13–22 nm apart, spanning intermembrane distances of 6–15 nm with indications of increments occurring in 5-nm steps (Table S1). Electron micrographs and tomograms of plastic-embedded liver (not depicted) and DT40 cells () indicate numerous regions of close association between mitochondria and both rough and smooth ER, but the noisy background (due to particle crowding) made tethers more difficult to detect than in isolated fractions. Tethers connecting OMM and smooth ER in situ (unpublished data) have lengths (9–16 nm) similar to those between attached vesicles and OMM of isolated mitochondria (Table S1). Spacings between OMM and rough ER in situ begin at 20 nm (the minimum distance to accommodate ribosomes), and the measured tether lengths are 19–30 nm (Table S1). Of the six tethers detected in the DT40 tomogram of (B–F), three appear to terminate at ribosomes on the ER. The DT40 cells used were IP3R triple knockout cells (IP3R-TKO), chosen to assess the role of IP3R in the interorganellar coupling. In electron micrographs of DT40 cells, the IP3R-TKO have ER–mitochondria associations similar to wild-type cells (Fig. S1, B–E), suggesting that an IP3R-independent linkage exists between ER and mitochondria. In summary, ET has revealed direct physical links of varying length between smooth and rough ER and mitochondria, both in normal tissue and IP3R knockout cells. Because IP3Rs are present to mediate Ca release at both smooth and rough ER, multiple coupling elements may be relevant for Ca signal propagation from ER to mitochondria. The heterogeneity in tether lengths points to an enticing new possibility, that ER–mitochondria communication may be controlled by varying the interorganellar distance. To test the functional significance of the tethers, we designed strategies for weakening and enhancing the physical coupling. To disrupt the ER–mitochondria physical coupling, limited proteolysis was used. Confocal images of isolated liver mitochondria preparations showed abundant overlapping immunoreactivity for both IP3R (type 1 and 2) and cytochrome oxidase, an enzyme of the inner mitochondrial membrane (), indicating that the IP3Rs reside in mitochondria-associated ER. When this preparation was trypsinized (40 μg/ml for 150 s followed by addition of soybean trypsin inhibitor [SBI] at 250 μg/ml) and recentrifuged, the IP3R immunoreactivity disappeared () and was recovered in the light membranes (not depicted). The IP-sensitive Ca store was also quantified by measurement of the IP + thapsigargin (Tg)-induced Ca release in the 10,000-g pellet (ER–mitochondria complex) and supernatant (ER only) of both the control and trypsin-pretreated liver mitochondria (). Trypsin pretreatment caused a twofold increase in Ca release in the supernatants and a significant decrease in the pellets (). Similar findings were obtained in RBL-2H3 cells (Fig. S2, A and B, available at ). Collectively, these data suggest that limited proteolysis disrupted the physical coupling between ER and mitochondria. We next evaluated the effect of proteolytic treatment on mitochondrial Ca signaling. IP3R-mediated Ca release effectively supports mitochondrial Ca uptake in permeabilized RBL-2H3 cells (). Suspensions of digitonin-permeabilized cells were treated with proteinase K, another serine protease (20 μg/ml for 150 s), and [Ca] and [Ca] were simultaneously monitored fluorometrically. In control cells, 8 μM IP evoked an abrupt increase in [Ca] that was paralleled by a rapid and substantial increase in [Ca] (, left). In the proteinase K–pretreated cells, the IP-induced [Ca] increase was preserved, but the [Ca] increase was practically eliminated ( and Fig. S2 C). Because proteinase K treatment did not affect IP-induced Ca release (either in the presence or absence of mitochondrial uncouplers) and failed to inhibit the mitochondrial uptake of directly added Ca ( and Fig. S2 C), it is likely that proteinase K inhibited the transfer of released Ca from IP3Rs to the mitochondria. Similar data were obtained when trypsin was used instead of proteinase K (). Consistent with earlier reports, trypsin-digested preparations retained IP-induced Ca release () but almost completely lost the IP-induced [Ca] signal, an effect that was prevented by SBI (, middle). Trypsin also failed to inhibit the [Ca] increase induced by elevation of the bulk [Ca] by addition of 10 μM CaCl (; initial rates were 0.96 ± 0.08 for control and 1.25 ± 0.17 μM/s for trypsin, respectively; = 3). The trypsin dose dependence and time course data (Fig. S2 C) further illustrate that the Ca transfer from IP3Rs to the mitochondrial matrix is very sensitive to trypsinolysis, whereas the Ca release or mitochondrial Ca uptake by itself is hardly inhibited. Thus, limited proteolysis disrupts the link between ER and mitochondria and suppresses the propagation of the IP3R-mediated Ca release to the mitochondria. To tighten the physical coupling between ER and mitochondria, we created a construct that encodes monomeric red fluorescent protein (mRFP) fused to the OMM targeting sequence of mAKAP1 at the N terminus and fused to the ER targeting sequence of yUBC6 at the C terminus (mAKAP1 [34–63]-mRFP-yUBC6, OMM–ER linker). Based on the size of the fluorescent protein (4.2 × 2.4 nm), the maximal length of this construct is <5 nm. As a control, the above construct was also prepared without the ER targeting sequence (mAKAP1 [34–63]-mRFP). Cells expressing the constructs showed red fluorescence localized to the mitochondria and displayed mitochondrial aggregation in some cells (unpublished data). To analyze the ER–mitochondria interface, the transfected cells were sorted and prepared for transmission EM (TEM). The mAKAP1(34–63)-mRFP transfected cells displayed numerous associations between ER and mitochondria, but the interface area only involved a small fraction of the mitochondrial perimeter (, left) similar to the situation in nontransfected cells (). The ER–OMM distance showed bimodal distribution, having the most frequent values at 10–15 and 25–30 nm, similar to the results of the ET analysis. In contrast, in the OMM–ER linker transfected cells, the ER formed a cap over large mitochondrial areas and the cleft between the ER membrane and the OMM was extremely narrow (, right). On average, the ER–mitochondria distance at these sites decreased from 24 ± 3 to 6 ± 1 nm and the interface area increased fourfold in the presence of the OMM–ER linker (). Thus, expression of the OMM–ER linker caused the associations to become tighter and the interface area to increase. To evaluate the effect of the enhanced physical coupling between the organelles on the Ca transport, we conducted imaging of [Ca] and [Ca] in permeabilized cells expressing either mAKAP1(34–63)-mRFP-yUBC6 or mAKAP1(34–63)-mRFP and the Ca probe, ratiometric pericam targeted to the mitochondrial or nuclear matrix. Synchronous Ca release evoked by maximal IP induced comparable [Ca] and [Ca] responses in both OMM–ER linker and control cells (unpublished data). However, a different picture emerged when gradual Ca liberation through IP3Rs was stimulated by adenophostin A (; ). The nuclear matrix [Ca] rise that closely follows the [Ca] signal was unaffected in the OMM–ER linker–expressing cells, but the [Ca] elevation was significantly enhanced (, top). In particular, the delay between the [Ca] and [Ca] elevation was shortened (, left). Thus, overexpression of the linker enabled mitochondrial Ca uptake during Ca mobilization conditions that normally are recognized by the mitochondria with low efficiency. Based on the effects of the synthetic linker, the quantity and the length of the tethers may exert a control on the ER–mitochondria Ca coupling. Bringing the ER closer to mitochondria by the physiological tethers effectively supports local Ca signaling. However, sustaining a gap between the organelles by the tethers may also bear significance for other cell functions. A too-close association between ER and mitochondrial membranes might cause continuous mitochondrial Ca uptake during background Ca release and, in turn, could facilitate mitochondrial Ca overloading and membrane permeabilization (; ). Also, the extent of the anchorage of ER to the mitochondria is relevant to the motility of the mitochondria, which allows dynamic redistribution of the mitochondrial ATP production and Ca buffering throughout the cell (; ). To test the idea that tightening of the ER–mitochondria coupling may affect the cells' ability to respond to challenges, RBL-2H3 cells expressing the ER–OMM linker were exposed to Tg, which gradually mobilizes the ER Ca store and, in turn, stimulates the store-operated Ca entry (as a control, either the mitochondria- or the ER-targeted part of the linker was overexpressed). In the cells expressing the OMM–ER linker, the [Ca] signal showed an initial elevation followed by a partial decay to a plateau. After a longer period of time, a gradual elevation appeared turning to a steep and robust [Ca] rise (, red). This second [Ca] rise began at different time points in the individual cells, causing a more gradual rise in the mean response (, bottom). In control cells, the first [Ca] rise was similar to those in the OMM–ER linker cells; however, the second [Ca] rise developed much more slowly (, black). Analysis of [Ca] signals in single cells showed an early onset of both the gradual [Ca] elevation and the steep and robust [Ca] increase (, right) in the cells expressing the OMM–ER linker. The second [Ca] elevation was prevented by the addition of either 5 μM carbonyl cyanide p-trifluoromethoxyphenylhydrazone and 2.5 μg/ml oligomycin or by 5 μM cyclosporin A, a drug interfering with the Ca-dependent activation of the permeability transition pore, suggesting that it depended on mitochondrial Ca uptake and was a result of Ca release from Ca-overloaded mitochondria (unpublished data). Hence, mitochondria were susceptible to Ca overloading and permeabilization in cells where the ER–mitochondria coupling was tightened by the OMM–ER linker. The mitochondrial Ca dysregulation was regularly followed by detachment of the cells, indicating the loss of viability. Because Ca transfer to the mitochondria is a key step in induction of many forms of cell death, we reasoned that tightening of the ER–mitochondria coupling may contribute to the execution of the cells induced by certain proapoptotic stimuli. To this end, RBL-2H3 cells were exposed to apoptotic conditions (serum starvation and tunicamycin treatment) and were fixed for EM at 24 h, before the onset of cell detachment. Analysis of the dimensions of the ER–mitochondria interface showed shortening of the mean distance between ER and mitochondria in both the serum-starved and tunicamycin-treated cells and an increase in the frequency of tight associations (<6 nm distance; ). The high incidence of the tight associations could not be attributed to the condensation of the apoptotic cells because the perimeter or area of the cell cross sections has not been altered yet. These results suggest that narrowing of the ER–mitochondria gap occurs in intact cells and may be an important step in the execution of some apoptotic mechanisms. The scheme in illustrates the novel aspects of the ER–mitochondria signaling uncovered in the present work. The association between ER and mitochondria is due to the presence of tethers that link both smooth and rough ER to the mitochondria. The length of the tethers displays some diversity, giving rise to varying distances between ER and mitochondria. In response to apoptotic agents the ER–mitochondria gap narrows, indicating dynamic regulation of the interorganellar junction. In healthy cells, the ER–mitochondria tethering ensures the propagation of IP3R-linked Ca signals to the mitochondria to coordinate ATP production with the stimulated state of the cell and to enable the mitochondrial Ca buffering. However, the gap between the organelles is sufficiently wide to isolate mitochondria from the slow Ca leakage from the ER. Relaxing the ER–mitochondria coupling suppresses the Ca signal propagation to the mitochondria, putting at risk the Ca-dependent control of mitochondrial metabolism. In contrast, tightening of the coupling invokes mitochondria in the handling of Ca under resting conditions, sensitizing mitochondria to Ca overloading and leading to permeabilization and committing the cells to a cell death pathway. Tightening of the connections seems to be relevant for several mechanisms of cell death. Thus, these results reveal an unexpected dependence of cell function and survival on the maintenance of a proper spacing between the ER and mitochondria. To construct the OMM–ER linker, mRFP was targeted to the ER by using the C-terminal ER localization sequence of the yeast UBC6 protein (X73234, residues 233–250: MVYIGIAIFLFVGLFMK), through the linker (SGLRSRAQASNSRV; ). This construct was complemented with the N-terminal mitochondrial localization sequence of the mouse AKAP1 protein (V84389, residues 34–63: MAIQLRSLFPLALPGLLALLGWWWFFSRKK) with the linker (DLELKLRILQSTVPRARDPPVAT). Ratiometric pericam targeted to the mitochondrial or nuclear matrix was provided by A. Miyawaki (Institute of Physical and Chemical Research, Wako-city, Japan). RBL-2H3 cells were cultured as described previously (). Cells were transfected with cDNA by means of electroporation in suspensions (4.5 × 10 cells + 20 μg of each cDNA in 250 μl medium). Electroporation was performed in a BTX-830 square-pulse generator in a 4-mm gap cuvette using a single 250-V 13-ms pulse. For FACS sorting (MoFlo FACS sorter [DakoyCytomation] equipped with a 488-nm laser), 8.5–12.5 × 10 cells transfected with a construct of interest and EGFP were cultured for 24 h. DT40 (wild type and IP3R knockouts alike were a gift from T. Kurosaki, Kansai Medical University, Moriguchi, Japan) cells were cultured in suspension, in RPMI 1640 with glutamine supplemented with penicillin/streptomycin, 2 mM -glutamine, 10% heat inactivated FCS, and 1% chicken serum (Invitrogen) in 5% CO and 95% air at 40°C. Experiments were performed as described earlier (). Cells grown overconfluent in tissue culture flasks (∼7 × 10 cells) were loaded with fura2FF/AM, harvested using trypsin/versene, and washed with Na-Hepes/EGTA. All further steps were performed at 4°C. The cells were exposed to hyposmosis for 10 min (14 ml intracellular medium [ICM; 120 mM KCl, 10 mM NaCl, 1 mM KHPO, 20 mM Tris-Hepes, 2 mM MgATP, and 1 μg/ml each of antipain, leupeptin, and pepstatin, pH 7.2]) diluted fivefold with dHO and supplemented with 200 μM EGTA and 5 mM MgCl). Subsequently, the cells were homogenized in a dounce glass/glass homogenizer (30–35 strokes, tight pestle). To restore osmolarity, 3 vol of 100% ICM supplemented with 125 mM sucrose, 200 μM EGTA, and 5 mM MgCl was added. To eliminate unbroken cells and nuclei, the homogenate was centrifuged at 1,000 for 10 min. The supernatant was further centrifuged at 10,000 for 15 min, and the pellet (mitochondrial fraction) was resuspended in 400–500 μl ICM plus protease inhibitors and 10 μM EGTA and stored on ice. Attachment to CellTak (BD Biosciences) coated coverslips was performed at room temperature for 5 min in the presence of 2 mM MgATP in 25–50 μl vol. The protocol was adapted from . The liver of a 350–400-g normal male Sprague-Daily rat was perfused with ∼200-ml Na-Hepes/EGTA and was removed. All the further steps were done at 4°C. The liver was cut up to small pieces with scissors and washed with ICM. After determination of the wet weight, a 1:4 homogenate was prepared in 350 mM sucrose containing 2.5 mM magnesium acetate and 10 mM Tris maleate, pH 7.4. Homogenization was performed in a 60-ml glass-Teflon homogenizer (11 strokes at 900 rev/min). The homogenate was filtered through two layers of sterile gauze and once more through one layer of Miracloth (Calbiochem). The mitochondrial fraction was obtained by centrifuging the supernatant of the 900- (10 min) fraction at 8,000 for 15 min. Particles were attached to CellTak-coated coverslips as described above in the previous paragraph. The membrane fractions attached to coverslips were fixed in 3% paraformaldehyde. A monoclonal anti-human cytochrome oxidase complex IV subunit 1 antibody (Invitrogen) was used to visualize mitochondria and polyclonal anti–IP3R-1 and -2 antibodies (Affinity BioReagents, Inc.) were used to visualize the IP3Rs. The secondary antibodies were fluorescently labeled (Alexa Fluor 488 and 568). Images were acquired using a confocal system (Radiance 2001; Bio-Rad Laboratories), and colocalization was evaluated using Lasersharp software (Bio-Rad Laboratories; ). A 20–25-μl aliquot of the crude mitochondrial fraction was transferred to 800 μl ICM supplemented with 1.5 μM fura2/FA, 2 mM Mg-ATP, 2 mM succinate, and protease inhibitors in a stirred cuvette at 35°C. Ratiometric recording of fura2 fluorescence was performed as described for the permeabilized cells. RBL-2H3 cells or mitochondrial fractions attached to coverslips were placed in 1 ml buffer to the heated stage (35 C°) of a microscope (IX70 [Olympus]; 40×; UApo340) connected to a cooled charge-coupled device camera (PXL; Photometrics). Ratiometric imaging of fura2FF and pericam was used to monitor [Ca] and [Ca] as described previously (; ). For embedding, a standard protocol was used (). Ultrathin sections for TEM were poststained with UA and sodium bismuth (). The sections were examined with either a scanning transmission electron microscope (model 7000; Hitachi) or a digital transmission electron microscope (Tecnai 12; Philips) driven by Gatan software. For cryo-EM of isolated mitochondria, 3–5-μl aliquots of mitochondrial suspensions (10–20 mg/ml in 0.225 M manitol and 0.075 M sucrose) containing 10-nm colloidal gold particles were deposited on freshly glow-discharged 300-mesh copper grids with holey carbon films. Grids were blotted with filter paper and immediately plunged into liquid ethane cooled by liquid nitrogen. Tilt series were collected over an angular range of ±60° at 2° intervals (total dose ∼24 electrons/Ǻ) using a transmission electron microscope (JEM-4000FX; JEOL) equipped with Gatan cryo-transfer unit and a TVIPS 1024 × 1024 cooled charge-coupled device camera. Images were aligned and tomographic reconstructions calculated as previously described (), using the weighted back-projection method as implemented in the SPIDER image processing system (). 3D models were generated by density thresholding using Iris Explorer (Numerical Algorithms Group) or surface rendering in Iris Explorer after manual membrane tracing in Sterecon (). In the case of plastic sections, the z dimension (section thickness) of the final models was increased by 20% to compensate for radiation-induced thinning of the plastic section. Lengths of tethers connecting mitochondrial outer membranes and ER membranes were determined using NIH ImageJ. Table S1 shows the dimensions of the ER–mitochondria interaction areas. Fig. S1 shows tight ER–mitochondria associations in quick frozen and chemically fixed isolated liver mitochondria and in wild-type and IP3R-TKO DT40 cells. Fig. S2 demonstrates protein linkage between ER and mitochondria in RBL-2H3 cells. Online supplemental material is available at .

The Rab family of small GTP-binding proteins is responsible for the spatial and functional organization of intracellular compartments and controls vesicular transport between organelles in eukaryotic cells (; ; ). The principles of Rab function as regulatory GTPases and how they control downstream processes through Rab effectors are well understood (; ). Details are emerging on the coordination of the action of Rab GTPases when they function within a contiguous organelle (; ; ). Less is known about the higher level organization when a complement of Rabs functions within a complex multistage trafficking pathway. It has been reported that maturation from early to late endosomes involves a novel process of abrupt, synchronous Rab5–Rab7 replacement on an entire early endosomal organelle (). This process, which is termed Rab conversion, ushers the maturation of an early endosome into a late endosomal organelle (). A functionally similar process of Rab handover seems to operate within the biosynthetic pathway in yeast (). The events that govern endosomal Rab conversion are only beginning to be elucidated (). In this context, phagolysosome biogenesis provides a convenient and morphologically tractable model system (). Maturing phagosomes closely mirror trafficking events observed within the endocytic pathway (; ; ). A phagosome, when it normally matures into a phagolysosome, undergoes a transition between the stages marked by Rab5 and Rab7 (; ). The switch between Rab5 and Rab7 on a phagosome correlates with functional changes from an organelle with early endosomal characteristics to a compartment with lysosomal, degradative properties. When taken up by the phagocytic cell, can arrest phagosomal maturation and prevent phagolysosome biogenesis (; ), providing an advanced model system to study the role of Rabs and their effectors in phagosome maturation. Initially, Rab5 was identified as one of the low molecular weight GTP-binding proteins present on mycobacterial phagosomes (), leading to the identification of Rab effectors () involved in mycobacterial phagosome maturation arrest. Rab7 is excluded from the phagosome (), indicating that mycobacterial phagosomes do not undergo Rab5–Rab7 conversion. We report that Rab22a, which is a member of the group V Rabs (), is a key Rab, accumulating on mycobacterial phagosomes and precluding their acquisition of Rab7 and maturation into phagolysosomes. We investigated the members of the group V Rabs (), Rab5, Rab21, and Rab22a, by live microscopy using a previously published approach (; ). The entry of a phagocytosed particle was identified as previously described (), and live imaging was initiated to record EGFP-Rab dynamics on nascent and maturing phagosomes containing latex beads (), followed by quantification using ratiometric analysis of intensities comparing phagosome and cytosol fluorescence values (R; ; ; ). The majority of the group V Rabs were transiently recruited to latex bead phagosomes (), with Rab5 and Rab21 () desorbing from the phagosomes by 10 min after the uptake (), and with Rab22a showing a diminutive initial peak (). Because of a very low-level EGFP-Rab22a recruitment to latex bead phagosomes, we wondered whether macrophages expressed Rab22a. Rab22a expression was confirmed by RT-PCR (Fig. S1 A, available at ). In addition, endogenous Rab22a was detected in macrophages by immunofluorescence (Fig. S1 B). Hence, low levels of Rab22a on latex bead phagosomes cannot be explained by a lack of Rab22a expression in macrophages. Thus, the group V Rabs are transiently recruited in small amounts to latex bead phagosomes during early time points after phagocytosis. We next tested group V Rab dynamics on mycobacterial phagosomes. Rab5 and Rab21 followed similar kinetics on both mycobacterial and latex beads phagosomes (). However, mycobacterial phagosomes displayed a marked difference relative to latex bead phagosomes by recruiting and retaining high quantities of Rab22a (). Enumeration of Rab5, Rab21, and Rab22a profiles () confirmed that Rab22a was persistently accumulating on mycobacterial phagosomes. This was accompanied by diminishing levels of EGFP-Rab22aWT fluorescence in other parts of the cell, a phenomenon that was augmented in macrophages infected with more than one bacillus. EGFP-Rab22aWT–positive profiles were observed to tether and fuse with mycobacterial phagosomes, increasing EGFP-Rab22aWT levels on these organelles (Video 1, available at ). These observations were confirmed by immunofluorescence detection of endogenous Rab22a on bacillus Calmette-Guérin (BCG) phagosomes (Fig. S1 B). The differential distribution of EGFP-Rab22aWT was not caused by phagosome size difference because 3-μm latex beads (Fig. S1 C) behaved similarly to the 1-μm beads. Thus, Rab22a is specifically enriched on phagosomes containing mycobacteria. Rab22a has been implicated in early endosomal and recycling pathways in nonphagocytic cells (; ). We tested Rab22a localization in macrophages and found that both EGFP-Rab22aWT and endogenous Rab22a overlapped with the early endosomal marker EEA1 (Fig. S1, D and E). This is in keeping with the reported early endosomal localization of Rab22a in other cells (). Immunofluorescence analysis using GM130, syntaxin 6, and TGN38 showed that in macrophages Rab22a was not on Golgi organelles (Fig. S1 F), and Golgi vesiculation did not occur in cells transfected with EGFP-Rab22aQ64L (Fig. S1 F), in contrast to a report that the Rab22a mutant vesiculates Golgi in CHO cells (). The early endosomal localization of Rab22a, and the increased fusion of early endosomal organelles with mycobacterial phagosomes stimulated by Rab14 (unpublished data), may partially explain Rab22a enrichment on BCG phagosomes. We examined whether expression of constitutively active Rab22a (EGFP-Rab22aQ64L) affected phagosomes harboring dead mycobacteria. Heat inactivation of incapacitates it to block phagolysosome biogenesis (; ; ). The constitutively active mutant of Rab22a accumulates on mycobacterial phagosomes (Fig. S1 G) in a manner similar to wild-type Rab22a. Expression of EGFP-Rab22aQ64L inhibited maturation of phagosomes containing dead BCG, as follows: (a) heat-killed BCG phagosomes showed reduced colocalization with the acidotropic dye LysoTracker Blue (), indicating impaired acidification; (b) phagosomes with dead BCG showed lower proteolytic activity in macrophages transfected with EGFP-Rab22aQ64L, as indicated by lower staining with DQ-Red BSA, which is an endocytic protease substrate whose fluorescence dequenches upon proteolysis (); (c) although phagosomes harboring dead mycobacteria normally do not retain transferrin receptor (), transfection with EGFP-Rab22aQ64L caused significant presence of transferrin receptor on dead mycobacterial phagosomes ( and Q); (d) A similar effect was observed with another previously mapped () early/recycling endocytic marker, syntaxin 13 (); and (e) Rab11, a GTPase that controls transferrin receptor (TfR) recycling (; ), accumulated on dead mycobacterial phagosomes in cells expressing EGFP-Rab22aQ64L when compared with control untransfected cells (34 ± 5% vs. 14 ± 4% colocalization; P = 0.04; Fig. S1 H), indicating that the constitutively active mutant of Rab22a conferred recycling endosomal characteristics upon dead mycobacterial phagosomes. In lieu of experiments with a dominant-negative Rab22a, which was found to cause macrophage detachment, we resorted to siRNA knockdown of Rab22a and examined its effects on maturation of phagosomes harboring live variant BCG. Rab22a knockdown () caused a threefold increase in the colocalization of live mycobacterial phagosomes with the most robust late endocytic marker CD63 (, B and E; ), indicating that live mycobacterial phagosomes were maturing into phagosomes with late endosomal characteristics. Live mycobacterial phagosomes also showed a doubling of V HATPase association with phagosomes containing live mycobacteria (; ; ). Rab22a knockdown did not cause indiscriminate mixing of endosomal markers (Fig. S2, A–C, available at ). Furthermore, colocalization of TfR with live mycobacterial phagosomes was diminished upon Rab22a siRNA knockdown compared with scrambled siRNA control (). The effects of Rab22a knockdown with SMARTpool (Dharmacon) Rab22a siRNA (a combination of four Rab22a-specific siRNA duplexes) was confirmed using individual siRNA duplexes (Fig. S2 D), which also caused an increase in live mycobacterial phagosome maturation (Fig. S3, A–D, available at ). Furthermore, transfection with siRNA against the closely related Rab22b, which is also expressed in macrophages (Fig. S2, E and F), did not alter mycobacterial phagosomes (Fig. S3, A–D). The effects of Rab22a knockdown on mycobacterial survival (Fig. S3, E–G) were mild within the period investigated, and although a trend was observed, no statistically significant differences could be established (Fig. S3, F and G). We conclude that Rab22a is necessary to maintain phagosome maturation block, but that maturation block override does not automatically translate into direct bacterial elimination by macrophages, in keeping with the early observations by . A prequel to endosomal maturation into late endosomal/lysosomal organelles is Rab conversion (). This term describes a process whereby an organelle synchronously sheds off early endosomal Rab(s) and concomitantly receives the late endosomal Rab, Rab7 (). The signals for this transition are currently unknown (). We wondered whether Rab conversion applies to phagosomes, and whether Rab22a, as a candidate terminal recycling Rab involved in cargo and membrane sorting from the early endosome (; ), could supply or contribute to such signals. To test this, we examined Rab7 acquisition by the mycobacterial phagosome, which was previously shown to exclude this critical late endocytic Rab (). Unlike in cells treated with control scrambled siRNA, Rab7 acquisition was increased to 80% on live mycobacterial phagosomes in macrophages in which Rab22a was knocked down by siRNA (). These findings are consistent with a functional role for Rab22a in mycobacterial phagosome maturation block. More generally, the conversion of live mycobacterial phagosomes into the Rab7 stage upon Rab22a knockdown suggests that Rab22a supplies signals preventing acquisition of Rab7 and precluding organellar maturation into a late endosomal/lysosomal compartment. Thus, Rab22a functions not only as a recycling Rab involved in cargo and membrane sorting from the early endosome (; ), but it also acts as a coordinator of Rab succession. In keeping with our findings with phagosomes, there is a lack of endosomal EGF degradation in Rab22aQ64L-transfected Hep2 cells (). In CHO cells, the expression of Rab22aQ64L causes endocytic tracers to remain in Rab22a-positive vesicles (). We propose that Rab22a is a central regulator of the transition to late endocytic organelles by signaling “all clear” and allowing the leftover sorting endosomal or phagosomal organelle to transit from an early compartment to a degradative organelle controlled by Rab7. RAW264.7 macrophages and variant BCG were maintained as previously described (). Mycobacteria were heat killed by incubation at 90°C for 5 min before labeling. Both live and dead mycobacteria were labeled with 0.5 mg/ml Texas red–succinimidyl ester and prepared as previously described (). Dead mycobacteria were also labeled with 0.25 mg/ml Alexa Fluor 647–succinimidyl ester in PBS for 1 h. Streptavidin-conjugated 1-μm polystyrene beads (Sigma-Aldrich) were labeled and prepared as previously described (). 3-μm polystyrene beads were opsonized in DME supplemented with 10% FBS before use. The plasmids pEGFP-Rab21WT, pEGFP-Rab22aWT, and pEGFP-Rab22aQ64L were obtained from J. Donaldson (National Institutes of Health, Bethesda, MD), pEGFP-hRab5WT was obtained from P. Stahl (Washington University, St. Louis, MO), and pEGFP-Rab7WT was obtained from A. Wandinger-Ness (University of New Mexico, Albuquerque, NM). For transfection, 5 × 10 RAW264.7 cells were resuspended in a nucleoporator buffer supplied by the manufacturer (Amaxa Biosystems) with 5 μg of plasmid DNA. Cells were nucleoporated according to the manufacturer's protocol and allowed to express the construct for 24 h before the imaging experiments. Rabbit polyclonal antibody to Rab22a was obtained from J. Donaldson, and polyclonal antibody to syntaxin 13 was obtained from R. Scheller (Genentech, South San Francisco, CA). Rabbit polyclonal antibodies to transferrin receptor and CD63 were purchased from Santa Cruz Biotechnology, Inc. Antibody to V was used as previously described (,). Mouse monoclonal antibody to transferrin receptor was purchased from Zymed Laboratories. Monoclonal antibodies against GAPDH, GM130, and TGN38 were obtained from Abcam. Lysotracker Blue, DQ Red BSA, and secondary antibodies conjugated to Alexa Fluor 488, 568, and 647 were purchased from Invitrogen. The acidotropic dye Lysotracker Blue was diluted in DME (1:10,000) and preloaded into macrophages for 2 h. DQ Red BSA was preloaded at 10 μg/ml for 3 h before infection. Cells were subsequently fixed and viewed using immunofluorescence microscopy. Rab22a and Rab22b knockdowns were achieved by using siGENOME SMARTpool reagent (Dharmacon) specific for Rab22a and Rab22b (Dharmacon). All effects of Rab siRNAs were compared with siCONTROL Nontargeting siRNA pool (Dharmacon), which is labeled as scrambled siRNA in figures. RAW264.7 cells were transfected with 1.5 μg siRNA by nucleoporation. Immunoblotting (30 μg of total protein) was performed as previously described (,). GAPDH immunoblotting was used as a loading control. Rab22a single siRNA duplexes used individually were as follows: duplex 1, sense (CAGCAGCCAUCAUCAUCGUUUAUU) and antisense (5′-PUAAACGAUGAUGAUGGCUGCUGUU); duplex 2, sense (GGGAACAAGUGCGAUCUUAUU) and antisense (5′-PUAAGAUCGCACUUGUUCCCUU); duplex 3, sense (GAGAUUAGUCGAAGAAUUCUU) and antisense (5′-PGAAUUCUUCGAAGAAUUCUU); and duplex 4, sense (GGAUACGGGUGUGGGUAAAUU) and antisense (5′-PUUUACCCACACCCGUAUCCUU). Imaging of 1-μm-thick optical sections was performed using an Axiovert 200M microscope with an Axioscope 63× oil objective and LSM 5 Pascal or LSM 510 META systems (Carl Zeiss MicroImaging, Inc.). At least 200 phagosomes from three independent experiments were analyzed for colocalization studies. A rotating disk confocal microscope (UltraView; PerkinElmer) that affords low photocytotoxicity and low photobleaching was applied for 4D imaging, as previously described (; ). For ratiometric quantitative analysis (; ) of a volume over time, z sections were collapsed into a single projection according to the published procedure (). Transfected RAW264.7 cells were synchronously infected by centrifugation of bacteria or beads onto macrophages adherent to coverslips at 1,000 rpm for 5 min. Coverslips were mounted into a perfusion chamber (Harvard Apparatus) set at 37°C. Identification of mycobacterial entry and image acquisition was performed as previously described (). To measure R, fluorescence intensity of the phagosomal membrane was divided by background cytosolic fluorescence. RAW264.7 macrophages were seeded at 2.0 × 10 cells/well in 12-well plates after transfection with either siGENOME SMARTpool Rab22a siRNA or scrambled siRNA. Cells were incubated for 24 h. Macrophages were infected with live variant BCG or live H37Rv (preincubated for 30 min at 37°C in DME), at a nominal multiplicity of infection of 10, followed by four washes using complete DME. Macrophages were hypotonically lysed using cold sterile water after a 2, 4, or 24 h chase period. Mycobacteria were plated for colony forming units on Middlebrook 7H11 agar (Difco) and incubated at 37°C for 2.5 wk. Bacterial viability was expressed as percentage of survival relative to scrambled siRNA control. Experiments were performed in triplicate. Results are from experiments performed in triplicate. All statistical analyses were calculated using Fisher's protected least significant difference post hoc test (analysis of variance, ANOVA) (SuperANOVA 1.11; Abacus Concepts). P values of ≤ 0.05 were considered significant. Fig. S1 shows that endogenous Rab22a and EGFP-Rab22aQ64L in macrophages is recruited to mycobacterial phagosomes and that Rab22a colocalizes with early endosomes, but not Golgi organelles. Fig. S2 shows an analysis of Rab22a knockdown effects on early and late endosomes, characterization of single duplex Rab22a siRNA knockdowns, and expression of Rab22b. Fig. S3 shows the effects of single-duplex siRNA Rab22a knockdown on mycobacterial phagosomal maturation and Rab22a knockdown on intracellular survival of mycobacteria. Video 1 shows an EGFP-Rab22aWT–transfected macrophage infected with live Texas red–labeled variant BCG.

Intraflagellar transport (IFT) of particles containing structural and signaling proteins is critical to building and maintaining cilia and flagella (; ; ). Defects in IFT in humans can give rise to dysfunctional cilia and produce a variety of disease states (). Sensory cilia in neurons, which are crucial for chemosensory function (), also rely on IFT for their formation and function (). Anterograde IFT transport in sensory neurons is driven by two Kinesin-2 family members: heterotrimeric Kinesin-II and homodimeric OSM-3 (; ; ). Retrograde IFT transport is driven by a special class of dynein molecules that recycle the anterograde motor and proteins destined for turnover back to the cell body (; ). The activities of the heterotrimeric and homodimeric Kinesin-2 motor appear to be carefully regulated during IFT. Microscopy studies in living have shown that both motors cooperate to move IFT particles along the middle segment of the cilia consisting of nine doublet microtubules, whereas only OSM-3 transports IFT particles along the distal segment consisting of nine singlet microtubules (; ). OSM-3 activity is most likely confined to axonemes in because –null animals only display a loss of the distal segments of their sensory cilia. Thus, it is possible that the OSM-3 motor is active only in the cilium, where it executes its transport functions. However, the mammalian homologue of Osm-3 (KIF17) is involved in both ciliary IFT () and membrane transport in neuron dendrites (). To better understand the molecular basis of IFT transport, it is important to characterize the properties of the IFT motor proteins in vitro. Although several studies have explored the motile properties of the heterotrimeric Kinesin-2 motor in vitro (; ; ), in vitro studies of OSM-3 and single-molecule studies of KIF17 have not been undertaken. In this study, using single-molecule techniques, we show that single OSM-3 molecules do not move processively (long distance movement along microtubules without dissociation) in vitro. However, motility can be activated by attaching OSM-3 via its tail domain to beads or by mutations to a hinge region in the middle of the molecule. These results support a model in which OSM-3 is autoinhibited in a folded conformation and processive movement is activated by attachment to IFT cargo. This mechanism is similar to an intramolecular, autoinhibitory interaction in Kinesin-1 that has been proposed to keep this motor in an inactive conformation until it interacts with its cargo (for review see ). The biological importance of OSM-3 autoinhibition is suggested by the finding that constitutive activation of OSM-3 motility through a single point mutation also causes severe defects in IFT transport. Here, we have characterized the motile properties of a recombinant OSM-3 construct consisting of full-length OSM-3 (aa 1–699) with a C-terminal GFP (). As expected, OSM-3 was an active plus end–directed motor in a microtubule gliding assay, although the velocity of movement (0.3 μm/s) in this gliding assay was lower than IFT transport by OSM-3 in the distal segment (1.3 μm/s; ). of the full-length OSM-3 motor was only 4 ATP/s/head ( and Fig. S1, available at ), which is much slower than would be expected for a motor capable of moving 300–1,300 nm/s (assuming that it takes 8-nm steps per ATP hydrolyzed as shown for Kinesin-1; ). We next examined OSM-3 processivity by imaging single GFP-labeled molecules using a total internal reflection fluorescence (TIRF) microscope. In this assay, a truncated construct of Kinesin-1 (K530; ) fused to GFP exhibited numerous processive movements with a mean run length of 1.2 μm, which is similar to previous results (). In contrast, full-length OSM-3 rarely showed any processive runs (Video 1). OSM-3's lack of processivity is somewhat surprising because many dimeric motors involved in long-range transport are processive in vitro (; ; ; ). Extensive in vitro studies on the long-range, unidirectional processivity of Kinesin-1 and -3 have identified three critical determinants of processivity: the presence of a stable neck coiled coil that joins the two kinesin motor domains immediately after the neck linker (kinesin's mechanical element) (; ; ); the ability of the two motor domains to coordinate their ATPase cycles (; ; ); and the presence of intramolecular interactions that inhibit processivity (; ; ; ). We sought to test whether any of these possible mechanisms could explain the lack of processivity by OSM-3. We initially suspected that an unstable neck may underlie the lack of OSM-3 processivity because the neck coiled coil of OSM-3 is more than two heptads shorter and is predicted to be much weaker than the neck coiled coil of Kinesin-1 (). The putative OSM-3 neck coiled coil is also less positively charged than the neck coiled coil of Kinesin-1, a factor shown to enhance its processivity (). We tested the hypothesis that a weak coiled coil may be responsible for the lack of processivity of OSM-3 by fusing the putative neck coiled coil of OSM-3 (and the subsequent C-terminal stalk and tail domains) to the motor domain and neck linker of Kinesin-1 (K-O; ). We reasoned that if the OSM-3 neck was incompatible with processivity, K-O should be a nonprocessive motor. Contrary to this prediction, K-O molecules moved processively in the single molecule TIRF assay with similar velocities (0.5 ± 0.2 μm/s), albeit with reduced run lengths (0.4 μm), compared with K530 (0.7 ± 0.3 μm/s and 1.2 μm). The processivity of K-O suggests that the OSM-3 neck coiled coil is not responsible for the lack of processivity in OSM-3. We next tested whether the OSM-3 motor domain was incompatible with efficient processive motility by fusing the OSM-3 motor domain (catalytic core and neck linker) to the neck coiled coil and stalk of Kinesin-1 (O-K; ). O-K exhibited robust processive movement with a mean run length (2.6 μm) that was approximately twofold higher than that of K530. The velocity measured in the single-molecule fluorescent assay (1.5 ± 0.3 μm/s) was similar to the velocity of OSM-3 cargo in vivo (1.3 ± 0.2 μm/s), suggesting that O-K may be mimicking the in vivo function of the motor. Thus, the OSM-3 motor domain is compatible with processivity in a dimeric construct that lacks its native stalk/tail domain. Interestingly, the O-K ATPase rate ( 75 ATP/s/head; and Fig. S1) was an order of magnitude greater than that of wild-type OSM-3 ( 4 ATP/s/head). This difference in rates indicates that some element C terminal to the motor domain of OSM-3 is inhibiting its catalytic activity and likely the processivity of the motor as well. To test whether the tail domain might inhibit the processivity of OSM-3, we made a construct of OSM-3 lacking the tail domain (OSM-3 aa 1–555) but found that it was an unstable dimer under conditions of the TIRF assay (low nanomolar range; unpublished data). An alternative strategy for investigating a possible autoinhibition of OSM-3 processivity was suggested by a comparison with Kinesin-1. The processivity of Kinesin-1 is inhibited by an autoinhibitory interaction between its neck coiled coil and tail domains (; for review see ). The interaction of these distant N-and C-terminal elements is achieved by folding about a flexible hinge in the stalk (, H2). The deletion of this hinge prevents autoinhibition and restores processivity to the motor (). A comparison of the overall architecture of the Kinesin-1 and OSM-3 stalk suggested the presence of an analogous hinge in OSM-3 (, H2). Therefore, we wondered whether the deletion of H2 and the in-phase fusion of the adjacent coiled coils (, cc1 and cc2) could restore processivity to OSM-3. Strikingly, this construct (OSM-3–ΔH2) exhibited robust processive movement with long run lengths (∼2 μm; and Video 2, available at ) in the single-molecule TIRF assay and displayed similar elevated ATPase rates (∼70 ATP/s/head) to O-K. The dramatic activation of processivity in OSM-3–ΔH2 demonstrates that the OSM-3 motor, in the absence of any fusion to Kinesin-1, has an intrinsic potential for processive motion. These results also show that H2 plays a critical role in repressing the processivity and microtubule-stimulated ATPase activity of OSM-3. Having demonstrated the importance of H2 for the regulation processivity in vitro, we searched through OSM-3 alleles for mutations in H2 that produce chemosensory defects in (). We noticed one allele () in H2 that changed a glycine to a glutamatic acid (G444E). When this H2 point mutation was introduced into full-length OSM-3 (OSM-3–G444E), we found a striking activation of processivity and ATPase activity (75 ATP/s/head) similar to that observed with the deletion of the entire H2 region. The velocity of OSM-3–G444E (1.1 ± 0.2 μm/s) is also similar to that of OSM-3–driven IFT transport in vivo, implying that this mutation could activate OSM-3 in a manner similar to its activation in vivo. This dramatic activation of motility by a single point mutant implies that a specific conformation of H2 is required for the regulation of OSM-3 motility. In Kinesin-1, H2 facilitates a conformational transition between a compact, nonprocessive form (S = 6.7) and a more extended processive form (S = 5.1; ) . The compact to extended conformational transition is favored by cargo binding () and high ionic strength (). To investigate whether a similar conformational change could underlie the regulation of OSM-3 processivity, we examined the hydrodynamic properties of wild-type and mutant OSM-3 by sucrose gradient sedimentation (). Similar to Kinesin-1, OSM-3 sediments with a higher S value at low ionic strength (S = 7.9) than at high ionic strength (S = 6.8), suggesting that it too has compact and extended conformations. In contrast to wild-type OSM-3, the H2 mutants (OSM-3–ΔH2 and OSM-3–G444E) sedimented with a low S value (S = 6.5–6.9) at both low and high ionic strengths. These results suggest that the processive H2 mutants stabilize an extended conformation of OSM-3. We attempted to visualize the two OSM-3 conformations by rotary shadow EM but unfortunately could not obtain satisfactory images as a result of aggregation of the motor on the mica surface (unpublished data). The correlation between the extended conformation of OSM-3 and in vitro processivity suggests that a reversible autoinhibitory interaction regulates the motility of OSM-3. Relief of OSM-3 autoinhibition in vivo might be stimulated by cargo binding. To mimic cargo binding in vitro, we attached wild-type OSM-3 and OSM-3–G444E via their C-terminal GFP to beads coated with GFP antibody. Motor-coated beads were then captured by an optical trap and positioned near axonemes immobilized onto a coverslip. As expected, OSM-3–G444E-coated beads moved processively at three different constant loads generated by force feedback (1–6 pN; ). In a stationary (nonfeedback) optical trap, the motor dissociated before reaching a stall force (cessation of forward motion when the opposing force of the optical trap equals the maximum force produced by the motor). However, the observation of movement with a force feedback trap at 6 pN indicates that the motor can exert forces in excess of that value and, thus, appears to have similar force-producing capability to Kinesin-1 (). Wild-type OSM-3–coated beads also moved processively with velocities comparable with OSM-3–G444E at all three force levels (), implying a similar mechanism. To determine whether the bead motion was driven by single or multiple motors, the fraction of moving beads was measured as a function of the motor/bead ratio. Poisson statistical analysis clearly revealed that a single OSM-3–G444E or wild-type OSM-3 motor is sufficient to move a bead (; ), although wild-type OSM-3 required a 60-fold higher motor/bead ratio to yield an equivalent probability of bead movement as OSM-3–G444E. The requirement for a higher concentration of wild-type OSM-3 could be caused by a lower probability of motor-bead attachment as a result of the folded conformation of the inactivated OSM-3 or as a result of inefficient activation of the repressed OSM-3 motors by their attachment to the GFP antibody-coated beads. The fact that wild-type OSM-3 bound to anti-GFP antibody-coated glass slides can move microtubules in a gliding assay also is consistent with the idea that surface attachment can activate repressed OSM-3, although the number of active motors was not investigated in this assay. In summary, our results demonstrate that wild-type OSM-3, once relieved of its inhibition by surface attachment, is capable of processive movement. In conclusion, we have shown that the processive movement and microtubule-stimulated ATPase activity of OSM-3 are repressed in solution. ATPase activity and processivity are both dramatically stimulated by mutations in hinge 2, which also change the conformation of the motor from a compact to an extended form. Wild-type OSM-3 also becomes processive when attached via its tail domain to beads. Collectively, these results suggest a model in which OSM-3 exists in the cytoplasm in a compact, autoinhibited state and that binding to an IFT particle relieves this autoinhibition, converting the motor to an extended conformation and enabling long-distance processive movement (). Although more work will be needed to understand the structural basis of this autoinhibition, the processivity of OSM-3–Kinesin-1 chimeras (O-K and K-O) suggests that inhibition requires both the OSM-3 motor and stalk/tail domains and perhaps involves interactions between the two. This proposed regulatory mechanism for OSM-3 is similar to that described for Kinesin-1 (for review see ). However, these two motors belong to different kinesin classes that share no sequence similarity in their nonmotor domains, having diverged very early in eukaryotic evolution (), and, thus, the structural details of autoinhibition are likely to differ. Kinesin-3 processivity also is repressed by the formation of an intramolecular coiled coil in its neck region, which inhibits dimerization (). Although the effects on motor activity are not known, heterotrimeric Kinesin-II also undergoes a salt-dependent conformational shift between a compact and an extended form (). Thus, autoinhibitory mechanisms, although differing in their precise intramolecular interactions, may be commonly used in motor regulation. It will be interesting to explore whether KIF17, the mammalian homologue of OSM-3 involved in olfactory cilia IFT and neuronal transport in dendrites (; ), is also regulated by an autoinhibitory mechanism similar to OSM-3. The regulation of OSM-3 processivity is likely to be important for its biological function in vivo. Strongly supporting this connection, the OSM-3–G444E allele (), which interferes with autoinhibition in vitro, behaves indistinguishably from the OSM-3–null allele () in chemosensory neurons (). Further supporting a loss of OSM-3 function, the distal ciliary segment (which is supplied exclusively by OSM-3 transport; ) does not form in OSM-3–G444E mutant animals, and IFT particles move along the remaining ciliary segment at velocities similar to IFT particles being transported by heterotrimeric Kinesin-II alone (Ou, G., and J. Scholey, personal communication). Although the OSM-3–G444E allele () behaves as a null mutation in vivo, our in vitro optical trapping experiments show that its velocity and force production are indistinguishable from the wild-type motor. Thus, the most plausible explanation for the in vivo mutant phenotype is a loss of motor regulation rather than motor domain dysfunction. Two possible mechanisms could explain how a loss of autoinhibition gives rise to a null phenotype. First, a constitutively active OSM-3–G444E motor may not be able to dock onto IFT cargo. Second, processive OSM-3–G444E motors may constitutively move along microtubules in the neuronal cell body and fail to be delivered to the cilium. Another open question is how OSM-3 is relieved of its autoinhibition. It has been shown previously that the DYF-1 protein is required to load OSM-3 onto IFT particles (), but purified DYF-1 did not activate wild-type OSM-3 in our single-molecule assay (unpublished data). Thus, the regulatory machinery may require additional proteins or posttranslational modifications. Consistent with this idea, two novel mutants with a similar phenotype to and worms were recently identified (Ou, G., and J. Scholey, personal communication). The ability to study OSM-3 autoinhibition in in vitro and in living provides powerful tools for dissecting the regulatory mechanism of this IFT motor. OSM-3 cDNA was obtained from J. Scholey (University of California, Davis, Davis, CA). From this clone, we constructed OSM-3–GFP in a pET-17bplasmid, which encodes aa 1–699 of OSM-3 with a C-terminal GFP (S65T variant) followed by a His tag. K530-GFP (aa 1–530 of Kinesin-1) was derived from K560-GFP (). In the O-K construct, aa 1–337 of the OSM-3 motor domain were followed by aa 337–530 of the K530 stalk. In the K-O construct, aa 1–336 of the human Kinesin-1 motor domain were followed by aa 338–699 of the OSM-3 stalk/tail. The OSM-3–G444E construct was created by QuikChange mutagenesis (Stratagene). For OSM-3–ΔH2, aa 428–447 were removed to maintain a continuous heptad repeat between cc1 and cc2 (). Protein expression and purification were performed as described previously (). For ATPase assays, single-molecule fluorescence and optical trap experiments, and gliding assays, motor proteins were further purified by microtubule affinity (). Motor concentration was determined either by Bradford assay or SDS-PAGE using BSA as a standard. Microtubule-stimulated ATPase activities were measured as described previously () using 1–20 nM of motor and 0–10 μM of microtubules in BRB12 (12 mM Pipes, pH 6.8, 2 mM MgCl, 1 mM EGTA, 1 mM DTT, and 1 mM ATP). and microtubule. The movements of single GFP-fused kinesin molecules along axonemes were visualized by TIRF microscopy using BRB25 buffer (BRB12 with 25 mM Pipes) as described previously (). The laser power for total internal reflection illumination was 9 mW. Motility was analyzed and corrected for photobleaching using ImageJ software (National Institutes of Health) as described previously (). Microtubule gliding assays were performed using a glass slide coated with GFP antibodies, and gliding velocities of Cy3-labeled microtubules were determined as described previously (). We found that the microtubule gliding velocities of wild-type OSM-3 are closer to those of the OSM-3 hinge mutant when the wild-type motor is adsorbed directly to glass rather than via surface-coated GFP antibodies. Sucrose density centrifugation was performed in BRB25 and 10 μM ATP with 0, 0.05, 0.5, or 1 M NaCl. OSM-3 and standard calibration proteins (ovalbumin, 3.7 S; albumin, 4.2 S; and catalase, 11.3 S) were mixed and loaded onto 12–33% sucrose gradients. After centrifugation at 50,000 for 6 h using a SW55Ti rotor (Beckman Coulter), fractions were analyzed by SDS-PAGE. S values of the standards were plotted versus their peak sedimentation fraction number and fit to a linear curve. S of the OSM-3 motor were calculated based upon their peak sedimentation fraction and the slope of the standard curve. The optical trapping bead assay was performed using a feedback-controlled single-beam trapping microscope. GFP-tagged OSM-3 and OSM-3–G444E were coupled to carboxylated latex beads (0.92-μm diameter; Invitrogen) via affinity-purified anti-GFP antibodies (). Trapped beads were positioned near rhodamine-labeled sea urchin sperm flagellar axonemes immobilized onto a coverslip. Bead displacement was sampled at 2 kHz with a quadrant photodiode detector. Trap stiffness was calibrated for each trapped bead from the amplitude of the thermal diffusion. Before each experiment, the trapped bead was scanned along the x axis (coinciding with the long axis of the axoneme) across the detection region to obtain the detector's response. Experiments were performed at dilutions at which the fraction of beads moving was ≤0.3 to ensure measurements on a single-molecule level (). Velocities were obtained from the slopes of the displacement traces of the beads moving under constant load. The bead-trap separation during force-clamp measurements was between 50 and 150 nm, depending on applied load (1–6 pN) and trap stiffness (0.022–0.06 pN/nm). The assay solution consisted of 80 mM Pipes, pH 6.8, 2 mM MgCl, 1 mM EGTA, 1 mM Mg-ATP, 1 mg/ml casein, 10 mM DTT, and an oxygen scavenger system (). Video 1 shows that wild-type OSM-3–GFP does not move processively along axonemes. Videos 2 and 3 show the processive motion of OSM-3–ΔH2-GFP and OSM-3–G444E-GFP, respectively, along axonemes. Fig. S1 shows the microtubule-stimulated ATPase activity of wild-type OSM-3 and OSM-3 mutants and chimeras. Table S1 is a summary of individual sucrose gradient data for wild-type and OSM-3 hinge mutants. Online supplemental material is available at .

Small ubiquitin-related modifier (SUMO) proteins have been implicated in a wide variety of processes (). Although budding yeast has a single SUMO, called Smt3p, there are three commonly expressed mammalian SUMO paralogues, called SUMO1, -2, and -3 (). SUMO2 and -3 are 96% identical, whereas SUMO1 is ∼45% identical to either SUMO2 or -3. (Where they are not distinguishable, SUMO2 and -3 are referred to jointly as SUMO2/3 in this paper.) Newly synthesized SUMO proteins are proteolytically processed to expose a C-terminal diglycine motif. Mature SUMO proteins are linked to their substrates through an amide bond between their C-terminal carboxyl group and an ɛ-amino group of target lysine residues within the substrate. This linkage is accomplished by a pathway that requires an activating enzyme (E1), a conjugating enzyme (E2), and SUMO protein ligases (E3s; ; ). The linkage between SUMO proteins and their substrates can be hydrolyzed by SUMO proteases (; ) and may therefore be dynamic in vivo. Individual SUMO paralogues appear to play distinct functions in vertebrate cells (; ), and many substrates are modified in a paralogue-specific fashion (; ). Because all paralogues share the same E1 and E2 (), the selectivity of E3 enzymes and proteases is likely to play key roles in regulating paralogue-specific conjugation patterns. Ubiquitin forms polymeric chains through the linkage of additional ubiquitin moieties to internal lysines of previously conjugated ubiquitins. The biological roles of ubiquitin chains depend upon the lysines chosen as acceptors during their extension (). Although the prevalence and physiological role of SUMO chains have not been established, it has been shown that Smt3p, SUMO2, and SUMO3 can form chains in vitro and in vivo (; ; ). The major acceptor lysines used in these chains are Lys15 in Smt3p and Lys11 in SUMO2 and -3. Although SUMO1 does not have a conserved lysine at the equivalent residue, it can also form chains in vitro through an uncharacterized linkage (). There are a limited number of reports indicating that chain formation by SUMO2 or -3 is required in vivo for correct regulation of substrate function (; ). The promyelocytic leukemia protein (PML) is a major SUMO-conjugation substrate and the defining constituent of PML bodies, which are nuclear structures of undefined function. It has been reported that the formation of SUMO3 chains may be particularly important for regulation of PML body structure and dynamics (). Ulp1p (ubiquitin-like protease 1p) and Ulp2p/Smt4p are budding yeast Smt3p proteases that share a conserved catalytic domain (, ). These enzymes are not functionally redundant. Ulp1p is likely to have an important role in posttranslational processing of Smt3p; overexpression of mature Smt3p weakly suppresses ulp1Δ mutants, whereas nonprocessed forms of Smt3p do not (). In contrast, Ulp2p has been implicated in the deconjugation of Smt3p from its substrates () and, specifically, in preventing the formation of poly-Smt3p chains (). ulp2Δ cells accumulate high-molecular-weight Smt3p-containing species, which are lost when conjugatable lysine residues within Smt3p are mutated (). Additionally, Smt3p mutants that do not form chains suppress some ulp2Δ phenotypes (), consistent with the notion that those phenotypes arise from inappropriate accumulation of Smt3p chains. Mammalian proteins related to Ulp1p and -2p have been called sentrin-specific proteases (SENPs; ). Mammals have seven distinct genes encoding SENP/Ulp family members (; ). Notably, some of these gene products act on other ubiquitin-like proteins (; ). Moreover, there are distinctions even among the SENP/Ulps that have been verified as SUMO proteases. First, they show distinct localizations (). Second, in vitro studies suggest that SENP/Ulps have specialized enzymatic activities. For example, SENP2 is significantly more efficient in processing SUMO2 than SUMO1 or -3 (); although SENP1 processes SUMO-1 and -2 efficiently, it is ineffective for processing of SENP3 (). In addition, SUMO-specific protease 1 (SUSP1, also known as SENP6; ) was reported to act effectively in vitro as a processing enzyme but not as a deconjugating enzyme for SUMO1 (). The enzymatic specificities of individual SENP/Ulps have not been systematically evaluated, nor have in vitro observations on individual SENP/Ulps been well correlated to their in vivo roles. We have examined the localization, biological function, and specificity of SUSP1, the largest human SENP/Ulp. We found that SUSP1 localizes to the nucleoplasm. Suppression of SUSP1 synthesis in cell lines stably expressing EGFP fusions to individual SUMO paralogues caused redistribution of EGFP-SUMO2 and -SUMO3 into nuclear foci. A similar redistribution was not observed in cells expressing EGFP-SUMO1. Immunofluorescence studies showed that the majority of EGFP-SUMO2 and -SUMO3 foci in the SUSP1-depleted cells corresponded to PML bodies. Notably, both the size and number of PML bodies increased under these circumstances. Fusion protein maturation was not required for this redistribution, suggesting that it resulted primarily from a deficit of deconjugation activity. We investigated the enzymatic specificity of SUSP1 using vinyl sulfone (VS) inhibitors and model substrates. We found that SUSP1 has a strong paralogue preference for SUMO2/3, and particularly for substrates containing three or more SUMO2/3 moieties. Our findings suggest that SUSP1 may play a highly specialized role in dismantling SUMO2 and -3 chains that is critical for PML body maintenance. It has been reported that SUSP1 localizes primarily in the cytoplasm (). Because all human SUMO paralogues are predominately nuclear (), we reexamined the localization of SUSP1 in HeLa cells () using both immunofluorescence staining and N- or C-terminal fusions between SUSP1 and EGFP. All of these approaches clearly indicated that SUSP1 is a nucleoplasmic protein, with minimal distribution to the cytoplasm, nucleolus, or nuclear envelope. To map the domain responsible for its nuclear localization, different fragments of SUSP1 were expressed as C-terminal EGFP fusions (). When we examined their distribution in fixed cells, we found that sequences between residues 84 and 448 of SUSP1 are required for its nuclear localization (). SUSP1's localization sequence is notable with respect to targeting sequences of other SENP/Ulps: the catalytic domains of all SENP/Ulps are localized toward their C termini (). In every case where the targeting requirements of SENP/Ulps have been determined, correct localization requires sequences within their N-terminal domains (Fig. S1, available at ). Our results for SUSP1 are consistent with this pattern, suggesting that the N-terminal domains of SENP/Ulps generally mediate their subcellular targeting. To examine the biological role of SUSP1, its expression was suppressed by siRNA-mediated gene silencing in human osteosarcoma-derived cells (U2OS cells) that stably express N-terminal EGFP fusions to SUMO1, -2, or -3 (). 48 h after transfection, cells that were treated with siRNAs against SUSP1 mRNA showed substantially lower levels of SUSP1 protein (<10%) than control cells that were treated with siRNAs directed against Lamin A/C mRNA (). In SUSP1-depleted cells, EGFP-SUMO1 distribution was indistinguishable from Lamin-depleted controls (). In contrast, EGFP-SUMO2 and -SUMO3 showed striking accumulation within nuclear foci in ∼30% of the SUSP1-depleted cells, although this redistribution was not observed in the control cells. Although the overall spectrum of EGFP-SUMO2– or EGFP-SUMO3–conjugated targets detected by Western blotting was not grossly different after depletion of SUSP1, we observed a moderate but consistent increase in very high molecular weight GFP-containing species (Fig. S2, available at ). No comparable accumulation of high molecular weight GFP-containing species occurred in EGFP-SUMO1–expressing cells (unpublished data). With prolonged incubations after siRNA treatment, we observed a higher level of cell death in the SUSP1-depleted cells expressing EGFP-SUMO2 and -SUMO3 than in Lamin-depleted controls (unpublished data), possibly suggesting that the accumulation of such SUMO2/3-conjugated species is detrimental to cell survival. Importantly, redistribution of EGFP-SUMO2 and -SUMO3 was not observed when siRNAs were cotransfected with a plasmid that expresses a fusion between SUSP1 and the red fluorescent protein (RFP), encoded by a mutant mRNA that is not degraded by the siRNAs (). This finding substantiates the conclusion that altered EGFP-SUMO2 and -SUMO3 distributions are a direct result of SUSP1 depletion. Together, these findings indicate that SUSP1 plays a paralogue-specific role in the regulation of nucleoplasmic SUMO2/3. To determine whether the processing function of SUSP1 was crucial for the redistribution of EGFP-SUMO2 or -SUMO3, we repeated the siRNA experiment in U2OS-derived cell lines stably expressing the processed forms of EGFP-SUMO2 and -SUMO3, which terminate with the mature diglycine motif (). Similar to cells expressing the unprocessed forms, EGFP-SUMO2(GG) or -SUMO3(GG) became concentrated strongly into nuclear foci. This result implies that accumulation of EGFP-SUMO2 or -SUMO3 into foci after SUSP1 depletion results from the inability to deconjugate these fusion proteins from their substrates rather than insufficient processing capacity. Consistent with this conclusion, we observed no redistribution after siRNA of nonconjugatable EGFP-SUMO2 or -SUMO3 that had only a single C-terminal glycine (). Together, our results suggested that insufficient levels of SUSP1 deconjugation activity caused the accumulation of EGFP-SUMO2 and -SUMO3 in nuclear foci. It was therefore of interest to characterize these structures. First, we examined the size of EGFP-SUMO2– or EGFP-SUMO3–labeled foci in cells depleted of SUSP1 or Lamin A/C (). We observed a bimodal distribution of foci size in the SUSP1-depleted cells, with a substantial peak of larger foci centered ∼0.8 μm in diameter. There were very few foci of this size in the Lamin-depleted cells. We reasoned that these foci might correspond to nuclear subcompartments where SUMO2/3 normally play a physiological role. To test this idea, we stained SUSP1-depleted cells with a variety of antibodies that recognize antigens characteristic of splicing foci (SC35; ), pericentric heterochromatin (trimethyl-Histone H3 [Lys9]; ), and centromeres (CREST sera; ). We did not observe colocalization of EGFP-SUMO2/3 with either SC35 or trimethyl-Histone H3 (unpublished data). We observed some colocalization with CREST sera staining, but the extent of this accumulation was not substantially different between SUSP1-depleted and control cells (unpublished data). We also stained the cells with antibodies directed against a variety of SUMO substrates, including Bloom's antigen (BLM), Wilms' tumor 1 (WT1), proliferating cell nuclear antigen (PCNA), p300, and PML (; ). We did not find redistribution of BLM, WT1, PCNA, or p300, nor their accumulation within the EGFP-SUMO2/3 foci (unpublished data). However, we saw a substantial increase in the number of PML bodies after SUSP1 depletion and extensive colocalization of PML with EGFP-SUMO2 or -SUMO3 (). Comparison of PML bodies versus EGFP-SUMO3 foci () showed that the size distribution of PML bodies after SUSP1 depletion mirrored the change in EGFP-SUMO3 foci. Indeed, the larger EGFP-SUMO3 foci in SUSP1-depleted cells almost universally correlated to PML-containing structures (), suggesting that enlarged PML bodies result from insufficient SUMO2/3 deconjugation after SUSP1 depletion. The smaller, non–PML-containing bodies may analogously represent other structures to which SUMO2/3 conjugates associate under normal circumstances. To directly determine the paralogue specificity of SUSP1, we used VS derivatives of tagged SUMO1 and -2 (HA-SU1-VS and HA-SU2-VS; ). VS reagents derived from ubiquitin-like proteins covalently react with the nucleophilic active site residues of their respective modifying enzymes, showing considerable preference toward deconjugation enzymes over E1 and E2 enzymes. To test the reactivity of SUSP1 for SUMO1 and -2, extracts of control and SUSP1-depleted U2OS cells were incubated with HA-SU1-VS or HA-SU2-VS. Samples were taken at different times and subjected to SDS-PAGE and Western blotting with anti-HA antibodies (). We observed a band in HA-SU2-VS–treated control extracts that migrated with an apparent molecular mass of ∼160 kD. This band was absent in the SUSP1-depleted extracts. Both the molecular mass of this band and its depletion through RNAi indicate that it was derived from a covalent attachment of HA-SU2-VS and SUSP1. Moreover, when we immunoblotted the same samples with anti-SUSP1 antibodies, we observed that SUSP1 was quantitatively shifted into a higher molecular mass band within 5 min of incubation with HA-SU2-VS, confirming this conclusion. Notably, an HA-reactive band of the same size was only weakly seen in the reaction containing HA-SU1-VS, suggesting that it has a significantly lower affinity for SUSP1. Although some SUSP1 migrated at a higher molecular weight after incubation with HA-SU1-VS, this conversion was significantly slower than in reactions containing HA-SU2-VS and did not proceed to completion within 30 min. Together, these results suggest that SUSP1 has a strong preference for SUMO2 over SUMO1. To test this conclusion using more stringent criterion, we preblocked U2OS cell lysates with a threefold molar excess (over the VSs) of untagged aldehyde derivatives of SUMO1 or -2, which bind reversibly to the active sites of SUMO proteases (). After blocking, the extracts were allowed to react with HA-SU2-VS and analyzed by Western blotting using anti-HA antibodies (). In these experiments, SUMO2 aldehyde, but not SUMO1 aldehyde, competed for the binding of HA-SU2-VS to SUSP1 and caused reduction in the intensity of the HA-reactive band at 160 kD, rigorously supporting the conclusion that SUSP1 has higher affinity for SUMO2/3 paralogues. We developed model substrates to further evaluate the enzymatic activity of SUSP1 in processing and deconjugation assays. To examine SUSP1 activity as a processing enzyme, we expressed SUMO1, -2, and -3 in bacteria, fused at their C termini to a T7 tag (SU1-, SU2-, and SU3-T7). We purified these substrates by affinity chromatography. We immunoprecipitated SUSP1 from HeLa extracts using anti-SUSP1 antibodies and confirmed its activity through reactivity with HA-SU2-VS (). We then tested whether the immunoprecipitated SUSP1 fraction could release the T7 tag from the model processing substrates. We observed negligible cleavage of the SU1-, SU2-, and SU3-T7 during the course of a 45-min reaction at 37°C (, bottom), indicating that SUSP1 works poorly as a processing enzyme under these conditions. As a positive control, we performed the same experiment using immunoprecipitated SENP1 protein (), which has been shown to be an efficient processing enzyme for SUMO1 but inefficient in its action against SUMO3 (). We judged that comparable concentrations of active enzyme were added to both reactions through equivalent reactivity with HA-SU2-VS, which would irreversibly label the active sites of both enzymes during the course of a 15-min reaction. Consistent with the earlier findings, we observed efficient cleavage of SU1- and SU2-T7 by SENP1, with minimal activity toward SU3-T7 (, top). Although these results do not strictly rule out the possibility that SUSP1 is ever involved in processing of any paralogue, they argue that SUSP1 is unlikely to be a major processing enzyme. To examine deconjugation, we produced a purified, recombinant C-terminal fragment containing the primary acceptor site Ran GTPase-activating protein 1 (RanGAP1; His-T7-RanGAP1-C2). We incubated this fragment in vitro with purified E1 and E2 enzymes plus SUMO1 or -2 (), resulting primarily in monoconjugated species containing each paralogue. We again found that immunoprecipitated SUSP1 showed little activity against these model substrates (unpublished data), leading us to conclude that SUSP1 works poorly as a deconjugating enzyme against substrates containing single SUMO moieties. These results suggested that SUSP1 might not act as a general processing or deconjugation enzyme but might act on a much more specialized subset of SUMO2/3-containing substrates. To test whether it might act on substrates that are multiply conjugated, we used polycistronic vectors for bacterial expression of His-T7-RanGAP1-C2 with SUMO E1 and E2 enzymes, along with SUMO1 or -2 (). In this system, His-T7-RanGAP1-C2 becomes highly conjugated with the coexpressed SUMO proteins (), as isopeptidases are absent. Formation of these conjugates requires the diglycine motif at the C termini of SUMO proteins and the primary SUMO acceptor lysine of RanGAP1 (Lys517 in ), indicating that SUMO conjugation in occurs through an isopeptide bond between Lys517 of RanGAP1 and glycine at the C terminus of SUMO1 (). To this extent, conjugation in and mammalian cells are similar, as both occur specifically through the single, conserved primary acceptor lysine (; ). This specificity is likely to reflect the strong and specific binding of Ubc9 to this region of RanGAP1 (; ). It is formally possible that conjugation of Lys517 in bacteria promotes the subsequent recognition and conjugation of other lysine residues within His-T7-RanGAP1-C2. However, we consider this scenario to be improbable because none of the other lysine residues lie within an optimal sequence context, nor do they become conjugated under any other circumstances, including in vitro assays with high concentrations of all conjugation pathway enzymes. Thus, it is most likely that the conjugates from bearing multiple SUMO polypeptides are configured in chains or branched structures. After bacterial lysis and affinity purification of His-T7-RanGAP1-C2–containing species, we obtained a preparation containing His-T7-RanGAP1-C2 conjugated to different numbers of SUMO moieties. We assayed whether immunoprecipitated SUSP1 would effectively deconjugate any or all of the species contained in this mixture (). We observed that the immunoprecipitated SUSP1 fraction cleaved high-molecular-weight SUMO2 conjugates very efficiently. Interestingly, conjugation products containing only one or two SUMO2 moieties were not efficiently deconjugated, even with relatively long incubations (, 60 min). Remarkably, we observed minimal deconjugation of any SUMO1-containing substrates (), indicating that SUSP1 is also not an efficient isopeptidase for SUMO1 conjugates, even those that linked to high numbers of SUMO1 moieties. We have used a coordinated approach to investigate in vivo function and in vitro biochemical properties of SUSP1. Our observations demonstrate that SUSP1 is a nucleoplasmic SUMO2/3 isopeptidase. The activity of this enzyme is required to maintain the distribution of SUMO-conjugated species between subnuclear compartments, and we observe the inappropriate accumulation of SUMO2/3-modified species in PML bodies and other structures in its absence. Moreover, in vitro analysis of paralogue specificity through VS derivatives of SUMO proteins strongly indicate that SUSP1 acts selectively on SUMO2. Further experiments with other model substrates revealed that SUSP1 has minimal activity as a processing enzyme or in the deconjugation of single SUMO moieties. On the other hand, assays of its activity against SUMO-conjugated substrates that were prepared using an –based expression system showed that SUSP1 could act effectively against species conjugated with multiple SUMO2 moieties. Collectively, these observations suggest that SUSP1 plays a highly specialized role in vertebrate cells in the dismantling of highly conjugated SUMO2/3 species and that the appropriate control of these species is critical for the accurate maintenance of nuclear structures, particularly PML bodies. An earlier report indicated that SUSP1 acts as a processing enzyme for SUMO1 in vitro but that it does not efficiently deconjugate SUMO1 from RanGAP1, leading to the conclusion that SUSP1 is primarily involved in SUMO1 processing (). In contrast, our observations imply that SUSP1 has little processing activity () but is important for SUMO2/3 deconjugation, especially from species containing three or more SUMO moieties (). This apparent conflict may reflect the fact that examined neither processing of SUMO2/3 nor deconjugation of monomeric or polymeric SUMO2/3 species. As a result, SUMO1 processing may have appeared to be the most robust activity of this enzyme, even though this activity is not substantial in comparison to SUSP1's capacity to dismantle highly conjugated SUMO2/3 species. also reported that SUSP1 localizes to the cytoplasm, whereas we find that the majority of this protein resides within nuclei of HeLa and U2OS cells (). We are confident in the conclusion that SUSP1 is a nuclear protein, because we have used multiple independent methods to assay its localization. Mapping of the sequences sufficient for targeting of SUSP1 defines an N-terminal domain that contains multiple putative nuclear localization signals (), providing some suggestion that SUSP1 gains entry to the nucleus through classical nuclear import pathways. Interestingly, we find that the depletion of SUSP1 alters the distribution of EGFP-SUMO2 and -SUMO3, causing marked accumulation of these fusion proteins in PML-containing bodies ( and ). It has previously been reported that RNAi-mediated depletion of SUMO3 specifically redistributes PML and causes loss of PML body integrity (). This defect could be rescued by the expression of exogenous wild-type SUMO3, but not by a mutant lacking lysine 11, the primary residue implicated in SUMO2/3 chain formation. In combination with our findings, these data suggest that the integrity of PML bodies may require the formation of SUMO2/3 chains. Because SUSP1 acts to dismantle multiply conjugated species (), its absence after RNAi-mediated depletion promotes not only the accumulation of such species within the PML bodies of cells expressing EGFP-SUMO2 or -SUMO3, but also the assembly of PML bodies of remarkably increased size and number. We have performed genomic searches across eukaryotic species to identify SENP/Ulp family members, to perform a comparison of their protein sequences (). The SENP8/Deneddylase 1–related branch of this family tree was clearly distinct, consistent with the finding that these enzymes act specifically on another ubiquitin-like protein, Nedd8 (; ). More notable, we found that another branch point divided the SUMO-specific SENP/Ulp family members into two distinct subsets, which contained budding yeast Ulp1p and -2p, respectively. Vertebrates possess four enzymes within the Ulp1p branch (SENP1, -2, -3, and -5), but only two within the Ulp2p branch (SUSP1/SENP6 and SENP7). Given this evolutionary relationship, it is interesting to note that our data indicate some conservation of function within one of these two branches: elegant genetic and biochemical studies suggest that the critical function of Ulp2p in budding yeast involves regulation of Smt3p chain elongation (). Our findings indicate that SUSP1 acts preferentially against multiply conjugated SUMO2/3 species (), which are likely to contain chained or branched SUMO structures (). Moreover, we find that changes in the profile of EGFP-SUMO2– or EGFP-SUMO3–conjugated substrates upon SUSP1 depletion are most pronounced among very high molecular mass species (Fig. S2), consistent with the possibility that polymerized EGFP-SUMO2 and -SUMO3 structures form in its absence, whereas other aspects of SUMO2 and -3 metabolism are largely unaltered. SENP/Ulps appear to share both the conserved catalytic region and a similar overall arrangement, wherein the catalytic domains are localized toward the C terminus of each protein and targeting sequences are found within their N termini (). Our finding that sequence features required for nucleoplasmic targeting of SUSP1 lie within its N-terminal domain is clearly consistent with this pattern (Fig. S1). Two other features of SUSP1's sequence may be further notable in light of its strong preference for multiply conjugated SUMO2 and -3 species. First, SUSP1 possesses four sequence motifs that conform to a previously identified SUMO-binding domain (; ). It is interesting to speculate that these motifs may confer a higher affinity of SUSP1 for multimeric SUMO2/3 chains or orient SUSP1 during the process of substrate recognition. Second, SUSP1 also contains a 195-residue insertion that splits the conserved region corresponding to the catalytic domain (Fig. S1). This insertion appears to be unique to SUSP1, although SENP7 possesses a smaller inserted sequence in a closely equivalent configuration. We do not yet know the function of this inserted sequence, although it is possible to speculate that it may either enhance SUSP1's activity against polymeric SUMO2/3 or restrict SUSP1's activity against other targets, including single SUMO conjugates or processing substrates. The most extensive analysis of SENP/Ulp functions have been performed in budding yeast (, , ; ); these studies have conclusively shown that Ulp1p and -2p are not functionally redundant, as elimination of either protein results in highly distinct phenotypic consequences. We believe that vertebrate SENP/Ulps may possess an even higher degree of specialization with respect to their enzymatic activity, paralogue preference, and targeted localization within the nucleus. In this case, it will be vital to characterize each of these enzymes accurately and completely, in order to interpret experiments involving their manipulation correctly and to understand fully their role within the whole SUMO pathway. Monoclonal mouse antibody (PG-M3) against PML was obtained from Santa Cruz Biotechnology, Inc. SUSP1 antibody was raised in rabbit against the N-terminal 499 amino acid of SUSP1 and was affinity purified using an antigen column. Anti-HA 3F10 rat monoclonal antibody was purchased from Roche Diagnostics Corporation. Alexa Fluor–labeled secondary antibodies were obtained from Invitrogen, HRP-conjugated mouse and rabbit secondary antibodies and -hydroxysuccinimide–Sepharose (FF) were obtained from GE Healthcare, HRP-conjugated anti-rat antibody was obtained from Pierce Chemical Co., and HRP-conjugated anti-T7 tag antibody was obtained from Novagen (EMD Biosciences). Anti-SUMO1, -SUMO2, and -Senp1 polyclonal antibodies were raised in rabbit and affinity purified (provided by M. Matunis, Johns Hopkins University School of Public Health, Baltimore, MD). All other reagents were obtained from Sigma-Aldrich unless otherwise stated. HeLa cells, U2OS cells, and their derivative stable lines were grown at 37°C in a humidified atmosphere of 5% CO in DME (Biosource International) with 2 mM glutamine supplemented with 10% fetal bovine serum (Gemini Bio-Products), 100 U penicillin/ml, and 100 μg/ml streptomycin. Cells were transfected with plasmids using Effectene reagent (QIAGEN) according to the manufacturer's instructions. For stable cell line selection, 0.5 mg/ml Geneticin was added to culture medium 24 h after transfection. Cells were incubated in Geneticin-containing culture medium that was refreshed daily for a period of 1 wk after resistant colonies were reseeded sparsely to culture single cell–derived colonies. Uniformly fluorescent colonies derived from single cells were marked and isolated under an inverted fluorescence microscope. Stably transgenic cells were maintained thereafter in medium containing 0.25 mg/ml Geneticin. Cells were grown either on poly--lysine–coated coverslips or LabTekII coverslip-bottomed chambers. For immunofluorescence, the cells were washed with PBS and fixed for 12 min at ambient temperature with 4% paraformaldehyde in PME buffer (PBS supplemented with 5 mM each of MgCl and EGTA). The cells were permeabilized with 0.5% Triton X-100 for 10 min. After washing with PME, the cells were blocked for 10 min in 1% normal horse serum (Vector Laboratories) in PME and incubated for 1 h with anti-PML antibodies diluted 1:200 in blocking solution. The coverslips were washed and incubated for 45 min with Alexa Fluor 568–conjugated secondary antibodies (Invitrogen) diluted 1:400 in blocking solution. Unbound antibodies were removed by washing, and the cells were briefly incubated in 100 ng/ml Hoechst 33258 DNA stain. Finally, the coverslips were mounted in Vectashield mounting medium (Vector Laboratories). Fluorescence microscopy was performed on a confocal microscope (LSM510 META; Carl Zeiss MicroImaging, Inc.) equipped with 40× Plan Neofluar objective. LabTekII coverslip-bottomed chambers were used for live analyses. Temperature (37°C) and CO concentration (5%) were maintained using a humidified environmental chamber. Confocal microscopy software (SP2 version 3.2; Carl Zeiss MicroImaging, Inc.) was used for capturing images, which were then analyzed by Photoshop 7.0 (Adobe). In all figures, scale bars represent 10 μm. We used a 543-nm HeNe laser (5 mW output; detection LP560 nm) for detection of RFP-SUSP1-rescue signal and indirect immunolocalization of PML by Alexa Fluor 568–labeled antibodies. The 488-nm line of an Argon laser (25 mW nominal output; detection BP 505–530 nm) were used for analysis of GFP-conjugated proteins. Hoechst 33258 images were captured using the 364-nm line of an ion laser (Enterprise II ML UV [Coherent, Inc.]; 80 mW nominal output; detection BP 385–470 nm). Expression of SUMO-intein-CBD fusion proteins in was as described in . Bacterial lysates were prepared as described previously () and bound to chitin bead columns (New England Biolabs, Inc.). The columns were washed with five volumes of lysis buffer (50 mM Hepes, 100 mM sodium acetate, pH 6.5, and 50 μM PMSF), followed by three volumes of lysis buffer containing 50 mM β-mercaptoethanesulfonic acid (MESNa; Sigma-Aldrich). The column was incubated overnight at 37°C in the buffer with MESNa to allow on-column cleavage. The SUMO-MESNa thiolesters were eluted with 1.5 column volumes of lysis buffer, and the fractions containing SUMO-MESNa products were concentrated on a Centriprep (3,000-molecular-weight cutoff; Millipore). To convert SUMO-MESNa products to their VS derivatives, a large excess of Glycine VS (final concentration 0.25 M) was added to concentrated SUMO-MESNa (1–3 mg/ml; 500 μl), followed by addition of 75 μl of 2 M -hydroxysuccinimide and 30 μl of 2 M NaOH (provided by H. Ovaa, Netherlands Cancer Institute, Amsterdam, Netherlands; ). The mixture was incubated for 1–2 h at 37°C, and the reaction was terminated by the addition of 30 μl of 2 M HCl. To obtain SUMO1 and -2 aldehydes, SUMO1 and -2 acetals were synthesized by reacting SUMO-MESNa thiolesters (1–3 mg/ml; 500 μl) with 0.2 ml of 4 M aminoacetaldehyde diethyl acetal, pH 8.5, and 0.14 ml of -hydroxysuccinimide, pH 7.0, at 37°C for 2 h. To obtain SUMO1 and -2 aldehydes, SUMO1 and -2 acetals were separately incubated with 0.15 M HCl at 37°C for 30 min. All steps of derivative syntheses were monitored by HPLC and by SDS-PAGE. VS and aldehyde derivatives of SUMO1 and -2 were prepared as described previously (). 48 h after transfection with siRNAs directed against Lamin or SUSP1, as indicated, U2OS cells were harvested and washed twice in PBS containing 1 mM AEBSF. The cells were resuspended in reaction buffer (10 mM Hepes, pH 7.4, 150 mM NaCl, 5 mM EDTA, 1 mM DTT, and 25 μg/ml each of leupeptin, aprotinin, and pepstatin) and sonicated. Cell lysates were clarified by centrifugation at 16,000 for 10 min. The protein concentration of the total cell lysate was maintained between 500 μg/ml and 2 mg/ml. SUMO-VS was added to a final concentration of 0.3 ng/μl and allowed to react for the indicated intervals. The reactions were terminated by the addition of SDS sample buffer and analyzed by Western blotting with the indicated antibodies. SUMO1- and SUMO2-conjugated (His)-T7-RanGAP-C2 fragment were produced in bacteria () and purified using Ni-NTA beads (provided by Y. Uchimura and H. Saitoh, Kumamoto University, Kumamoto, Japan). SUSP1 was immunoprecipitated from HeLa cell lysate using antibody conjugated to -hydroxysuccinimide–Sepharose (2 μg/ml of beads). 10 μl SUSP1 beads were incubated with 500 ng of SUMO-conjugated T7-RanGAP-C2 in a total reaction volume of 30 μl at 23°C for the indicated times. The reaction was terminated with sample buffer and analyzed by Western blotting with antibodies against T7, SUMO1, or SUMO2/3. Full-length SUMO1, -2, and -3 cDNAs were subcloned into pET28b expression plasmids that had been cut with NcoI–NdeI, NcoI–NdeI, and BspH1–NdeI restriction endonucleases, respectively. These plasmids were transfected into , and the expression of the fusion proteins was induced with 0.4 mM IPTG under standard conditions. The expressed C-terminally tagged SUMO-T7-His proteins were purified on a Ni-NTA column followed by Q-Sepharose and MonoQ columns to generate pure tagged substrate for the processing reactions. HeLa cells from confluent 15-cm dishes were harvested by trypsinization and washed twice with ice-cold PBS. The cells were snap frozen in liquid nitrogen and then sonicated in 500 μl lysis buffer (10 mM Hepes, pH 7.5, 150 mM NaCl, 1 mM DTT, 1 mM AEBSF, and 5 mg/ml each of leupeptin, pepstatin, and aprotinin). The cell lysates were centrifuged at 120,000 rpm for 5 min at 2°C. The cleared supernatants were incubated with 5 μg of rabbit IgG, affinity-purified anti-SENP1 antibody, or affinity-purified anti-SUSP1 antibody for 1 h at 4°C. 50 μl preblocked protein A beads were added to the lysates and incubated for another hour and subsequently washed three times in lysis buffer. The beads were reacted with 0.3 ng/μl HA-SUMO2-VS in lysis buffer containing 100 μg/ml BSA for 15 min at RT, and the reactions were terminated with sample buffer. Using the different dilutions of the HA-SUMO-VS adducts thus obtained, beads having an equivalent enzymatic activity for SENP1 and SUSP1 were determined by anti-HA Western blot. Beads having equivalent amount of HA-SUMO2-VS reactivity and control beads were then incubated with 40 ng/μl SUMO1, -2 or -3 T7 for different time points at 23°C in 20 μl reaction volume. The reactions were stopped by the addition of sample buffer, and Western blots were performed using anti-T7, -SUMO1, or -SUMO2/3 antibodies. The eukaryotic genome databases of the National Center for Biotechnology Information were searched using psi-BLAST using human Senp protein sequence and query with e-value inclusion cutoff of 0.001, for 6–10 cycles. Whenever any specific genome of interest failed to give any hit, the nucleotide genome sequence of corresponding organism was searched using TBLASTN with human Senp protein sequences as query. The collected protein sequences were aligned using MUSCLE () with 100 iterations. The generated multiple alignments were manually corrected using the conserved C48 peptidase domain as anchor. Phylogenetic analyses were performed using minimum evolution (least-square) method as implemented in MEGA3.1 (), with Poisson correction model and pairwise deletion of gaps and 1,000 bootstrap replicates. Fig. S1 shows aligned sequences of Ulp1p and Ulp2p/Smt4p from and mammalian SENP/Ulp family members. Fig. S2 shows EGFP-SUMO3 in SUSP1-depleted cells. Online supplemental material is available at .

Nuclear pore complexes (NPCs) mediate the bidirectional transport of proteins, RNAs, and ribonucleoprotein cargos across the double-membrane nuclear envelope (NE) of eukaryotic cells. NPCs are large (∼60–120 megadaltons) structures with octagonal rotational symmetry. They are composed of at least 30 different nuclear pore proteins (Nups), each present in an integer multiple of eight copies (; ; ). As observed by electron microscopy, the pore itself is ∼90 nm in length and is ∼50 nm wide at its narrowest point. Flexible filaments extend ∼50 nm into the cytoplasm, and a filamentous open basket structure extends ∼75 nm into the nucleoplasm (; ). Molecules smaller than ∼20–40 kD (diameter ∼4–5 nm) transit through the NPC without specific recognition (“passive diffusion” or “signal-independent transport”). Larger molecules up to ∼25 megadaltons (diameter ∼40 nm) typically form a complex with at least one transport receptor for transit through the NPC (“facilitated translocation” or “carrier-mediated, signal-dependent transport”); in most cases, the transport receptor belongs to the importin/exportin superfamily (; ; ). Importins and exportins (also known as karyopherins) recognize nuclear localization sequences and nuclear export sequences, respectively, and migrate with cargos through NPCs (). Import complexes (ICs), consisting of cargo and importins, are disassembled after transit through the NPC by RanGTP, the GTP-bound form of the G-protein Ran. Export complexes, consisting of cargo, exportin, and RanGTP, release cargos after transport through the action of Ran-binding proteins (RanBPs) and RanGAP, Ran's cytoplasmically localized GTPase activating protein (; ; ; ; ; ). A RanGTP concentration gradient is established and maintained by cytoplasmic RanGAP and nucleoplasmic RanGEF, the chromosome bound guanine-nucleoside exchange factor for Ran, which catalyzes GDP/GTP exchange (). This RanGTP concentration gradient across the NE is quite steep, as the nuclear RanGTP concentration typically exceeds the cytoplasmic concentration by ∼1,000-fold (). The direction of net cargo flux is determined by the RanGTP concentration gradient because transport directionality was reversed by gradient inversion (). An extensive network of thousands of phenylalanine-glycine (FG) repeat motifs located on almost half the Nups (FG-Nups) provides binding sites for importins and exportins. These FG-Nups are distributed throughout the NPC structure on both the cytoplasmic filaments and nuclear basket and within the pore itself (; ; ; ; ). The FG-repeat regions are natively unfolded (disordered) and highly flexible (). Over half of the mass of the FG-repeat domains on the FG-Nups can be deleted without loss of viability or the permeability barrier (). The structure, distribution, and properties of the FG-Nups have been intensely debated. It is clear that these FG-Nups regulate passage through the ∼50-nm-diameter open pore visualized by structural studies (; ) and that they play critical roles in selectivity and permeability regulation (). Members of the importin (Imp) β superfamily of transport receptors allow signal-dependent cargos to gain access to the pore and, ultimately, to migrate across the NE via interactions with the FG-Nup network (; ; ; ; ; ; ; ; ; ). Numerous distinct models have been postulated to explain the precise mechanism whereby molecules gain access to and subsequently translocate through the NPC (; ; ; ; ; ; ; ; ; ). None of these models addresses the possible functional effects, such as changes in transport speed or import efficiency, that could result when multiple cargo complexes or transport cofactors simultaneously interact with the same NPC. Studies using cargo-decorated colloidal gold particles indicated that multiple import and export cargos can simultaneously bind to an NPC (; ), suggesting that multiple import and export cargos can simultaneously transit through the NPC. Consequently, considering the thousands of FG-repeat motifs providing binding sites, the many transport cofactors that must be recycled through the NPCs, and the massive amount of material that must pass through NPCs (), the general picture that has emerged is one of two-way traffic in which multiple cargos, as well as multiple empty transport cofactors (), all simultaneously migrate through a given NPC. This picture of multiple-cargo, simultaneous two-way traffic suggests that the identity and amount of material in transit at any one time could significantly affect transport speed or import efficiency. Here, import efficiency is defined as the number of cargos that enter the nucleus after ICs interact with an NPC divided by the total number of ICs that interact with that NPC. As a result of the RanGTP gradient, RanGTP is more readily available near the nucleoplasmic exit of the pore rather than near the cytoplasmic exit. Thus, for cargos that require RanGTP to exit from the NPC, a reasonable prediction is that IC dissociation and cargo exit occur preferentially near the nucleoplasmic exit. Such a situation would seemingly promote a relatively high import efficiency, though it could be significantly influenced by the distribution of RanGTP within or near the pore or by the traffic level within the NPC. We showed earlier that the model cargo NLS-2xGFP interacts with NPCs for ∼9 ms at low IC concentrations. At high IC concentrations, this cargo accumulates in the nucleus at a rate of up to ∼1 translocation event per NPC per ms (). One interpretation of these findings is that ∼9 ICs can transit simultaneously through a single NPC at high IC concentrations, consistent with the picture gleaned from the colloidal gold experiments discussed in the previous paragraph. However, this explanation implicitly assumes that the cargo translocation time is invariant with the IC concentration. Herein, we show that the translocation time of both signal-independent and -dependent cargos is not invariant and that one factor that influences the translocation time is the Imp β concentration. We report import efficiency measurements and demonstrate that the Imp β concentration also affects the import efficiency. The most parsimonious explanation is that the Imp β concentration within the NPC can directly and/or indirectly influence the transport speed and import efficiency. We used single-molecule fluorescence (SMF) microscopy to determine interaction times and import efficiencies for an IC consisting of NLS-2xGFP(4C), Imp α, and Imp β. The four maleimide-reactive cysteines of NLS-2xGFP(4C) were used for dye labeling (Alexa 555 or 647). Images were obtained at 500 or 1,000 frames per second. Solution viscosity was increased through the addition of 25% glycerol to reduce the bulk diffusion rate. Under these conditions, an IC's originating and destination compartment was identified for 37% of NPC interaction events ( and Videos 1 and 2, available at ). For the remaining interaction events, the cargo could not be tracked before and/or after interaction with an NPC because of diffusion outside the focal plane. Unless otherwise noted, all single-molecule transport videos were collected within the 1-min time window beginning ∼1 min after initiating transport. A series of control experiments indicated that transport was minimally affected by the presence of 25% glycerol or the dyes on the cargo protein. In addition, single-molecule diffusion measurements indicated that our in vitro viscosity conditions very closely matched the in vivo viscosity (Figs. S1 and S3 and the supplemental text). Under these conditions, in vitro single-molecule import experiments revealed that some cargos crossed the NE (entry events), whereas other cargos returned to the cytoplasmic compartment after interaction with NPCs (abortive events; ). An overlay of trajectories for entry and abortive events suggests that the regions within the NPC accessible to cargos that passed through to the nucleus and those accessible to cargos that aborted transport were similar (). Thus, cargos appear to be able to penetrate fairly deeply into the NPC in cases of abortive transport. These data are consistent with a picture in which cargos randomly diffuse within a relatively large region of the NPC before exiting. Under the conditions of , the NPC interaction times for entry and abortive events were the same (∼8.3 ms), within experimental error (), and the import efficiency was 51 ± 5%. Having established that the interaction time and import efficiency could be measured at a very low cargo concentration, we measured these parameters at higher, more physiological cargo concentrations. Because very low fluorescent cargo concentrations were essential for detection of single molecules, nonfluorescent cargo was added to the import reactions to increase the total cargo concentration. The interaction time dropped from 8.6 ± 0.4 to 2.2 ± 0.1 ms when the IC concentration was increased from 0.1 nM to 15 μM. The import efficiency increased from 51 ± 5 to 77 ± 5% over the same concentration range (). The interaction time, import efficiency, and bulk nuclear accumulation rate indicated that an average of ∼1–2 ICs at a time interacted with the NPC when the IC concentration was ∼5 μM. Therefore, we sought an explanation as to how the interaction time could decrease approximately threefold as the IC concentration was increased from 0.1 nM to 5 μM (, red) with so few ICs within the large NPC structure. One possibility we considered was that the NPC structure exhibits hysteresis, i.e., that it has a memory, such that a second IC rapidly following the first can transit faster than if the transport events are more infrequent. However, because Imp β–Imp α and Imp β–RanGTP complexes, as well as Imp β alone, bind to NPCs (, ; ; ; ), we tested whether the Imp β concentration alone could modulate the interaction time and the import efficiency. The cargo concentration was fixed at 0.1 nM, and the Imp β concentration was increased from 0.1 to 15 μM. The results () were similar to those obtained by increasing the cargo and cofactor concentrations simultaneously, though the changes were slightly larger. These data are consistent with the hypothesis that similar initial IC and cargo-free Imp β concentrations rapidly lead to similar numbers of Imp β molecules/complexes within the NPC and that these Imp β molecules/complexes within the NPC influence interaction times and import efficiencies. The sharp decrease in interaction time observed as the Imp β concentration was increased from 0.5 to 1.5 μM (, red) is consistent with a picture in which the binding of many Imp β molecules/complexes ( ∼1 μM, approximated from the transition point in this figure; similar to the of ≥4 μM estimated for transportin []) to the NPC results in a decrease in cargo interaction time. The significant decrease in interaction times and increase in import efficiencies observed at high Imp β concentrations were not a consequence of an increased Imp β/Ran ratio because the interaction times and import efficiencies at 1.5 μM Imp β with 2 or 6 μM Ran (and 1 or 3 μM NTF2, respectively) were similar (Table S1, available at ). At 15 μM Imp β, the NPC interaction times for entry and abortive events were 1.0 ± 0.1 and 1.4 ± 0.1 ms, respectively (; see the supplemental text for an explanation of differences between the interaction times of entry and abortive events). Interaction time and import efficiency were not linearly related (), indicating that different sets of parameters determine these two transport characteristics. We next tested the effect of RanGAP and RanBP1 on interaction time and import efficiency. RanGAP activates the Ran GTPase, causing rapid GTP hydrolysis (). RanBP1 binds Imp β–RanGTP complexes and recruits RanGAP (). We first determined the interaction time and import efficiency in the presence of increasing concentrations of RanGAP (0–15 μM) at a high initial cargo-free Imp β concentration (15 μM). A significant increase in the interaction time and decrease in the import efficiency was observed (). These data are consistent with the hypothesis that RanGAP promotes the conversion of RanGTP to RanGDP and that the reduced RanGTP concentration in the pore results in an increased interaction time and decreased import efficiency. Because RanBP1 enhances the RanGAP-induced GTP hydrolysis rate of RanGTP by ∼10-fold (), we tested the effect of RanBP1 on the interaction time and import efficiency. At 0.5 μM RanGAP and 3 μM Imp β, the interaction time increased, and the import efficiency stayed constant as the RanBP1 concentration was increased from 0 to 15 μM (). These results support the hypothesis that RanBP1 enhances RanGAP activity at the NPC and that a reduced RanGTP concentration within the NPC leads to an increased interaction time. The fact that the import efficiency stayed constant (within our errors) provides additional support for the conclusion drawn from that interaction time and import efficiency are distinct transport characteristics determined by different sets of parameters. In contrast, at 0.5 μM RanGAP and 15 μM Imp β, no effect of 0–15 μM exogeneous RanBP1 was observed (). These results are consistent with the hypothesis that the RanBP1 enhancement of RanGAP activity at the NPC is kinetically slower than the influx of RanGTP under these higher Imp β conditions. To confirm the picture that activation of the Ran GTPase increased the rate at which RanGTP was converted to RanGDP and that the RanGTP concentration at the NPC affected the interaction time and import efficiency, we examined the effect of a nonhydrolyzable GTP analogue, guanosine 5′-(β-γ-imido)-triphosphate (GMP-PNP), on these transport characteristics. We first examined the role of GTP hydrolysis in NLS-2xGFP import at 0.5 μM Imp β. We observed an interaction time of 3.7 ± 0.2 ms and import efficiency of 63 ± 5% in the presence of 1 mM GMP-PNP. These data indicate that the NLS-2xGFP cargo was efficiently transported through NPCs in the presence of nonhydrolyzable GTP, as is expected for a 57-kD signal-dependent cargo (). We next examined the effect of RanGAP at 15 μM Imp β and 1 mM GMP-PNP. We observed an interaction time of 1.3 ± 0.1 and 1.2 ± 0.1 ms and import efficiencies of 78 ± 5 and 80 ± 5% at 0 and 15 μM RanGAP, respectively. Thus, unlike when hydrolysable GTP was present (), RanGAP had no effect in these experiments. These data therefore confirm that RanGAP increased the interaction time and decreased the import efficiency through its ability to promote GTP hydrolysis. The shorter interaction time and higher import efficiency observed with 1 mM GMP-PNP and 0.5 μM Imp β (3.7 ± 0.2 ms; 63 ± 5%) compared with that observed with 1 mM GTP and 0.5 μM Imp β (8.6 ± 0.4 ms; 51 ± 5%; see ) is consistent with the picture that the GMP-PNP effects arise from a high RanGTP (RanGMP-PNP) concentration in the NPC as a result of the inability to form RanGDP. To estimate the extent to which transport cofactors bound to NPCs under the conditions of the single-molecule transport measurements, confocal microscopy was used to examine the localization of fluorescent versions of Ran, Imp α, and Imp β. At both low and high Imp β concentrations, the Ran concentration at the NE increased to a steady-state value within ∼1 min after reaction initiation (). In contrast, the Imp β concentration at the NE continuously increased over a 15-min time frame. Imp α accumulated at the NE to a lesser extent than Imp β under these conditions, indicating that at least some Imp β accumulated in the absence of Imp α (). Despite the variation in Imp β concentration at the NE during the acquisition of single-molecule transport data, a higher mean Imp β concentration at the NE was present at higher bulk Imp β concentrations (). In addition, RanGAP reduced both the Imp β and Ran concentrations at the NE ( and ), indicating that an increased Ran GTPase activity led to a decrease in both the Ran and Imp β concentrations at the NE. The Ran concentration at the NE varied within a narrower range and, in some cases, was lower than the Imp β concentration at the NE ( and ), suggesting that some of the Imp β that accumulated in the NPCs in our single-molecule experiments was not complexed with Ran (consistent with the findings of ). Comparing these results with the single-molecule transport measurements, the general conclusion is that shorter interaction times and greater transport efficiencies were obtained when higher Imp β concentrations were observed at the NE. As an additional test of the hypothesis that the accumulation of Imp β within the NPCs influenced the interaction time and transport efficiency, single-molecule transport measurements were made at 0.5 μM Imp β 15–16 min after reaction initiation, where the mean Imp β concentration at the NE was approximately fivefold higher than at 1–2 min after reaction initiation (). Under these conditions, the interaction time was 1.9 ± 0.1 ms and the import efficiency was 71 ± 6%. These values are significantly different from the values obtained at 0.5 μM Imp β 1–2 min after reaction initiation (). These data support the hypothesis that a higher Imp β concentration in the NPC leads to a shorter interaction time and greater transport efficiency. However, a similarly high concentration of Imp β was observed at the NE at 0.5 μM Imp β after a 15–16-min time delay and at 15 μM Imp β plus 15 μM RanGAP with a 1–2-min time delay (), and yet the interaction time was significantly different under these conditions. This finding supports the hypothesis that the concentration of Imp β within NPCs is not the sole determinant of the interaction time in these experiments. We next monitored the diffusion of a 10-kD dextran and the S13 protein of the small ribosomal subunit (rpS13; ∼18 kD) through NPCs under very low concentration conditions (0.1 nM) in the absence of transport cofactors, GTP, and glycerol. The NPC interaction times were 2.2 ± 0.1 and 3.3 ± 0.1 ms for the dextran and rpS13, respectively. When 25% glycerol was added to the import buffer, the interaction times were 2.2 ± 0.2 and 3.2 ± 0.3 ms, respectively, indistinguishable from the previous measurements made in the absence of glycerol (). These data indicate that an approximately twofold increase in bulk solution viscosity did not affect the interaction time of the dextran and rpS13 cargos. Import efficiency values in the 25% glycerol import buffer for dextran and the rpS13 were 51 ± 6 and 50 ± 6%, respectively. The interaction frequencies of the dextran and rpS13 cargos were ∼2.6- and ∼1.8-fold higher, respectively, than that of the NLS-2xGFP cargo at the same concentration (Table S1). These data are consistent with the two- to threefold larger diffusion constant expected for the small cargos compared with that of the NLS-2xGFP ICs based on their molecular masses. Therefore, the measured interaction times for the signal-independent cargos do not arise from a few long, statistically unlikely events of a process occurring faster than our time resolution. We then tested the effect of Imp β on dextran transport. As the Imp β concentration was increased from 0 to 3 μM Imp β, the interaction time decreased and the import efficiency increased (). At 15 μM Imp β, fluorescent dextran molecules were observed in the nucleus, demonstrating that the dextran molecules could still pass through the NPCs but that the interaction time was too fast to be measured (<0.5 ms). These data indicate that the interaction time for signal- independent cargos was strongly dependent on the concentration of Imp β. Because there was no Ran or GTP added in these experiments, the observed effects were due to Imp β alone. As a control to the applicability of our results to in vivo nuclear import, cargos were microinjected into live HeLa cells at 37°C. Individual molecules were first observed diffusing in the cytoplasm. With time, cargos were observed interacting with the NE and transporting into the nucleus. The interaction time and import efficiency of NLS-2xGFP were 8.1 ± 0.1 ms and 51 ± 6%, respectively. The interaction time and import efficiency of the 10-kD dextran were 1.8 ± 0.1 ms and 50 ± 4%, respectively (). In eggs, the in vivo Imp β concentration is ∼3 μM (). In vitro measurements at 3 μM Imp β yielded significantly shorter interaction times and higher import efficiencies for NLS-2xGFP and the dextran than the in vivo values (; and ). Possible (and nonexclusive) explanations that the in vivo data resemble in vitro data obtained at low Imp β concentrations include the following: the in vivo Imp β concentration in HeLa cells is <3 μM, or a significant portion of Imp β is sequestered in our live cell experiments; a low concentration of transport cofactors exists within NPCs with a full complement of transport cofactors and cargos (in vivo conditions), thereby resulting in an increased interaction time and decreased import efficiency; and competition between additional import pathways and/or export pathways (in vivo conditions) results in a longer interaction time and lower import efficiency than expected for a relatively high concentration of transport cofactors within the NPC. Our SMF investigations of nuclear import have directly elucidated several fundamental properties of nucleocytoplasmic transport. The major findings reported here are as follows: the import efficiency of both signal-independent and -dependent cargos is ∼50% in vivo and at low Imp β concentrations in vitro; the interaction times of signal-independent cargos are not affected by an approximately twofold increase in solvent viscosity; for both signal-independent and -dependent cargos, the interaction time decreases and the import efficiency increases as the Imp β concentration increases; for signal-dependent cargos, the interaction time can be modulated by changes in the GTP hydrolysis rate; and though the interaction time and import efficiency are often negatively correlated, they are distinct characteristics determined by different sets of parameters. The implications of these findings are discussed below. The primary message of the results reported here is that the interaction time and import efficiency vary with conditions. At sufficiently high cofactor and cargo concentrations, interaction time and import efficiency must limit the maximum rate of cargo transport ( ). and the mean time to cross the NPC as predicted by the interaction time and import efficiency, the aforementioned data indicate that maximum signal-dependent cargo import rates can be modulated at least ∼10-fold by the Imp β concentration. is variable. This conclusion could explain why the import rate does not follow simple Michaelis-Menten saturation kinetics for Imp β ICs and transportin but, rather, that “cooperativity” is observed at high transport cofactor concentrations (; ). increases. For all our NLS-2xGFP transport data, we observed a similar NPC interaction frequency, which suggests that an increased transport efficiency implies an increased bulk import rate. , the interaction time is not expected to limit import rate. Thus, the magnitude by which changes in interaction time affect bulk transport rates remains uncertain. The most parsimonious explanation for the observed changes in interaction time and import efficiency is that an NPC's transport properties can vary significantly depending on the particular molecules bound to it at any given moment in time. Thus, we introduce “pore occupancy,” a term that recognizes that the number, identity, and distribution of molecules within the FG network affects cargo translocation through the NPC. The term pore occupancy is thus multivariate, explicitly including the concentration and distribution of all molecules within the pore, such as RanGTP, RanGDP, importins, exportins, and cargos. We have shown here that higher Imp β concentrations lead to altered pore occupancy conditions. The altered pore occupancy conditions produced by high Imp β concentrations resulted, in general, in faster cargo translocation and higher transport efficiency. We have determined that higher numbers of Imp β molecules exist at the NE at high bulk Imp β concentrations. Though it is uncertain whether the distribution or concentration of RanGTP in the NPC changed as the bulk Imp β concentration changed, the total Ran concentration at the NE remained essentially unchanged (). Thus, the distribution of transport cofactors within the NPC under various pore occupancy conditions remains unclear and is likely different for different conditions. Our findings with regard to signal-independent cargos challenge the assumption present in most published models (implicitly or explicitly) that small cargos diffuse through an open central channel. From nuclear accumulation measurements on dextrans, the putative central aqueous channel permitting passage of small cargos has been estimated to have a diameter of ∼10.8 nm and a length of ∼45 nm (). The viscosity of our import buffer with 25% glycerol is ∼3.5 cP, and we consider this a reasonable conservative estimate for the viscosity within the putative central channel in this buffer (see the supplemental text). Based on these channel dimensions, assuming a viscosity of 3.5 cP, and correcting for drag arising from the channel walls (), a 10–18-kD cargo is predicted to one-dimensionally diffuse a distance corresponding to the length of this channel significantly faster (i.e., in ∼40–70 μs) than our time resolution (1 ms). Based on this reasoning, the residence time of small cargos within the NPC's putative central channel should be unmeasurable with a 1-ms time resolution. Thus, the measured interaction times for signal-independent cargos at low Imp β concentrations were significantly longer than those expected for diffusional transport through the putative open central channel. Consistent with this observation, found that, based on size alone, the transport flux of GFP (diameter ∼5 nm) through NPCs is >10-fold slower than expected if transport occurs through the putative central channel. In addition, according to the central channel hypothesis, the interaction times of signal-independent cargos should increase with an increase in the viscosity within the channel. However, we observed that these interaction times were unchanged when the bulk viscosity was changed approximately twofold. Though these data appear inconsistent with the central channel model, we cannot rule out that edge effects within an open channel determine the transport characteristics of small cargos. In such a situation, the faster translocation of 10-kD dextran when the Imp β concentration within the FG network is high could be explained by structural changes or by a reduction in the strength of edge interactions (e.g., by Imp β coating the walls of the central channel). Other investigators have reported Nup-deletion mutants where either the signal-independent or -dependent pathway, but not both, is affected, thus suggesting that the two pathways are distinct and unlinked (; ) and supporting the central channel hypothesis. In contrast, the Imp β effects on signal-independent cargo transport suggest that these pathways partially overlap. Collectively, these data support the hypothesis that the two transport pathways are not identical, but they do share some elements. For example, signal-independent cargos could migrate through the same FG-Nup network through which signal-dependent cargos migrate. A relatively open, mesh-like, FG-Nup network could allow passage of small cargos while simultaneously prohibiting passage of larger cargos and allowing the passage of ICs (). A convoluted migration path through such a meshwork is expected to be an inherently slower process than migration through an open channel, consistent with our data. An increase in bulk viscosity could have little effect on the microviscosity within the FG network because of an already relatively high viscosity introduced by molecular crowding and the ordering of solvent molecules in the porous network (; ). The exact mechanism whereby the Imp β concentration influences interaction time is likely to be complex. As described more fully in the supplemental text, we consider that Imp β could play a direct and/or indirect role in influencing pore occupancy. As an example of a direct effect, the presence of a large number of Imp β molecules within the NPC could reduce the volume accessible to transiting cargos. A decreased available volume implies a lower entropy and, thus, an increased free energy for transiting molecules in the pore, which in turn suggests easier access to the rate-limiting transition state and, hence, faster escape (). This mechanism is consistent with the shorter interaction times of the 10-kD dextran at high Imp β concentrations. As an example of an indirect effect, the presence of Imp β in the bulk solution or within the pore itself could change the distribution of RanGTP within the NPC. observed that under excess Imp β concentrations (e.g., [Imp β] > 2 μM), RanGTP is shuttled to the cytoplasmic compartment by cargo-free Imp β. Thus, both the distribution and concentration of RanGTP in the NPC or at the faces of the NPC could be significantly affected by the bulk Imp β concentration (without the total Ran concentration at the NE changing significantly; ). A second possible indirect effect of higher Imp β concentrations is that Imp β molecules within the NPC could disrupt energetically favorable filament interactions (). Consequently, high Imp β concentrations could increase the fluidity of the FG network (and, hence, the diffusion constant of the cargo within the pore). Thus, the faster translocation of signal-independent cargos in the presence of high Imp β concentrations could be explained, at least in part, by Imp β–induced structural changes in the FG network These Imp β–induced structural perturbations could also change the RanGTP distribution within the FG network or promote an increased productive interaction rate between ICs and RanGTP. The simple presence of many macromolecules of any type within the FG network may be sufficient to alter the physical properties of this network. In conclusion, we emphasize that the pore occupancy model does not postulate a transport mechanism but, rather, suggests a general means by which cargo interaction times and transport efficiencies can be modulated. Therefore, the pore occupancy model is in principle entirely consistent with existing mechanistic models. Our data challenge the assumption of an open central channel for small cargos, but the basic size-dependent exclusionary principles behind other mechanistic models are applicable with either an open central channel or a porous network allowing passage of small cargos, as we suggest here. The fact that the Imp β concentration influences interaction time and import efficiency for both signal-independent and -dependent cargos implies that a complete mechanistic model of NPC transport must consider the influence of other molecules within the NPC on cargo migration through the pore. The concentration range in which Imp β causes the greatest changes in interaction time is close to the estimated physiological Imp β concentration (∼3 μM; ), suggesting that in vivo mechanisms in which Imp β is alternately sequestered and released, or in which expression levels are modulated (), could have dramatic effects on nuclear transport rates. Whether cells actively alter cargo interaction times and import efficiencies to regulate transport fluxes in response to cellular needs remains an open question. Unless updated here, we followed our previously published methods (; ). NLS-2xGFP(2C) has two exposed cysteines reactive with maleimides (). To obtain a brighter cargo, two additional cysteine residues were added to NLS-2xGFP(2C) by mutating the surface serines on the bottom of the two GFP β-barrel domains (S175C) to yield NLS-2xGFP(4C) (). The NLS-2xGFP(4C) mutant tagged with four Alexa 647 maleimide (Invitrogen) molecules is ∼1.8-fold more fluorescent than doubly labeled NLS-2xGFP(2C). Ribosomal protein S13 (rpS13) of the small ribosomal subunit was cloned (IMAGE clone 2899987; American Type Culture Collection) into pET9a (Novagen) with a HHHHHHC C-terminal tag, expressed in BL21(λDE3) by 1 mM IPTG induction and purified by nickel-nitrilotriacetic acid Superflow (QIAGEN) and MonoS (GE Healthcare) chromatography. After labeling rpS13 with Alexa 555 maleimide (Invitrogen) and electrophoresing the protein on a polyacrylamide gel, a single fluorescent band was observed. All coding regions were confirmed by DNA sequencing. RanGAP and RanBP1 were expressed with N-terminal 6xHis-tags (plasmids were a giflt from D. Görlich, University of Heidelberg, Heidelberg, Germany) and purified as described previously (). Fluorescent versions of Imp α, Imp β, and Ran were obtained by 10–60-min reaction with Alexa 555 maleimide, yielding a mean labeling of ∼3.3, ∼0.5, and ∼2 dye molecules per protein molecule, respectively. The fluorescent transport cofactors were functional in nuclear transport assays. The 10-kD dextran was obtained from the manufacturer (Invitrogen) labeled with a mean of 1–1.5 Alexa 647 molecules. The buffer used for import experiments was 20 mM Hepes, pH 7.3, 1.5% polyvinylpyrrolidone (360 kD), 110 mM KOAc, 5 mM NaOAc, 2 mM MgOAc, and 1 mM EGTA (import buffer). Glycerol was added where indicated. The SMF microscope setup included a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) equipped with a 1.45 NA 100× oil-immersion objective (Carl Zeiss MicroImaging, Inc.), an on-chip multiplication gain charge-coupled device camera (Cascade 128; Roper Scientific), and the MetaMorph software package (Universal Imaging Corp.) for data acquisition and processing. An I-Pentamax (Roper Scientific) was used for some in vivo experiments. High spatial resolution bulk transport images were obtained with a camera (CoolSnapES; Roper Scientific). A microinjector (FemtoJet; Eppendorf) and objective warmer (Bioptechs) was used for live cell injection at 37°C. Sample temperature was maintained by thermal contact with the immersion oil. Using 1.0-μm TetraSpeck microspheres (Invitrogen), we determined that the bright-field and SMF images were aligned to within 13 ± 2 nm ( = 10). In single-molecule experiments, a 300-mW, 532-nm solid-state laser (Coherent Radiation) was used for Alexa 555 excitation (9 kW/cm measured at the specimen). A 2.5-W ArKr mixed-gas ion laser (Spectra-Physics) was used for Alexa 647 excitation (647 nm; 4 kW/cm). Control experiments revealed indistinguishable interaction times for cargo labeled with Alexa 555 or 647 (Fig. S1 and the supplemental text). The Alexa 647 dye was necessary for high “nonfluorescent” cargo concentrations because of the leakage of GFP fluorescence onto the Alexa 555 fluorescence channel. Accurate import efficiency estimates are obtained only when the analysis includes single-molecule trajectories in which a cargo complex has interacted with an NPC. For molecules that crossed the NE, an NPC interaction was assumed because NPCs are the only known route of passage across the NE. The major challenge was to reliably infer when a molecule interacted with an NPC but did not transport across the NE. Our solution was as follows. Based on fluorescence recovery after photobleaching measurements (; ) and our earlier results (), we assumed that the NPCs remained within a small circular area ( < 100 nm) for the duration of our experiments. An overlay of all the trajectories in which molecules clearly passed through an NPC identified the location of all functional NPCs within the imaging area (). At least three such NE crossing events were observed for every NPC included in our analysis. We defined three classes of trajectories for cargos that approached within 100 nm of the NE from the cytoplasmic side. The first class (, red) consists of trajectories for which the first and last points are at least 100 nm from the NE, and a line between these points crosses the NE (entry and exit compartments different). This class is considered to consist of those molecules that transported across the NE (entry events). The second class (, blue) consists of trajectories for which the first and last points are at least 100 nm away from the NE and both are on the cytoplasmic side of the NE and at least one point is within 100 nm of an NPC location. Because the span from the tips of the cytoplasmic filaments to the center of the NPC is ∼100 nm (the NE, i.e., the red line in , was assumed to pass through the center of the NPC), this class of trajectories identifies those cargo complexes that potentially interacted with an NPC for at least one frame but did not transport. Therefore, this second class is considered to consist of molecules that interacted with an NPC but did not transport through the NPC (abortive events). The third class consists of trajectories that did not fall into either the first or second class (i.e., the exit compartment was unclear) and were discarded from further consideration in import efficiency calculations. Thus, import efficiency is the number of class 1 trajectories divided by the number of class 1 and 2 trajectories and is reported as a percentage. The trajectory alignment in was performed as follows. For entry events, all trajectories were aligned based on their NE crossing point. If there were multiple NE crossing points in a given trajectory, the mean crossing point was used. For abortive events, the horizontal-centroid of the points within the NE ± 100-nm region was placed on the vertical axis defined by the NE crossing points of the entry events. All trajectories were rotated so that the plane of the NE in the alignment corresponds to the plane of the NE at the NPC from which the trajectories were obtained. This procedure centered the trajectories about a single NPC position. Freshly split HeLa cells were grown overnight at 37°C with 5% CO in DME (Invitrogen) supplemented with 4.5 g/l glucose, 862 mg/l GlutaMAX-I, 15 mg/ml phenol red, 100 U/ml penicillin, 100 μg/ml streptomysin, and 10% (vol/vol) newborn calf serum in glass-bottomed Petri dishes (MatTek). Immediately before microinjections, the media was replaced with identically supplemented DME without phenol red. Cargo was diluted to ∼0.1 nM (single-molecule transport) or 0.5 μM (bulk transport) with 10 mM Tris, pH 7.3, 100 mM NaCl, and 0.1% polyvinylpyrrolidone (40 kD) and injected in a single 0.8-s pulse. All errors are 68% confidence intervals, unless otherwise indicated. Fig. S1 shows control experiments that suggest that NLS-2xGFP transport was not affected by 25% glycerol or dyes on the protein. Fig. S2 shows that nuclear accumulation of cargo was optimal at ∼2 μM Ran at both low and high Imp β concentrations. Fig. S3 demonstrates how viscosity was estimated from single-molecule diffusion measurements. Fig. S4 compares NE localization by bright-field and fluorescence imaging methods. Table S1 summarizes the interaction times, import efficiencies, and interaction frequencies for the various conditions discussed in this paper. Video 1 shows the entry event in . Video 2 shows the abortive event in . Online supplemental material is available at .

The endoplasmic reticulum (ER) is a major site of protein folding and assembly in eukaryotes. Polypeptides entering the ER often encounter various folding problems, resulting in aggregated or misfolded proteins. To preserve ER homeostasis, eukaryotes have evolved a conserved quality control pathway termed retrotranslocation, dislocation, or ER-associated protein degradation (ERAD), which efficiently eliminates misfolded ER proteins by exporting them into the cytosol for degradation by the ubiquitin–proteasome system (). Retrotranslocation is initiated when misfolded polypeptides are selectively targeted to the site of translocation at the ER membrane from where they are subsequently dislocated into the cytosol. The export of a subset of substrates requires the Derlin proteins, members of a highly conserved multi-spanning membrane protein family postulated to form a channel (; ; ). Other substrates might use different routes to exit the ER. Most substrates undergoing retrotranslocation are modified by polyubiquitination, which is achieved by the sequential action of a ubiquitin-activating enzyme (E1), a ubiquitin-conjugating enzyme (E2), and a ubiquitin ligase (E3) (). Polyubiquitination occurs at the cytosolic side of the ER membrane after a portion of the substrate has emerged into the cytosol. This modification is required for the subsequent dislocation of polyubiquitinated substrates from the ER membrane, a process that is mediated by the p97–Ufd1–Npl4 ATPase complex (). During Derlin-dependent retrotranslocation, the ATPase complex is recruited to the Derlins together with a single spanning membrane protein called VIMP and certain ER-specific ubiquitin ligases (, ; ). The assembly of a retrotranslocation complex containing both the “pulling” ATPase p97 and the ubiquitination machinery ensures that substrate ubiquitination and dislocation are tightly coupled. For misfolded glycoproteins, their dislocation is also associated with the removal of the attached N-linked glycan by a cytosolic N-glycanase (). After entering the cytosol, polypeptides need to be transferred to the proteasome for degradation. This process is poorly defined. In , the p97 homologue cdc48p does not seem to form a stable complex with the 26S proteasome (). Substrates bound by p97 thus need to be shuttled to the proteasome, which likely involves interactions between ubiquitin conjugates and various ubiquitin binding proteins (). Several proteasome-associated proteins such as rad23p in yeast contain ubiquitin binding motifs (), and can serve as receptors to collect ubiquitinated substrates (). However, little is known about the “shuttling factors” that deliver substrates to these receptors. One candidate for a shuttling factor is ataxin-3 (atx3) (), a poly-glutamine (poly-Q) containing deubiquitinating enzyme (DUB) that interacts with both p97, and HHR23A and B, mammalian homologues of rad23p (; ; ; ). Expansion of the poly-Q segment in atx3 has been linked to a dominantly inherited form of spinocerebellar ataxia, a member of the poly-Q– induced family of neurodegenerative diseases (), but how such poly-Q expansion causes the disease is unclear. In addition, although the enzymatic activity of atx3 has been extensively characterized (; ; ; ), its physiological function remains elusive. This study was initiated to explore the potential involvement of atx3 in ERAD. We conclude that atx3 acts in conjunction with p97 to regulate the degradation of misfolded ER proteins. To test the potential involvement of atx3 in ERAD, we first characterized its interaction with p97 using purified components. Indeed, recombinant protein containing the GST fused to atx3 (GST-atx3) bound purified wild-type (wt) p97 as well as several p97 variants that were defective either in ATP binding (AA and KA) or in ATP hydrolysis (QQ) (). In contrast, a p97 mutant lacking its N-terminal domain (N-domain) failed to bind GST-atx3 (). These results suggest that atx3 binds p97 via its N-domain. Because many cofactors bind the p97 N-domain in a mutually exclusive manner (), we tested whether atx3 would influence the interaction of p97 with Ufd1–Npl4 (U/N), a binary cofactor complex required for ERAD. Recombinant p97 was incubated with purified U/N in the presence of increased concentrations of either GST-atx3 or of a recombinant protein containing the GST fused to SVIP, a small p97 cofactor of unknown function (). The p97–Ufd1–Npl4 complex immunoprecipitated with anti-Ufd1 antibodies was analyzed by immunoblotting. As previously reported (), addition of SVIP resulted in reduced association of p97 with U/N, presumably because these cofactors compete for the same binding site on p97 (, lanes 5–7). In contrast, GST-atx3 did not affect the assembly of the p97–Ufd1–Npl4 complex (, lanes 1–4), suggesting that atx3 either uses a different mode to interact with the p97 N-domain or binds p97 with an affinity significantly lower than other cofactors. The lack of competition between atx3 and U/N raises the possibility that these cofactors may act in conjunction with p97 to regulate ERAD. Therefore, we tested whether atx3 also interacts with other known components of the ERAD pathway, such as the recently identified Derlin–VIMP complex. Detergent extracts of 293T cells expressing FLAG-tagged atx3 were subjected to immunoprecipitation with anti-FLAG antibodies. Immunoblotting with specific antibodies showed that a fraction of Derlin-1, VIMP, and p97 was coprecipitated with atx3 (, lane 4). Endogenous atx3 could be coimmunoprecipitated with Derlin-1 and p97 from detergent extract of dog pancreatic ER membranes (, lane 3). When FLAG-tagged atx3 was coexpressed with Myc-tagged Hrd1, an ER-specific ubiquitin ligase (), both proteins could be coimmunoprecipitated (). Together, these data suggest that a fraction of atx3 is associated with the ER membranes via interactions with components of the retrotranslocation machinery, including Derlin-1, VIMP, p97, and an ER-specific ubiquitin ligase. Because only a small fraction of the Derlin complex could be coprecipitated with atx3, and immunoprecipitation using the VIMP antibody pulled down little atx3 (see ), atx3 appears to transiently interact with the ER membrane (see Discussion). Because knock-down of atx3 by ∼80% using atx3-specific siRNA was not sufficient to cause an apparent defect in retrotranslocation (Fig. S1, available at ), we attempted to interfere with the function of endogenous atx3 with a catalytically inactive atx3 mutant (atx3 C14A). We reasoned that this atx3 variant was likely to exert a dominant-negative effect because it binds ubiquitinated substrates with high affinity, but cannot process them (). We first tested whether the expression of atx3 C14A inhibits the p97-dependent degradation of misfolded TCRα (Fig. S2) (), a type I transmembrane protein targeted for retrotranslocation due to the lack of its assembly partners (). To this end, we expressed TCRα together with either FLAG-tagged wt atx3 or the atx3 C14A mutant in 293T cells. Quantitative immunoblotting using fluorescence-labeled secondary antibodies showed that the FLAG-tagged atx3 variants were expressed ∼12-fold higher than the endogenous protein (unpublished data). When the level of TCRα was determined, it was indeed elevated in cells expressing atx3 C14A (, lane 2 vs. lane 1; , lane 2 vs. lane 1). A pulse-chase experiment confirmed that TCRα was more stable in atx3 C14A–expressing cells (Fig. S3, t ∼60 min in atx3 C14A cells vs. ∼30 min in control cells). Because a significant fraction of the stabilized TCRα appeared to carry polyubiquitin conjugates as determined by immunoblotting the precipitated TCRα with ubiquitin antibodies (, lane 2; , lane 2), the stabilization effect by atx3 C14A was likely underestimated in the pulse-chase experiment because only the nonubiquitinated TCRα species could be analyzed. In contrast to atx3 C14A, overexpression of wt atx3 only moderately stabilized TCRα (Fig. S3), and did not increase the level of ubiquitinated TCRα (, lane 3). atx3 C14A–induced accumulation of ubiquitinated TCRα was more apparent when cells were treated with the proteasome inhibitor MG132 (, lane 4). Although MG132-treated control cells also contained a significant amount of the ubiquitinated TCRα, most substrates in these cells remained nonubiquitinated, which included both the glycosylated retrotranslocation precursor molecules and the dislocated deglycosylated degradation intermediates (, lane 3). In contrast, atx3 C14A–expressing cells treated with MG132 accumulated more ubiquitinated TCRα molecules at a compensatory loss of the nonubiquitinated species (, lane 4 vs. lane 3). These data suggest that the deubiquitination of TCRα is inhibited in atx3 C14A–expressing cells, leading to the stabilization of TCRα in part as polyubiquitinated species. To test for a direct involvement of atx3 in ERAD, we examined its interaction with TCRα. Detergent extracts of cells expressing HA-tagged TCRα together with either wt atx3 or atx3 C14A were subjected to immunoprecipitation with HA antibodies. Immunoblotting analysis showed that only a small fraction of wt atx3 was coprecipitated with TCRα (, lane 3), whereas significantly more atx3 C14A was bound to TCRα (, lane 2). These results are consistent with the concept that the atx3 C14A mutant is able to sequester its substrates in a stable complex. Treatment of cells with MG132 stabilized TCRα (, lanes 4–6 vs. lanes 1–3), and further enhanced the interaction between TCRα and atx3 C14A (, lane 5 vs. lane 2). In contrast, the association of TCRα with wt atx3 remained weak (, lane 6). These data suggest that atx3 transiently binds TCRα to directly regulate its degradation. To see whether atx3 also plays a role in degradation of soluble ERAD substrates, we examined the degradation of misfolded β-site amyloid precursor protein cleaving enzyme (BACE457Δ) () in cells expressing either wt atx3 or atx3 C14A. Similar to TCRα, BACE457Δ was stabilized in atx3 C13A–expressing cells, and a significant fraction of the accumulated BACE457Δ contained polyubiquitin conjugates (, lane 3). In contrast, expression of wt atx3 only moderately increased the level of nonubiquitinated BACE457Δ (, lane 2 vs. lane 1). Thus, atx3 is also involved in the degradation of lumenal ERAD substrates. Because a defect in ERAD is usually associated with the accumulation of misfolded proteins in the ER, which triggers ER stress, we monitored the ER stress level in cells expressing atx3 C14A to further confirm its involvement in ERAD. Indeed, expression of atx3 C14A caused strong induction of an ER stress reporter gene, pGL3-GRP78 (-132)-luciferase (). Expression of wt atx3 also induced ER stress, albeit to a much lesser degree. This is consistent with the observation that wt atx3 caused a moderate stabilization of TCRα and BACE457Δ. Together, these results further support the notion that the deubiquitinating activity of atx3 may be essential for efficient elimination of misfolded ER proteins. The inhibition of ERAD by atx3 C14A cannot be attributed to general perturbation of the ubiquitin proteasome system by the overexpressed defective DUB, because the proteasome-dependent degradation of an N-end rule substrate Ub-R-GFP () was not affected (). Interestingly, the degradation of Ub-R-GFP could be inhibited by a p97 mutant that was defective in ATP hydrolysis (QQ). Thus, the inhibitory effect of atx3 C14A on protein degradation cannot be simply due to nonselective disruption of p97 function by the overexpressed mutant protein. Because atx3 is a p97-associated DUB (), it may facilitate ERAD by promoting deubiquitination of substrates bound by p97. To examine p97-associated deubiquitination (PAD), p97 and its associated proteins were immunoprecipitated from cell extracts and analyzed by immunoblotting. A fraction of the precipitated material carried polyubiquitin conjugates. Because these ubiquitinated proteins were not detected when immunoprecipitation was performed under denaturing condition, they were likely substrates bound to p97 (Fig. S4, available at ). Next, we tested whether these ubiquitinated proteins could be subjected to deubiquitination in vitro by DUBs coprecipitated with p97. Indeed, the amount of p97-bound ubiquitinated material was slowly reduced during incubation (, lanes 1–4), a process that could be blocked by -ethylmaleimide, an inhibitor of deubiquitination (unpublished data). To test whether atx3 can facilitate this process, p97 and its associated substrates purified from cells expressing either wt atx3 or atx3 C14A mutant were tested for in vitro deubiquitination. Expression of wild type, but not the catalytically inactive atx3 mutant, accelerated the decay of these ubiquitin conjugates (, lanes 5–8 vs. lanes 1–4; ; Fig. S5). These data indicate that atx3 indeed facilitates PAD. Notably, a smaller amount of ubiquitinated proteins was coprecipitated with p97 from cells expressing wt atx3 (, lane 5 vs. lane 1; , lane 3 vs. lane 1), suggesting that atx3-mediated PAD may also occur in intact cells. Expression of USP14, a DUB primarily associated with the proteasome (; ), did not decrease the amount of ubiquitinated proteins bound to p97 (, lane 5 vs. lane 1), and did not increase the rate of PAD in vitro (, lanes 5 and 6). Together, these data suggest that atx3 can promote PAD both in intact cells and in vitro. To understand how atx3 cooperates with p97 to promote deubiquitination, we wished to determine the order between atx3-mediated deubiquitination and p97-dependent ATP hydrolysis. We reasoned if atx3 acts subsequent to the ATPase cycle of p97, inhibition of p97 ATPase activity would likely prevent ubiquitinated substrates from reaching the enzymatic site of atx3 because substrates cannot be released from p97 under this condition (). To test this idea, FLAG-tagged wt atx3 was coexpressed together with either His-tagged wt p97 or a p97 mutant defective in ATP hydrolysis (QQ) (). The amount of ubiquitinated substrates associated with His-tagged p97 was determined. The expression of atx3 reduced the amount of ubiquitinated proteins associated with wt p97 (, lane 2 vs. lane 1), but not of those bound by p97 QQ (lane 4 vs. lane 3), although more atx3 was bound by p97 QQ (lane 4 vs. lane 2). Moreover, the expression of p97 QQ resulted in more ubiquitinated proteins in association with atx3 and p97 (, lane 5 vs. lane 1), and these ubiquitinated proteins were more resistant to deubiquitination in vitro (, lanes 5–8 vs. lanes 1–4; ). These data indicate that ubiquitinated substrates accumulated upon p97 QQ expression cannot be efficiently deubiquitinated by atx3. Although these substrates can be coimmunoprecipitated with atx3, they are likely still bound by the p97 QQ mutant, which is also present in the complex. Thus, ATP hydrolysis by p97 appears to precede atx3-mediated deubiquitination. Perhaps p97-dependent ATP hydrolysis is coupled to the transfer of substrates to the enzymatic site of atx3. Because the p97 ATPase cycle is required for the extraction of misfolded proteins from the ER membranes (), atx3 likely acts at a step downstream of dislocation. We noticed that more atx3 was bound to the p97 QQ than to wt p97 (, lane 4 vs. 2; ). In addition, an increased association between atx3 and VIMP was observed in cells expressing p97 QQ (, lane 3 vs. lanes 1 and 2). Consequently, more atx3 was found in association with the ER membrane in these cells (, lane 5 vs. lanes 1 and 3). These data indicate that the association of atx3 with p97 and VIMP at the ER membrane may be regulated by the p97 ATPase cycle, a conclusion that is also supported by the recent observation that the interaction of p97 with atx3 in brain extracts is enhanced when ATP or ATPγS is present (). Because atx3 bound purified wt p97 and the p97 QQ mutant with similar affinity in vitro (), the enhanced interaction between atx3 and p97 QQ mutant may require additional factors present in cells. Interestingly, mutant atx3 lacking its deubiquitinating activity (C14A) also exhibited an increased affinity to p97, VIMP, and Derlin-1 (Fig. S5; ). Consistent with this finding, atx3 C14A also accumulated at the ER membrane to a much higher level than wt atx3 (, lane 3 vs. lane 1). The simplest explanation of these observations is that a small fraction of atx3 is transiently associated with the ER membrane via interactions with p97 and VIMP. atx3 may shuttle on and off p97 to promote PAD, a process that is linked to the ATPase cycle of p97 and to the subsequent delivery of substrates to the proteasome. Mutations affecting the enzymatic activity of either p97 or atx3 appear to interfere with the release of atx3 from p97, which may also prevent substrates from leaving p97 (see following paragraph). To test whether atx3 C14A inhibits ERAD by interfering with the transfer of substrates from p97 to the proteasome, we determined the relative amount of ubiquitinated substrates bound to p97 versus those bound by the proteasome. We reasoned that a defect in substrate transfer would lead to an accumulation of ubiquitinated substrates in complex with p97. Detergent extracts of 293T cells were subjected to immunoprecipitation with antibodies to either p97 or α6, a subunit of the 20S proteasome. Immunoblotting showed that the α6 antibodies precipitated α7, another subunit of the 20S proteasome, together with Mms1, a subunit of the 19S particle, but not p97. Likewise, antibodies against p97 precipitated the ATPase itself, some Mms1, and no α7 (). These data confirm that p97 and the 26S proteasome do not form a stable complex. When ubiquitinated proteins were examined, control cells contained little polyubiquitinated proteins, and they were evenly distributed between the proteasome and p97 (, lane 2 vs. lane 1). In contrast, atx3 C14A–expressing cells contained more ubiquitinated proteins, and they were almost entirely associated with p97 (, lanes 3 and 4; ). The increased association of p97 with ubiquitinated proteins cannot be simply due to the accumulation of ubiquitinated proteins in cells expressing a defective DUB, because overexpression of the catalytically inactive USP14 mutant (USP14 C114A) also increased the level of ubiquitinated proteins, but did not cause an apparent accumulation of ubiquitinated proteins in complex with p97 (). These data indicate that atx3 C14A likely blocks the transfer of substrates from p97 to the proteasome, leading to their stabilization. We have investigated the role of atx3 in retrotranslocation of misfolded ER proteins. We show that atx3 interacts with the recently identified retrotranslocation complex consisting of Derlin-1, VIMP, certain ER-specific ubiquitin ligases, and the pulling ATPase p97 in the ER membrane. A direct involvement of atx3 in ERAD is suggested by its binding to model ERAD substrates en route to the proteasome. Moreover, expression of a mutant atx3 defective in deubiquitination (atx3 C14A) dramatically inhibits the degradation of both membrane and lumenal ERAD substrates. Consequently, ER stress is induced in these cells. ERAD substrates stabilized by mutant atx3 are accumulated in part as polyubiquitinated degradation intermediates, suggesting that atx3 acts on these substrates to remove their ubiquitin conjugates. Intriguingly, ubiquitinated proteins accumulated upon mutant atx3 expression are sequestered in complex with p97, raising the possibility that atx3-mediated deubiquitination may be coupled to the transfer of substrates from p97 to the proteasome. The observation that overexpression of wt atx3 moderately inhibits ERAD is unexpected, but is not without precedence. In fact, overexpression of rad23, an atx3-interacting partner that facilitates the delivery of ubiquitinated proteins to the proteasome (), also inhibits protein turn-over. Recent studies suggest that the effect of rad23 on protein degradation is highly dependent on its concentration, with low concentrations being stimulatory and high concentrations inhibitory (). The inhibition of ERAD by wt atx3 may be similarly caused by its overexpression, which may titrate out certain factors essential for ERAD. Alternatively, the deubiquitinating activity of atx3 may need to be tightly regulated in conjunction with other reactions to mediate the degradation of ERAD substrates (see following paragraph). The presence of excessive atx3 may uncouple deubiquitination from these processes, which may cause a defect in ERAD. We propose that atx3 may promote deubiquitination of p97-bound substrates to facilitate their transfer to the proteasome during retrotranslocation (). In this model, atx3 would first interact with p97, which is prebound to its substrates together with Derlin-1 and VIMP at the ER membrane. In the next step, substrates may be transferred from p97 to atx3, a process that is likely associated with the p97 ATPase cycle. Next, atx3 would act on the ubiquitinated substrates, which might be subsequently transferred to a downstream ubiquitin receptor such as rad23. Because ubiquitin chains consisting of at least four ubiquitin molecules are required for efficient targeting of modified substrates to the proteasome, deubiquitination of substrates en route to the proteasome by atx3 is likely tightly coupled to the transfer process to ensure that the targeting signal is not prematurely erased. In , ubiquitin chains on substrates of cdc48p appear to be progressively shortened while the substrates are in transit to the proteasome (). In light of this finding, we speculate that atx3 may function as an editing enzyme to trim ubiquitin conjugates on substrates to guide them to the proteasome. Although atx3 immunoprecipitated from mammalian cells appears to deubiquitinate its substrates completely, recombinant atx3 purified from bacteria indeed exhibits chain trimming activity, leading to a shortening of ubiquitin chains rather than to their complete removal (). Because atx3 binds the ubiquitin-like domain of rad23, which also mediates the interaction between rad23 and the proteasome (), it is likely that rad23 may not interact with atx3 and the proteasome simultaneously. Our analysis demonstrates a physiological function for atx3. We postulate that the expression of the poly-Q–expanded atx3 mutants may also impair ERAD and trigger ER stress, which may contribute to the pathogenesis of spinocerebellar ataxia. Further analysis of these mutants will help to clarify the long-suspected connection between ER stress and neurodegenerative diseases. The plasmids pcDNA3.1 His-p97 wt and pcDNA3.1 His-p97 QQ were described previously (). The plasmid pRK-FLAG atx3 wt (consisting of 20 CAG repeat) was constructed by cloning the human atx3 cDNA (a gift from R. Pittman, University of Pennsylvania, Philadelphia, PA; ) into the SalI and NotI sites of the pRK-FLAG vector (a gift from H. Shu, the National Jewish Medical Center, Denver, CO). pRK-FLAG-USP14 was constructed by cloning the human USP14 cDNA into the SalI and NotI sites of the pRK-FLAG vector. pRK-FLAG-atx3 C14A and pRK-FLAG-USP14 C114A were constructed by site-directed mutagenesis using the QuikChange mutagenesis kit from Stratagene. All plasmids were sequenced. HA-tagged TCRα clone was a gift from R. Kopito (Stanford University, Stanford, CA; ). pGL-3-GRP78 (-132) luciferase plasmid was provided by K. Mori (Kyoto University, Kyoto, Japan). Plasmids expressing Ub-R-GFP, GST-SVIP, and BACE457Δ were provided by N. Dantuma (Karolinska Institute, Stockholm, Sweden), M. Nagahama (Tokyo University of Pharmacy and Life Science, Tokyo, Japan), and M. Molinari (Institute for Research in Biomedicine, Bellinzona, Switzerland), respectively. Anti-BACE antibody was a gift from P. Paganetti (Novartis Pharma AG, Basel, Switzerland). Antibodies to ubiquitin, p97, Ufd1, and VIMP were described previously (, ). Rabbit polyclonal antibodies to Derlin-1 were raised against a peptide corresponding to residues 238–251 of the human Derlin-1. All antibodies used in this study were affinity purified using the corresponding antigenic peptides. FLAG, HA, and GFP antibodies were purchased from Sigma-Aldrich, Roche, and Invitrogen, respectively. MG132 was purchased from Calbiochem. The purification of His-tagged p97 variants and the Ufd1–Npl4 complex has been described previously (). GST-atx3 was purified according to standard procedure using glutathione beads. GST pull-down experiments were performed as described previously (). In brief, GST fusion proteins (5 μg) bound to glutathione beads were incubated with various purified recombinant proteins in 50 mM Hepes (pH 7.3), 150 mM potassium chloride, 2.5 mM magnesium chloride, 5% glycerol, and 0.1% Triton X-100. Bound material was analyzed by SDS/PAGE and stained with Coomassie blue. 293T cells were maintained according to standard procedures. Transfections in 293T cells were done with Trans IT (Mirus). For fractionation experiments, cells were homogenized by passing through a 25G needle in a low salt buffer (50 mM Hepes, pH 7.4, 10 mM potassium chloride, 1 mM DTT, and a protease inhibitor). Cell extracts were then subjected to sequential centrifugation at different speeds. For coimmunoprecipitation experiments, tissue culture cells or dog pancreatic microsomes were extracted with buffer N (30 mM Tris/HCl, pH 7.4, 150 mM potassium acetate, 4 mM magnesium acetate, 1% DeoxyBigCHAP, and a protease inhibitor cocktail). To detect ubiquitinated TCRα, cell extracts were adjusted to contain 0.2% SDS, 0.5 mM DTT, and 5 mM -ethylmaleimide. Samples were heated at 95°C before being subjected to immunoprecipitation. To isolate the 26S proteasome, 2 mM ATP was included in the extraction buffer to preserve the interaction between the 19S and the 20S proteasome particles. All cell extracts were cleared by centrifugation (20,000 ) and subjected to immunoprecipitation with various antibodies. Immunoblots were visualized with a cooled CCD digital camera system (LAS-3000; Fuji). Results were quantified using MultiGauge v3.0 software (Fuji). To detect p97 or atx3-associated deubiquitination, cells were extracted with buffer N. Cell extracts were subjected to immunoprecipitation with either p97 antibodies or anti-FLAG antibody to isolate p97 or FLAG-tagged atx3. Proteins bound to the beads were incubated in a deubiquitinating buffer (50mM Tris/HCl, pH 7.4, 20mM potassium chloride, 5mM magnesium chloride, 1mM DTT, 2.5% BSA) at 37°C. Deubiquitination was stopped by addition of SDS sample buffer. Samples were analyzed by immunoblotting. Fig. S1 shows that knock-down of atx3 by ∼80% by siRNA has little effect on the degradation of TCRα. Fig. S2 shows that the degradation of TCRα requires p97. Fig. S3 provides evidence that the accumulation of TCRα in atx3 C14A–expressing cells is due to its stabilization. Fig. S4 shows the association of p97 with ubiquitinated proteins. Fig. S5 demonstrates that an atx3 mutant defective in deubiquitination fails to promote PAD. Online supplemental material is available at .

Vesicular trafficking provides a continuous exchange of proteins and lipids between membranes in a eukaryotic cell (for review see ; ). Coat proteins are believed to confer much of the specificity associated with protein sorting into transport vesicles (,; ). To date, three classes of coated vesicles have been identified: clathrin/adaptor-coated vesicles mainly involved in traffic between the TGN and the endosomes (); coat protein complex I (COPI), which is responsible for both retrograde transport from the Golgi back to the ER and intra-Golgi transport (; ); and COPII, which mediates anterograde transport from the ER to the Golgi apparatus (; ). Coat assembly is initiated by activation of the ADP ribosylation factor (ARF) family of small G proteins (Arf1p and the closely related Sar1p) by which membrane-selective nucleotide exchange catalysts activate Arf1p or Sar1p for membrane attachment (; ; ; ). Arf1p regulates the recruitment of COPI and most clathrin-containing coats, leading to membrane deformation into coated buds and vesicles. Likewise, COPII vesicles form when Sar1p-GTP recruits the inner coat complex (the Sec23/24p heterodimer) and the outer coat (the Sec13/31p heterotetramer; ; ; ). Although coat proteins account for much of the vesicular traffic in a cell, no such involvement of coat proteins has been documented in the formation of vesicles or tubules that convey membrane and secretory proteins directly from the TGN to the cell surface. As an example of this limb of the secretory pathway, we have studied the transport of a cell wall biosynthetic enzyme, Chs3p (chitin synthase III), from the TGN/endosome membranes to the plasma membrane of the mother–bud junction in yeast. Chs3p is a multispanning transmembrane protein that is required for chitin ring formation during the G1 phase of the cell cycle and, subsequently, in lateral cell wall chitin synthesis (). Unlike other cell surface proteins, Chs3p export is regulated in response to cell cycle and stress signals (; ). However, throughout the cell cycle, it is maintained in an intracellular reservoir by being recycled between the TGN and the early endosomes. This recycling is mediated by clathrin and an adaptor protein complex (AP-1). Chs5p and Chs6p are peripheral proteins that are required to transport Chs3p from the reservoir to the cell surface (; ). and mutants accumulate Chs3p in the TGN/endosome membranes, and the deletion of clathrin or subunits of AP-1 restores Chs3p traffic to the cell surface by some unknown bypass pathway (). Thus, at least two mechanisms of traffic from the TGN/endosome membrane to the cell surface are possible for Chs3p. Each pathway involves an unexplored protein-sorting event that packages Chs3p into secretory vesicles that are delivered to the bud plasma membrane by the standard secretory pathway (). The Chs5 and Chs6 proteins are restricted to yeast and fungi, which may imply an organism-specific role such as the biosynthesis of yeast cell wall chitin. Yeast cells have three additional Chs6-like proteins (Bch1p [], Bud7p, and Bch2p []), and their roles relative to Chs6p and the traffic of Chs3p are not yet understood (Satchatjate and Schekman, 2006; ). To pursue the role of Chs5p and Chs6p in the sorting and packaging of Chs3p, we have cloned and characterized the gene products, including the three paralogues of Chs6p. We have demonstrated that these Chs5 and Chs6 proteins are associated with each other in a complex that may make direct contact with Chs3p (). In this study, we identified Sec7p, a TGN-localized Arf1p nucleotide exchange factor that is required for the membrane association of Chs5p and regulating the interaction of Chs5p and Arf1p. We elucidated the biochemical requirements for membrane recruitment of a complex of Chs5p, Chs6p, and the Chs6p paralogues and describe a novel coat structure termed exomer, which forms when the complex is recruited to synthetic membranes in the presence of Arf1p and GTPγS. In a previous study, found Chs5p localized to puncta in cells marked by the TGN/endosome resident protein Kex2p. We confirmed the late Golgi localization of Chs5p-RFP by comparing its localization in vivo with Sec7p-GFP (late Golgi marker) and Anp1p-GFP (early Golgi marker) using double staining live cell imaging (Fig. S1, A–C; available at ). To determine whether this localization is perturbed when Golgi traffic is disrupted, we used a functional, integrated Chs5p-GFP to examine its localization in wild type and in cells defective in secretory protein traffic from the Golgi complex. In wild-type cells, Chs5p-GFP localized to spots dispersed within the cytoplasm, as was determined previously (), and this distribution was similar at 26 and 37°C. We also expressed Chs5p-GFP in two temperature-sensitive mutants in which the secretory function of the Golgi apparatus can be altered. A temperature-sensitive allele of the () gene blocks protein secretion and accumulates exaggerated Golgi structures (; ; ; ). Similarly, temperature-sensitive alleles of arrest secretory traffic and accumulate large Golgi stacks at a restrictive temperature (; ). In the strain at 37°C, Chs5p-GFP coalesced into large punctae, as was previously observed for the Golgi marker Och1p (; ). In contrast, Chs5p-GFP dispersed in a diffuse pattern in cells incubated at 37°C, although Chs5p-GFP localized normally at 26°C in both mutant strains (). The Golgi marker proteins Anp1p-RFP and Kex2p-GFP accumulated in exaggerated structures at 37°C in , indicating that the Golgi membrane did not disperse (Fig. S1 D). Attempts to localize Chs6p-GFP were unsuccessful because the fluorescence of Chs6p-GFP was too weak for microscopy as a result of a low expression level (unpublished data). encodes a nucleotide exchange factor for Arf1p. One explanation for the difference between and is that the mutation resides within the conserved Arf1 guanine nucleotide exchange factor (GEF) domain (Sec7 domain; ). Thus, we considered the possibility that Chs5p was restricted to Golgi membranes through an interaction with activated Arf1p. To test this possibility, we created a strain harboring protein A fused to the C-terminal codon of the chromosomal copy of . A cytosol fraction obtained from this strain was incubated with GTP or one of two nonhydrolyzable analogues, GMP-PNP and GTPγS. Arf1p–protein A was recovered by binding to IgG-coated Dynabeads, and samples corresponding to the input, unbound wash fraction and bound (bead) materials were separated and evaluated by SDS-PAGE followed by Arf1p and Chs5p immunoblotting. A clear copurification of Chs5p with Arf1p was detected in incubations containing GTPγS and, to a lesser extent, with GMP-PNP (). Nucleotide hydrolysis in the GTP sample may explain the poor binding of Chs5p to Arf1p. To further confirm the nucleotide-dependent interaction between Chs5p and Arf1p–protein A and to support the role of Sec7p in regulating this association, we next evaluated the recovery of bound Chs5p in incubations containing the ARF GEF inhibitor brefeldin A (BFA). We observed a reduction in the retention of Chs5p in an incubation containing GTPγS and 2 μg/ml BFA (). A similar effect of BFA in vivo has been reported by . These experiments suggest that activated Arf1p-GTP, presumably by contact with Sec7p, recruits Chs5p to the Golgi membrane to initiate its role in the traffic of Chs3p. In addition to binding to activated Arf1p, Chs5p may interact by additional contact with the Golgi membranes. To determine whether Chs5p interacts with lipids, we expressed GST hybrid forms of Chs5p and Chs6p in , and the purified GST hybrid proteins were probed for lipid interaction using an overlay assay on PIP strips. GST-Chs5p but not GST or GST-Chs6p showed significant interaction with most anionic lipids (unpublished data). A similar spectrum was seen with 6× His-tagged Chs5p expressed in yeast (unpublished data). Fragments of Chs5p were fused to GST to define the region interacting with lipids (). The potential anionic lipid interaction domain was mapped to the C-terminal one third of Chs5p. This region may correspond to the C-terminal lysine-rich tail (). Although the C-terminal domain may facilitate the membrane recruitment of Chs5p, it appears to be dispensable for chitin synthesis because a truncated version of corresponding to amino acid residues 1–401 complemented a strain based on the growth sensitivity of yeast cells making chitin to the chitin-binding dye calcofluor (unpublished data). To understand the interaction between Chs5p and Chs6p, fragments of Chs5p used in the lipid-binding assay were evaluated for interaction with Chs6p using the yeast two-hybrid assay. Binding domain–Chs5p (aa 1–79) and binding domain–Chs5p (aa 1–260) showed two-hybrid interaction with activation domain–Chs6p (). These two domains (the lipid-binding domain within the C terminus of Chs5p and the Chs6 interaction domain within the N terminus of Chs5p) were compared by a competition assay evaluating their functional importance in chitin synthesis. The overexpression of GST hybrids containing the N-terminal domain of Chs5p interfered with Chs3p traffic as judged by the calcofluor growth test, whereas a GST hybrid containing the C-terminal domain of Chs5p did not impair chitin synthesis (). Thus, the interaction of Chs5p and Chs6p may be crucial for the transport of Chs3p. To further investigate the role of activated Arf1p in Chs5p membrane recruitment (), we sought to isolate the Chs5p- and Chs6p-containing complex () for functional tests. Efforts to express the proteins in stable oligomeric forms in and yeast resulted in poor yields. However, expression in baculovirus proved more reliable. We created baculovirus vectors containing N-terminally 6× His-tagged Chs5p and untagged versions of one or more copies of Chs6p and its paralogues (Bch1p, Bud7p, and Bch2p; ). Recent evidence has suggested that Chs5p interacts with Chs6p and each of the Chs6 paralogues and that complexes include more than one copy of Chs6p and its paralogues (; Trautwein et. al., 2006). The baculovirus system allowed us to evaluate complex formation by coexpressing multiple combinations of these recombinant Chs5p and Chs6p proteins. Consistent with previous observations (; Trautwein et. al., 2006), we found that Chs5p copurified with each Chs6p or its paralogue when the two were coexpressed and that multiple paralogues of Chs6p copurified with Chs5p from cells expressing two, three, or all Chs6p and Chs6p paralogues ( and not depicted). In most cases, the apparent abundance of Chs5p, based on Sypro red staining intensity, approximated the abundance of the sum of the Chs6 and Chs6p paralogues. Bch2p was a notable exception, perhaps because of a lower virus titer. Infection with a larger Bch2 baculovirus stock increased the abundance of this species in the isolated Chs5p complex (unpublished data). The ratios of the Chs6p species in the complex differed from those detected in the complex isolated from wild-type yeast cells (), probably reflecting the different level of gene expression in these circumstances. The recombinant expression of Chs6p and its paralogues without Chs5p also resulted in complexes including Chs6p and one or more Chs6p paralogues (unpublished data). Affinity-purified complexes were evaluated by gel filtration on a Superose 6 fast protein liquid chromatography (FPLC) column to determine the size and composition of the complex. Complex isolated from cells expressing Chs5p and all four Chs6p and its paralogues (Chs5–Chs6[all]) fractionated at a position consistent with a size slightly >1 MD with coincident chromatography of the most abundant Chs6p species (). Similar patterns of filtration were seen with complexes containing only one Chs6p paralogue, although a complex of Chs5p and Chs6p fractionated somewhat heterogeneously. An independent method of separation, velocity sedimentation on a sucrose density gradient, confirmed that Chs5p, Chs6p, and Chs6p paralogues (Chs5–Chs6[all]) were present in a large complex (). This pattern of cosedimentation was not altered in the Chs5–Chs6[all] samples treated with 3 M urea, 1% Triton X-100, or 1 M KCl (unpublished data). Overall, recombinant Chs5p, Chs6p, and Chs6p paralogues, like those isolated from wild-type yeast cells, appear to be self-organized into a large and stable complex. Because genetic evidence suggests that more than just one of the Chs6p and its paralogues is required for traffic of Chs3p to the cell surface (; ), we used the complex including Chs5p, Chs6p, and three Chs6 paralogues (Chs5–Chs6[all]) to evaluate the role of activated Arf1p in membrane recruitment. Recombinant myristoylated Arf1p (mArf1p; Q71L; GTPase deficient) was purified from and mixed with GTPγS and liposomes formulated with synthetic phospholipids with various levels of selected acidic phospholipids (; ). EDTA was included to stimulate spontaneous GTP/GDP exchange on Arf1p (). After 1 h at 30°C, MgCl was added to stabilize Arf1-GTPγS, and the samples were supplemented with Chs5–Chs6[all] and incubated for a further 10 min at 22°C. For each liposome formulation, three samples were prepared: complete, without Arf1p, and without GTPγS. Liposomes and bound proteins were separated from unbound materials by flotation sedimentation on a sucrose density shelf. documents the recovery of proteins bound to liposomes, and displays a quantitative representation of proteins recovered in the floated fractions. In most samples, Arf1p bound to liposomes in the presence or absence of GTPγS. However, recruitment of the Chs5–Chs6[all] complex was optimum in incubations that contained Arf1p and GTPγS. Certain liposome formulations (e.g., phosphatidylcholine/phosphatidylethanoamine [PE]/phosphatidylserine/phosphatidic acid [PA] with a high concentration of PIP or phosphatidylinositol-4,5-bisphosphate) recruited Chs5–Chs6[all] in the absence of Arf1p or GTPγS (). Other formulations (e.g., major-minor mix optimized for COPII assembly; Matsuoka et. al., 1998) displayed a substantial GTPγS requirement for recruitment of the Chs5–Chs6[all] complex. From these results, we conclude that Arf1p binds directly to the Chs5–Chs6[all] complex and facilitates the association of the complex with liposomes, particularly with certain formulations containing one or more of several acidic phospholipids. The major-minor mix formulation, which was optimized for the recruitment of activated Sar1p and COPII to liposomes () and which also works well for the recruitment of activated Arf1p and coatomer (COPI; ), was reexamined for wild-type Arf1p and Chs5–Chs6[all] assembly in the presence of various nucleotides. shows a substantial dependence on GTPγS (or GMP-PNP), with much less of the complex recruited in the presence of GDP or GTP. GTP hydrolysis during recruitment and liposome sedimentation may explain the failure to retain comparable amounts of the Chs5–Chs6[all] complex on liposomes incubated with Arf1p-GTP. Likewise, a Golgi-enriched membrane fraction incubated with the Chs5–Chs6[all] complex confirmed that the binding of Chs5–Chs6[all] in a reaction was stimulated by GTPγS (unpublished data). We next compared recruitment of the Chs5–Chs6[all] complex to the other subcomplexes formed with Chs5p, Chs6p, or fewer paralogues of Chs6p (). Chs5p alone was very inefficiently recruited to major-minor liposomes. Combinations including one paralogue were recruited in a manner largely independent of GTPγS. However, two paralogues known to be important in Chs3p traffic (Chs6p and Bch1p) were recruited to membranes in an Arf1p-GTPγS–dependent manner comparable with Chs5–Chs6[all] (). To provide evidence that the recruitment is under control by a specific, regulated process, we varied the concentration of Arf1-GTPγS and the Chs5–Chs6[all] complex and the time of incubation to discover the optimum conditions of assembly on major-minor mix liposomes. Two-stage recruitment assays were performed. Liposomes were incubated with ∼1 μM mArf1p and GTPγS in a first-stage binding reaction as in and were mixed with the Chs5–Chs6[all] complex for 15 min at 22°C. Membrane-bound protein complexes were collected by flotation on a step sucrose cushion, and the protein content was measured by Sypro red staining of gels (). Chs5–Chs6[all] complex binding to liposomes was saturated at ∼0.5 μM in this experiment. Conversely, at a fixed Chs5–Chs6[all] concentration of ∼0.8 μM, Arf1p binding increased nonsaturably (, ii), whereas binding of the Chs5–Chs6[all] complex appeared to approach saturation at around 1.5 μM Arf1p (, i). Excess bound Arf1p may not be functional or accessible to the Chs5–Chs6[all] complex. Titration at the low range of Arf1p, where the linear membrane association of Arf1p and Chs5p was shown (, iii), indicated an ∼7.5:1 molar stoichiometry of the Arf1/Chs5–Chs6[all] complex, reflecting either a substantial fraction of bound Arf1p that is not accessible to the Chs5–Chs6[all] complex or incomplete density gradient recovery of Chs5–Chs6[all] compared with activated Arf1p bound to liposomes. The relationship of Arf1p exchange and Chs5–Chs6[all] recruitment in a reaction was evaluated in a time course experiment. Protein binding to liposomes was conducted with Arf1p at approximately the maximum Chs5–Chs6[all] ratio achieved in the titration experiment in . Incubations included EDTA to promote Arf1p nucleotide exchange. Samples were collected at the indicated times at 30°C, mixed with MgCl, and chilled on ice for the duration. The kinetics of binding paralleled the rate of EDTA-stimulated Arf1p nucleotide exchange as measured by the change in tryptophan fluorescence of activated Arf1p ( and not depicted; ). These results support our conclusion that GTPγS-activated Arf1p recruits the Chs5–Chs6[all] complex to the membrane. Given the similarity between the Chs5–Chs6[all] complex and the COPs (coatomer and COPII) in regard to Arf1p-GTPγS (or Sar1p–GMP-PNP)–dependent recruitment to liposomes, we examined the influence of this assembly on the buoyant density of liposomes. COPII proteins that assemble on liposomes in the presence of Sar1p–GMP-PNP cause membranes to shift to a higher buoyant density, reflecting the formation of protein-coated surfaces and synthetic COPII vesicles (). We formulated major-minor mix liposomes with Texas red–PE and conducted assembly incubations for the complete reaction containing Arf1-GTPγS and Chs5–Chs6[all] as described above. Samples were applied to a 10–50% linear sucrose gradient and centrifuged to equilibrium for 16 h at 4°C. In control incubations, in samples of liposomes incubated with the Chs5–Chs6[all] complex without Arf1p or GTPγS or liposomes incubated with Arf1p and GTPγS alone (unpublished data), fluorescent lipid peaked at the top of the gradient (fraction 1). In a complete reaction, most of the fluorescent lipids and Arf1p sedimented into the gradient but remained in the low density liposome fractions (, C and D; fractions 2–8), whereas most of the unbound Chs5–Chs6[all] complex sedimented to a high density with no apparent lipid cofractionation (). Chs proteins and Arf1p were detected in fractions 2–8 but not in a sample of Chs5–Chs6[all] incubated with liposomes in the absence of Arf1p (). Thus, at least some liposome-bound Chs proteins may coat membranes sufficiently to influence membrane buoyant density. The unambiguous assignment of a membrane coat requires inspection by thin section electron microscopy. Samples were prepared from a complete incubation (), an incubation of liposomes and Arf1p-GTPγS alone (), and an incubation with Arf1p, Chs5–Chs6[all] complex, and GDP (). A spiky coat appeared uniformly distributed along the surface of liposomes incubated under conditions in which the Chs5–Chs6[all] complex binds to the membrane but not in the control conditions. Although some membrane profiles appeared elongated, no coated buds or small coated vesicles were evident. Thus, the Chs5–Chs6[all] complex, which we now call exomer, appears to form coats on the membrane. However, unlike the COPs, exomer by itself does not deform membranes to induce the formation of buds and transport vesicles. Vesicular traffic in several limbs of the secretory pathway is initiated by the GTP-binding protein-dependent recruitment of coat proteins that sequester cargo molecules in a bud and pinch the membrane to form a transport vesicle. Clathrin, COPI, and COPII are well-known examples of this seemingly general feature (for review see ; ). However, one clear gap in our knowledge concerns the mechanism of sorting and transport of membrane and secretory proteins from the TGN to the cell surface. Although some proteins use clathrin to traverse the endosome en route to the cell surface (), others do not, and, until now, the general view has been that the direct path out of the TGN may involve tubular carriers formed without the intervention of coat proteins. Some regulatory proteins control the formation of transport carriers at the TGN by modulating lipid composition (for example, activation of phospholipase D by protein kinase C; ). Other proteins, such as FAPPs (four-phosphate adaptor proteins), are phosphoinositide-binding proteins that are found associated with TGN carriers and also interact with Arf1p (). Protein kinase D is recruited to the TGN through interaction with diacylglycerol and is subsequently activated by phosphorylation to promote carrier fission (). We have investigated traffic of the plasma membrane enzyme that makes the chitin ring at the mother–bud junction in growing yeast cells. This protein, Chs3p, has an interesting itinerary and set of genetic requirements for its traffic that are somewhat distinct from the requirements of other cell surface proteins. Specifically, the action of two peripheral membrane proteins, Chs5p and Chs6p, suggest a special machinery to convey Chs3p from the TGN to the cell surface (). Chs5p and Chs6p form subunits of a complex, including one or more additional Chs6p paralogues that also facilitate the traffic of Chs3p (; Trautwein et. al., 2006). Chs5p is required for the transport of at least two other proteins in addition to Chs3p, one of which, Fus1p, a cell surface protein required for yeast cell fusion, does not depend on Chs6p (). By analogy to the paralogues of Sec24p that form dimers with Sec23p in the COPII coat (), we propose that Chs5p, Chs6p, and its paralogues may recognize distinct sets of membrane cargo proteins at the TGN for packaging into mature secretory vesicles. Our strategy has been to identify conditions that promote the recruitment of Chs5p and Chs6p to the TGN/endosomal membrane and then to reconstruct this interaction with synthetic membranes. Two lines of evidence support a role for activated Arf1p in Chs5p–Chs6p assembly at the TGN. found that Arf1p-GTPγS binds Chs5p and Chs6p in a crude lysate of yeast. In vivo, nucleotide exchange on Arf1p is promoted by several proteins sharing a GEF domain that was first identified in a peripheral Golgi protein, Sec7p, which is required for secretory traffic within the Golgi complex. Mutations within the Sec7 GEF domain block traffic of all or most secretory proteins and disperse Chs5p from its normal punctate TGN/endosomal localization (). BFA blocks GEF activity and reduces the recruitment of Chs5p to immobilized Arf1p-GTPγS (). Arf1p-GTP may tether Chs5p to membrane much as it does for coatomer in COPI vesicle formation and adaptor proteins (AP-1 and -2) for clathrin-coated vesicles and as Sar1p-GTP does for Sec23p–Sec24p in COPII coat assembly. Coat recruitment and assembly on artificial membranes is stimulated by acidic phospholipids. Recombinant Chs5p interacts with Chs6p and with a variety of acidic phospholipids ( and ). Recombinant complex expressed in baculovirus-infected Sf-9 cells as combinations of Chs5p, Chs6p, and one or more of the Chs6p paralogues fractionates as an ∼1-MD complex that binds to artificial membranes in an Arf1-GTPγS (or GMP-PNP)–dependent manner (). Optimal interaction occurs on liposomes that are similar in composition to those formulated for the recruitment and assembly of the COPII and I coats (; ). Complexes formed with binary combinations of Chs5p and Chs6p or one other Chs6p paralogue are somewhat less strictly dependent on Arf1p-GTPγS for membrane recruitment. Thus, we suggest that the native mixed complex (i.e., Chs5–Chs6[all]) is the likely form recruited to TGN membranes in vivo. The Chs5–Chs6[all] complex binds to and perturbs the density of liposome membranes (). Thin section microscopy of fixed, membrane-bound complex reveals an electron-dense coat whose morphology is quite distinct from that seen for other coating complexes associated with cargo traffic (). Unlike COPII and I, which assemble onto and vesiculate artificial liposomes, the Chs5–Chs6[all] coat forms a spiky structure but does not pinch the membrane into buds and small vesicles. We propose to call the Chs coat the exomer to reflect its role in the exocytosis of Chs3p and select additional proteins. Although this coat also traffics a subset of other cargo proteins, the subunits are dispensable for normal cell viability, and the and genes are largely restricted to fungi that make chitin. It seems likely that other coats of similarly restricted roles will be uncovered in yeast and in other organisms, but the null phenotype of such coat subunits may not be nearly as dramatic as one normally associates with a general block in secretion. The absence of coated buds and small vesicles in the preparation of exomer-coated liposomes suggests that other factors cooperate with the exomer to convey cargo proteins out of the TGN donor compartment. Chs3p interacts with Chs5p–Chs6p (; ), and it is possible that cargo-coat contact may promote the shape change that accompanies vesicle morphogenesis. Alternatively or in addition, the exomer may engage elements of the cytoskeleton, perhaps actin directly, to draw cargo molecules into tubules much as has been shown for cooperative interaction of clathrin, actin, and the Arp2/3 complex in cell surface invagination and endocytosis in yeast (, ). Indeed, genetic studies link Chs5p and Chs6p to elements of the cytoskeleton (; ). A biochemical reconstitution approach with exomer, liposomes, and cytoskeletal proteins may now be used in an effort to recapitulate the formation of more fully developed TGN to cell surface traffic intermediates. GFP was integrated at the C-terminal codon of the locus by a PCR-based one-step transformation procedure (). Primers for C-terminal GFP integration were 5′-AAGAAGAATAAGAAGAATAAGAAGAAAGGGAAAAAGAAACGGATCCCCGGGTTAATTAA and 3′-ATAAAAAATAGATTATATTTGCTGAGGGATTCTCAGTCGGAATTCGAGCTCGTTTAAAC. A similar approach was applied to generate the Arf1-PA and strains. Primers for C-terminal Arf1 integration were 5′-GGTTTGGAATGGTTAAGTAACAGTTTGAAAAACTAAACTCGGATCCCCGGGTTAATTAAand 3′-CTTTATGTTTCATTTAGTTTATACAAGCGTATTTGATCCGAATTCGAGCTCGTTTAAAC. Deletion primers to generate were 5′-GTCTTCAGTTGATGTACTGTTAACAGTAGGTAAGTTGGACGGATCCCCGGGTTAATTAA and 3′-ATAAAAAATAGATTATATTTGCTGAGGGATTCTCAGTCGGAATTCGAGCTCGTTTAAAC. Yeast strains used in this study include the following: SEY6210 (), CWY512 (), CWY559 (), CWY612 ( ), CWY506 (), CWY624 ( ), and PJ69-4A ( ). All chemical reagents were purchased from Sigma-Aldrich unless specified. Lipids were obtained from Avanti Polar Lipids, Inc., and PIP strips were purchased from Echenlon. Sypro red protein staining dye was purchased from Invitrogen. Complete protease inhibitor cocktail was obtained from Roche Molecular Biochemicals. BL21 (DE3) coexpressing yeast -myristoltransferase and Arf1 (wild type and Q71L) were provided by R. Kahn (Emory University, Atlanta, GA). Anti-Chs5 antiserum was described previously (). Anti-GST antibody was purchased from Santa Cruz Biotechnology, Inc. DH10Bac Competent cells and the Bac-to-Bac baculovirus expression system were purchased from Invitrogen. Glutathione-Sepharose fast flow, pGEX vector, and the Superose 6 gel filtration column were obtained from GE Healthcare. Dynabead M-500 subcellular was purchased from Dynal. GST, GST-Chs5, GST-Chs6, and other GST-Chs5 fragments shown in were constructed in the pGEX vector (GE Healthcare), and proteins were purified from BL21. In brief, 1 L of cell culture was grown to A = 0.5–1.0 followed by 200 μM IPTG induction for 3 h at 22°C. Cells were resuspended in 20 ml PBS buffer, lysed by sonication, and centrifuged at 12,000 rpm for 10 min. The clear cell lysate was incubated with 5 ml glutathione-Sepharose (prewashed by PBS) at 4°C for 3 h. Beads and adsorbed proteins were poured into a column, and 3 × 30 ml PBS aliquots were applied to the column to remove nonspecific material. Bound proteins were eluted with 12 ml PBS + 10 mM of reduced glutathione. For the lipid-protein overlay assay, we adjusted purified proteins to 150 μg/ml and followed the procedures recommended by the manufacturer (Echelon). mArf1p was purified from an BL21 (DE3) strain that coexpressed yeast -myristoyltransferase and either wild-type or dominant-activated (Q71L) Arf1p. The purification procedures have been published previously (), except a Sephacryl S-100 column was used for the gel filtration step. Based on the mobility shift by electrophoresis and in agreement with the literature, we confirmed that >75% of the –purified mArf1p, either wild-type or the dominant-activated mutant, was -myristoylated. The purification resulted in ∼80% pure mArf1p. CWY506 cells were harvested at A = ∼1.0 from 100 ml of culture. Cells were lysed by agitation with glass beads in 2.5 ml of lysis buffer B88, 1× complete protease inhibitor cocktails, and 1 mM PMSF. Total cell lysates were centrifuged at 20,000 for 10 min at 4°C. To the resulting supernatant, we added 0.5 mM of nucleotide where indicated and incubated at 30°C for 10 min. 20 μl IgG-coated Dynabeads were added and incubated at 4°C for 2 h. Procedures for IgG coating on Dynabeads were provided by the manufacturer. Dynabeads were recovered by binding to a magnet, and beads were washed with 1 ml B88 buffer three times. Bound proteins were eluted in 100 μl SDS-PAGE resuspension buffer, and 10 μl was loaded on SDS-PAGE followed by immunoblot analysis. Liposomes were prepared as described previously except 1 mol percentage of Texas red–PE was substituted for NBD phospholipids (; ). In brief, 2 mM of lipid mixtures based on the lipid composition indicated in were prepared, and the organic solvent was evaporated by a rotavapor. HK buffer (20 mM Hepes, pH 6.8, and 160 mM KOAc) was applied to dried lipids, and the suspension was incubated at room temperature overnight followed by 19 passages through a 400-nm Nuclepore polycarbonate membrane. 20 μl of liposomes consisting of ∼1 mM of lipids were incubated for 1 h at 30°C with 1.9 μg of purified mArf1p (final concentration of ∼1 μM in a 80-μl standard reaction) and 0.1 mM nucleotide (GDP, GTP, or GTPγS) in a buffer composed of ∼20 mM Hepes, pH 7.4, 1.2 mM MgCl, 2.5 mM EDTA, 50 mM NaCl, and 15 mM KOAc. Samples were returned to ice, and 2 mM MgCl was added to stabilize the nucleotide-loaded mArf1p. The reaction was then adjusted to 80 μl by the addition of 0.5 μM Chs5–Chs6[all] complex (mol wt of ∼300 kD) or as indicated. In a second stage, complex recruitment was conducted in a buffer composed of ∼38 mM Hepes, pH 7.4, 1 mM EDTA, 1.7 mM MgCl, 18.75 mM NaCl, 35 mM KOAc, and 6.25% glycerol and followed by the binding step at room temperature for 15 min unless otherwise indicated. Samples were then transferred to ice and mixed with 50 μl of 2.5 M sucrose in HKM buffer (20 mM Hepes, pH 6.8, 160 mM KOAc, and 1 mM MgCl). A 110-μl sample was removed to a tube (TLA-100; Beckman Coulter) and layered with 100 μl of 0.75 M sucrose in HKM and 20 μl HKM, sequentially. Samples were centrifuged in a TLA-100 rotor (Beckman Coulter) at 100,000 rpm for 25 min at 24°C. The upper 30-μl fractions were carefully removed after centrifugation, 5 μl of which was used for the determination of lipid recovery based on the fluorescence of Texas red–PE. The remaining floated fraction (20 μl) was resuspend in SDS-PAGE resuspension buffer, and the amount applied to SDS-PAGE was based on lipid recovery. Gels were stained with Sypro red, images were taken using a Typhoon imager (GE Healthcare), and quantification was performed using ImageQuant software (GE Healthcare). The gel filtration experiment was performed using an AKTA FPLC system (GE Healthcare). The mol wt standards for this experiment contained 1 mg/ml blue dextran (2,000 kD), 2.5 mg/ml thyroglobulin (669 kD), 0.15 mg/ml ferritin (440 kD), 1 mg/ml catalase (232 kD), 1 mg/ml aldolase (158 kD), 4 mg/ml BSA (74 kD), 0.15 mg/ml ribonuclease (13.7 kD), and 1 mg/ml cytochrome C (12.4 kD). 200-μl samples (∼0.5–1.0 mg/ml) of a purified complex were injected into the system using a constant flow rate at 0.2 ml/min and collected into fractions of 250 μl. A 10-μl aliquot of each fraction was analyzed by SDS-PAGE followed by Sypro red staining. Quantification was performed using a Typhoon imager and ImageQuant software. To check the integrity of the purified Chs5–Chs6[all] complex, we performed a sucrose velocity gradient analysis. A 120-μl sample containing ∼60 μg of purified complex was loaded on top of a 1.89-ml 10–50% linear sucrose gradient in HKM, and centrifugation was performed for 16 h at 4°C in a rotor (TLS55; Beckman Coulter) at 55,000 rpm. Similar gradient conditions were used to check lipid and protein distribution in the standard incubation, and 160 μl of the reaction mixture was applied instead. A total of 20 fractions (20 × 100 μl) were collected from the top of this gradient, and the sucrose concentration was determined using a refractometer (Fisher Scientific). To check lipid recovery, we removed a 50-μl sample from each fraction into a microtiter plate for quantification of the fluorescence of the Texas red–PE. SDS-PAGE resuspension buffer was added to the remaining fraction, and proteins were analyzed by SDS-PAGE and Sypro red staining. Quantification was performed using a Typhoon imager and ImageQuant software. All strains used for microscopy were grown in synthetic dextrose (SD) medium to mid-log phase. Cells were incubated either at 26°C or shifted to 37°C for 40 min before examination. Microscopy was performed using a fluorescence microscope (Eclipse E600; Nikon). Images were captured by a CCD camera (C4742-95; Hamamatsu) using Image-Pro software (Media Cybernetics). For thin section microscopy, we fixed a standard reaction as described in the Liposome recruitment assay section with 2% glutaldehyde and 1% osmium tetroxide in cacodylate buffer for 1 h on ice. Samples were centrifuged using a TLA100.3 rotor (Beckman Coulter) at 55,000 rpm for 30 min. Membrane pellet fractions were processed for thin section electron microscopy as described previously (). Fig. S1 shows that Chs5p colocalized with Sec7p to the late Golgi compartment. Although the Chs5p-GFP/RFP signal was diffusely distributed at the restrictive temperature in the strain, other Golgi marker signals were focused and more exaggerated, indicating that the Golgi membrane did not disperse in the strain. Online supplemental material is available at .

Human cytomegalovirus (CMV) received its name from its capacity to induce two characteristic and successive cytopathic effects. The early cytopathic effect (ECE) consists of the rounding of infected fibroblasts, whereas the late cytopathic effect (LCE) is characterized by the appearance of granular or dense intracytoplasmic and intranuclear inclusion bodies, as well as by an increased cell volume (; ). The molecular mechanisms accounting for these cytopathic effects are elusive. CMV is an opportunistic pathogen that establishes life-long latent infection without overt clinical disease in immunocompetent individuals, but can cause severe illness in utero, in neonates, and in patients with acquired or iatrogenic immunodeficiency. CMV infection can be associated with colitis (), retinitis (), and encephalitis () accompanied by local cell deaths. CMV encodes two antiapoptotic proteins, especially the viral mitochondria–localized inhibitor of apoptosis (vMIA; pUL37 × 1; ). vMIA protects CMV-infected cells from apoptosis in the late phase of the viral life cycle (), and thus, vMIA-deficient CMV cannot replicate (because it kills the infected cells) unless it infects cells that overexpress Bcl-2–like apoptosis inhibitors such as E1B19K (). Although there is some functional similarity between Bcl-2 and vMIA, which both inhibit apoptosis-associated mitochondrial outer membrane permeabilization (MOMP), there is no obvious sequence similarity between the two proteins (; , ; ; ). Moreover, in contrast to Bcl-2, vMIA induces the fragmentation of the tubular mitochondrial network, reducing its connectivity (; ). The vMIA protein is largely confined to the mitochondrial compartment, and it coimmunoprecipitates with the adenine nucleotide translocase (ANT; ; ), which is the antiporter responsible for the exchange of ADP and ATP at the inner mitochondrial membrane (for review see ). In addition, vMIA has been shown to physically interact with the protein Bax, recruiting it to mitochondria while neutralizing its proapoptotic function (). Because vMIA loses its antiapoptotic action in Bax-deficient cells (), it appears that vMIA exerts its antiapoptotic function solely by neutralizing Bax. Based on these premises, we decided to evaluate the contribution of vMIA to CMV-induced cytopathic effects. We report that vMIA mediates the ECEs and LCEs of CMV infection through a novel effect on mitochondrial bioenergetics that is independent from its antiapoptotic function. Mitochondria of two stable cell lines constitutively expressing vMIA, i.e., a human cervical carcinoma cell line (HeLa) and an immortalized mouse fibroblast cell line (NIH3T3), look rounder and smaller than control mitochondria (). They present a highly disturbed organization, with mitochondrial fragmentation, matrix swelling, and reduction of the number of cristae (). This phenotype is not associated with a reduction of global mitochondrial mass, as determined by staining of the mitochondria with the potential-independent dye MitoTracker green (). Moreover, the abundance of proteins from the respiratory chain was not altered by vMIA (). We observed a reduction in the size of vMIA-expressing cells. This applied to both the cellular volume () and the cytoplasmic membrane surface () as measured by flow cytometry, as well as to the mean contact surface area of adherent cells (). There was no difference in the cell cycle distribution between vector-only and vMIA-transfected HeLa cells (Fig. S1 A, available at ), and the difference in size between vMIA-expressing and control cells was found in both the G1 and the G2/M phases of the cell cycle (Fig. S1 B). vMIA-expressing cells showed an altered actin cytoskeleton with fewer stress fibers and poorly polymerized cortical actin (). vMIA-expressing NIH3T3 cells exhibited a significant delay in their adherence compared with vector-only–transfected control cells (). Moreover, vMIA-expressing NIH3T3 cells were less efficient in “wound healing” in vitro, meaning that they migrated less rapidly into a cell-free area of the culture substratum, which was generated by scratching the monolayer (). Similarly, vMIA-expressing HeLa cells exhibited a reduced motility, as determined by videomicroscopy (unpublished data). The effects of vMIA on cellular morphology appear to be independent from its impact on apoptosis regulation, as indicated by two independent lines of evidence. As compared with Neo cells, HeLa cells expressing vMIA were resistant against a panel of apoptosis inducers, including CD95 ligand, the lysosomal toxin ciprofloxacin (), and anoikis. In contrast, vMIA had no apoptosis-inhibitory effects in NIH3T3 cells (Fig. S2, available at ). Thus, the vMIA effects on NIH3T3 cell morphology must be independent from its apoptosis-regulatory function. Accordingly, the siRNA-mediated knockdown of Bax, which is the target of vMIA-mediated apoptosis modulation (; ), did not reduce the size difference between Neo and vMIA-transfected HeLa cells (). siRNA-mediated knockdown of the Bax homologue Bak, which does not interact with vMIA (), also failed to affect the size of vMIA-expressing cells (). Transient transfection of vMIA imposed a cell-size reduction accompanied by a transition from a filamentous to a punctate mitochondrial morphology in both HCT116 cells, which express Bax, and HCT116 cells, in which was removed (). Altogether, these data provide strong evidence that the morphological effects of vMIA are independent from its apoptosis-inhibitory, Bax-mediated function. Both control and vMIA-expressing HeLa cells were enlarged when plated on either collagen type I– or fibronectin-coated tissue plates, which increase adhesion to the culture substratum, but the difference in size between the cell types persisted (Fig. S3, available at ), pointing to an intrinsic metabolic alteration imposed by vMIA. Actin rearrangements that occur during migration are mainly controlled by the Rho-GTPase family (; ). To check if these pathways were impaired in vMIA cells, we transfected HeLa cells with GFP-coupled wild-type RhoA and Rac-GTPase or with mutant constitutively activated GTPase Rho- and Rac-Q. The distribution of actin was not influenced by Rho-WT or Rac-WT, neither in control nor in vMIA-expressing cells (Fig. S4 A, available at ). This demonstrated that the vMIA-induced defect in actin polymerization (and hence in migration) was not secondary to a defect in Rho-GTPase function. The percentage of cells showing enhanced stress fiber polymerization in response to Rho-Q or increased cortex actin polymerization and cell surface adhesion in response to Rac-Q were similar in control and vMIA-expressing cells (Fig. S4 B). Nonetheless, in qualitative terms, stress fibers were less numerous in vMIA cells than in Neo cells transfected with Rho-Q. Moreover, cortex actin was less polymerized and cell surface adhesion was comparatively reduced in vMIA cells transfected with Rac-Q (Fig. S4 B). Thus, the Rho-GTPase–induced actin polymerization was functional in HeLa vMIA cells, although at a reduced level. No defect in Rho-GTP could be measured in HeLa vMIA cells (Fig. S4 C). Activation of integrins by extracellular matrix components, such as fibronectin or collagen, is a prominent migration inducer upstream of Rho-GTPase proteins (; ). Both fibronectin and collagen type I improved cortex actin polymerization and lamellipod extension in Neo and vMIA HeLa cells; they also enhanced focal adhesions, as assessed by the relocalization and activating phosphorylation (Y397) of FAK (Fig. S4 D). However, we again noticed that lamellipod extension and cortex actin repolymerization induced by collagen type I or fibronectin was less pronounced in vMIA than in Neo HeLa cells. These experiments suggested that signaling pathways controlling actin polymerization and migration were functional in vMIA HeLa cells, although these cells showed a reduced, delayed, and incomplete actin polymerization. Inhibition of actin disassembly with jasplakinolide completely restored stress fiber polymerization and the adherence surface of vMIA-expressing HeLa cells (Fig. S4 E), indicating that vMIA had no direct catabolic effect on actin. Because vMIA-expressing cells exhibited a modified mitochondrial morphology, we tested whether oxidative phosphorylation would be compromised in these cells. No obvious changes in the composition of respiratory chain complexes of HeLa cells could be detected by two-dimensional (native blue vs. SDS-PAGE) electrophoresis (Fig. S5, available at ). Polarographic measurements of mitochondrial oxygen consumption indicated that ADP was unable to stimulate the respiratory activity of mitochondria from vMIA-expressing cells. Subsequent blockade of the FFATP synthase by oligomycin only poorly decreased oxygen uptake in vMIA-expressing cells (respiratory control [RC] = 1.3), as compared with control (RC = 3.2). The decreased phosphorylating ability of the vMIA-expressing cells was further indicated by the similar effect of the uncoupler CCCP, functionally shunting the FFATP synthase on both cell types (RC = 3.3 and 3.4 for vMIA-expressing and control cells, respectively; ). This result suggested that vMIA compromises the function of the ATP synthasome. The ATP synthasome is a multiprotein ensemble composed of complex V (FFATP synthase), ANT, and phosphate inorganic carrier (PiC; ). No difference in the abundance of the α-subunit of the FFATP synthase was found between vMIA-expressing and control cells, as determined by immunoblotting. Moreover, vMIA-expressing cells and control cells possess a similar FFATP synthase activity, as determined by spectrophotometric assays (). vMIA reportedly interacts with mitochondria, particularly with ANT (; ). Hence, we determined the ANT activity on isolated mitochondria in an in vitro assay that does not yield information on the absolute ANT activities, yet allows for the comparison between control and vMIA-expressing cells. vMIA did not compromise the relative ANT activity and rather enhances the uptake of ADP into isolated mitochondria (by a factor of ∼2), which can be explained by an increased abundance of ANT protein (). Inhibition of the adenylate kinase, using P, P-Di (adenosine-5′-) pentaphosphate or that of the F part of the ATP synthase by oligomycin, did not abolish the difference in the ADP/ATP exchange rate between mitochondria from control and vMIA-expressing cells, whereas carboxyatractyloside totally inhibited the ANT activity and the ATP efflux measured (unpublished data). The data obtained on ANT activity of isolated mitochondria () were backed up by experiments performed on intact cells (). Real-time measurements of the intramatrix ATP levels using permeabilized cells transfected with a mitochondrion-targeted luciferase () revealed that the baseline matrix ATP levels were reduced in vMIA-expressing cells, suggesting defective mitochondrial ATP synthesis. The rate of import of external ATP into the mitochondrial matrix was enhanced in vMIA-expressing cells as compared with control cells (). Thus, ANT activity was enhanced by vMIA, much in the way that this has been observed for Bcl-2 overexpression, which also stimulates ANT activity (). Next, we investigated the vMIA effect on the third component of the ATP synthasome, the mitochondrial phosphate carrier. vMIA had no effect on the abundance of the phosphate carrier (). However, vMIA reduced the uptake of [P]phosphate into isolated mitochondria by ∼80%, compared with control mitochondria (), as determined by an in vitro assay. This observation was confirmed in permeabilized cells that were initially kept in a phosphate-depleted medium, in which ATP synthesis (as detected by matrix- targeted luciferase) was stimulated by the addition of increasing amounts of inorganic phosphate (). The phosphate-stimulated ATP generation was reduced in vMIA-expressing cells. Altogether, these data support the notion that vMIA inhibits mitochondrial ATP production, correlating with a decreased activity of the phosphate carrier. Correlating with the defect in oxidative phosphorylation induced by vMIA (), vMIA-expressing cells exhibited a lower steady-state ATP level, which was ∼70% of that in control cells (). The ATP levels in cells transfected with a vMIA mutant lacking the mitochondrion-targeting domain (Δ23-34) () were similar to that of control cells (). Cells expressing vMIA produced more lactate than control cells, indicating that the defect in mitochondrial ATP generation was partially compensated for by an increased anaerobic glycolysis (). The partial decrease in ATP induced by vMIA, however, was not sufficient to cause a manifest activation of the AMP kinase (AMPK)–mammalian target of rapamycin (mTOR)–p70 kinase pathway, as suggested by the absence of an increased phosphorylation of AMPK and p70 kinase in vMIA-expressing cells (). Having established that vMIA-expressing cells had a reduced capacity to generate ATP, we then evaluated the impact of a partial ATP defect on cellular morphology. Exposure of HeLa Neo cells (not depicted) or NIH3T3 control cells to 10 nM oligomycin, which inhibits mitochondrial ATP generation, led to an ∼30% reduction of the intracellular ATP concentration and to a decrease in mean cell size (), accompanied by reduced actin polymerization () and a diminished wound healing capacity () similar to that found in vMIA-expressing cells (). Inhibition of the respiratory chain complexes I and II by rotenone and TTFA, respectively, also reduced ATP levels and cell size (). Thus, expression of vMIA and exposure of cells to inhibitors of ATP production led to similar morphological changes, suggesting that the effects of vMIA on cellular morphology stem from the vMIA-induced changes in mitochondrial energetics, especially from reduced ATP levels. To formally prove that the defect in the mitochondrial carrier induced by vMIA is responsible for the cytopathic effect, we knocked down the phosphate carrier with two different siRNAs (), showing that this manipulation reduced the size of control HeLa cells, yet had no significant effect on the size of HeLa vMIA cells (). These data suggest that the effect of vMIA on cell size is secondary to its inhibitory effect on the phosphate carrier. To examine the effects of vMIA in the context of CMV infection, we infected human fibroblasts, which are a cell type permissive for CMV, with either a CMV virus that encodes functional vMIA or with a vMIA-deficient CMV strain. To prevent apoptosis, we used MRC5 fibroblasts expressing E1B19K, which is a strong cell-death suppressor that has previously been shown to suppress apoptosis induced by vMIA-deficient CMV (). MRC5 cells infected by CMV exhibited a marked rounding-up ∼24 h after infection. This ECE coincided with the onset of strong vMIA expression, as detected by immunofluorescence (). In contrast, vMIA-deficient CMV, while infecting the cell productively (not depicted), failed to induce ECE, indicating that the observed morphological alteration is truly attributable to vMIA expression (). In addition to its effects on the global cellular morphology, the expression of vMIA in cells infected with CMV caused a rapid (10 h after infection) fragmentation of the mitochondrial network, from a filamentous to a punctuate pattern, in accordance with previously published data (). To determine the temporal order between the mitochondrial and the cellular effects of vMIA, we performed confocal microscopy of the vMIA expression in CMV-infected cells, followed by three-dimensional reconstruction of the staining patterns of vMIA and actin. At a first step, vMIA was expressed in mitochondria without any major morphological effect. At a second step, vMIA-decorated mitochondria fragmented, while cells conserved a normal shape and a normal actin cytoskeleton. At a final stage, the fragmented mitochondrial network collapsed around the nucleus when cells rounded up (). This indicates that the mitochondrial change precedes cytoskeleton alterations and rounding of the cells. The LCE of CMV, nuclear and cytoplasmic granulation, and enlargement of infected cells, associated with a complete loss of phalloidin-detectable actin polymers, was observed 5 d after infection (). vMIA was required for the manifestation of LCE because infection with a vMIA-deficient CMV strain failed to induce granulation and actin depolymerization (). Altogether, these data demonstrated a major contribution of vMIA to the pathognomonic cytopathic effects of CMV. vMIA is a CMV-derived protein endowed with the capacity of disabling apoptosis through inhibiting Bax-mediated MOMP. We unraveled a hitherto unexpected effect of vMIA that affects mitochondrial bioenergetics and, hence, multiple cellular functions, including shape, size, adhesion, and mobility. vMIA possesses some functional similarities with Bcl-2 (or Bcl-X), yet exhibits important differences in respect to this family of endogenous apoptosis inhibitors. Both vMIA and Bcl-2 act on mitochondria to inhibit MOMP (; ; ) and diversion of these proteins to other subcellular localization largely reduces their antiapoptotic potential (; ; ). Both vMIA and Bcl-2 reportedly interact with ANT (; ), and both enhance the antiporter activity of ANT (; this study). This correlates with the observation that Bcl-2 and Bcl-X inhibit the pore-forming function of ANT, while enhancing ATP exchange on the membranes of isolated mitochondria (, ; ). Similarly, the antiapoptotic function of vMIA has been correlated with a reduced permeabilization of mitochondrial membranes, through an effect on either ANT (; ) or Bax (; ). In contrast to Bcl-2, vMIA partially reduces cellular ATP levels. Moreover, Bcl-2 has no effects on mitochondrial dynamics and cell size, whereas vMIA causes fragmentation of the mitochondrial network and cell size reduction. Thus, paradoxically, vMIA behaves like Bax or t-Bid, which can fragment mitochondria during apoptosis induction (; ; ; ). Nonetheless, the effect of vMIA on cellular bioenergetics is independent of its interaction with Bax because Bax-negative cells still manifest the mitochondrial fragmentation and bioenergetic effects induced by vMIA. Importantly, vMIA compromises the function of the phosphate carrier, which is the mitochondrial inner membrane protein that exchanges phosphate anions by hydroxyl anions and that oligomerizes with ANT (and the FFATP synthase) within the so-called ATP synthasome. This structure generates ATP within the matrix, while guaranteeing for its export (which is limited by the import of its precursors, i.e., ADP and inorganic phosphate). Thus, a strong inhibition of phosphate carrier activity by vMIA can limit the generation of ATP in the matrix and cause a bioenergetic deficiency in the entire cell. How vMIA (which is located in the mitochondrial outer membrane) affects the function of the phosphate carrier (in the inner mitochondrial membrane) remains a conundrum. However, it has been found that proteins that are located in the inner membrane can interact with outer membrane proteins within contact sites, and allosteric effects across the two membranes have been documented (; ; ). One plausible explanation of the findings obtained in this paper is portrayed in the following scenario. Upon CMV infection, the UL37 exon 1 is transcribed and translated into vMIA protein, which is imported into mitochondria, where it interacts with the ATP synthasome, reducing the enzymatic activity of the phosphate carrier. This ultimately causes a reduction in mitochondrial ATP production and, hence, whole ATP levels. By analogy with other inhibitors of mitochondrial respiration that induce reversible mitochondrial fragmentation without apoptosis induction (; ), vMIA induces mitochondrial fission. On theoretical grounds, the disruption of the mitochondrial network could be caused by either ATP depletion or the effects on mitochondrial proteins. Thus, reduced ATP levels can inhibit mitofusin-1–dependent fusion without affecting Drp1-dependent fission, causing mitochondrial fragmentation as the net result (). Moreover, the antiretroviral drug nelfinavir, which affects ANT structure (), can induce mitochondrial fragmentation without reducing ATP levels (; ). Thus, it is not clear whether the vMIA-induced ATP depletion or the vMIA interaction with the ATP synthasome accounts for the alteration of mitochondrial dynamics. Shortly after mitochondrial fission, cells manifest rounding-up and accumulate a series of alterations that are similarly induced by vMIA expression or ATP depletion, namely a reduction in cell size, adhesion, and mobility. These changes correlate with a depolymerization of stress fibers and a reduced actin cortex, two changes that are not secondary to alterations in Rho or Rac signaling. Although ATP and GTP levels may be expected to correlate (), no detectable defect in Rho-GTP levels was induced by vMIA. Previous studies have unraveled the deleterious effects of ATP depletion on the actin cytoskeleton, resulting in lamellipod retraction, stress fiber collapse, and ventral F-actin aggregation, e.g., in conditions of ischemia (; ; ). Hence, the effects induced by vMIA on the cytoskeleton are likely to be fully explained by ATP depletion. At first glance, it appears counterintuitive that a virus might be “interested” in reducing ATP levels, which might compromise cellular functions and, hence, reduce viral replication. Indeed, herpesviridae other than CMV encode apoptosis-inhibitory proteins with a marked homology to Bcl-2. This applies, for example, to two nononcogenic herpesviruses (Herpesvirus Saimiri and Herpesvirus 68), as well as to two oncogenic viruses, namely, Kaposi's sarcoma associates herpesvirus and Epstein Barr virus (; ; ). Transfection of the two Epstein Barr virus Bcl-2 homologues, BALF or BHRF1, has no measurable effect on mitochondrial or cellular morphology (unpublished data), indicating that they act like endogenous Bcl-2. As a matter of speculation, a partial reduction of ATP levels, by ∼30%, might be “useful” for CMV. As a possibility, actin depolymerization might facilitate the anterograde and retrograde traffic of viral particles and products throughout the CMV life cycle, as has been suggested previously (; ; ; ). Moreover, reduced adhesion or migration of CMV-infected cells might participate in the theratogenic effects of CMV, e.g., on the fetal brain (). Among the β-herpesviridae, CMV is characterized by a long life cycle (4–5 d) and a peculiar cytopathic effect, which is observable on many different CMV-infectable cell types (; ). This characteristic cytopathic effect shaped the name of the virus. The data reported in this study provide a molecular explanation for the cytopathic effect of CMV, which can be attributed to the expression of one single viral protein, vMIA. HeLa and BJAB vMIA-expressing cells were previously described (). NIH3T3 cells were stably transfected with pLncx/vMIAmyc, coding for vMIA-Myc (provided by F. Subra, Ecole Normale Supérieure, Cachan, France). MRC5 human fibroblasts stably expressing E1B19K protein (MRC5) were infected with vMIA-deficient CMV (CMV) or its parental strain, AD169ATCC (CMV WT; ). HCT116 cells were provided by B. Vogelstein (Johns Hopkins University, Baltimore, MD; ). MRC5, HeLa, BJAB, and HCT116 were cultured as previously described (). NIH3T3 cells were cultured in DME supplemented with 10% newborn calf serum. To knock down Bax, Bak, or PiC expression, two siRNAs were used (hBax-384 for Bax1, hBax-208 for Bax2, hBak01-258 for Bak1, [] Qiagen hp_Bak_1_5 for Bak2; and sense strands CUGGCGCACAUCACUAUAU and CCAGGUUAUGCCAACACUU for PiC1 and PiC2, respectively). As a control, siRNA luciferase (Proligo, siRNA luc) was used. 0.5 μl Oligofectamine (Invitrogen) and 30 pmol siRNA were added to a 24-well plate. Plasmid transfections were performed in a 6-well plates using Lipofectamine (Invitrogen) at a ratio of 3 μg Lipofectamine for 2 μg plasmid. NIH3T3 cells were treated with oligomycin, TTFA, and rotenone (Sigma-Aldrich). Cells were incubated for 15 min at 37°C with 150 nM MitoTracker green (Invitrogen). The cell size was quantified as the forward scatter channel (FSC). Plasma membranes were labeled with the PKH2 Green Fluorescent Cell Linker kit (Sigma-Aldrich). Immunofluorescence experiments were performed as previously described (). vMIA Myc-tag, actin filaments, and DNA were visualized using Myc-FITC (Abcam), Alexa Fluor 568 phalloidin (Invitrogen), and Hoechst 33342 (Invitrogen). The primary antibodies used were as follows: vMIA (), immediate early antigen (IEA), early antigen (EA; Argene), and complex III core 2 (Invitrogen). Secondary Alexa Fluor–coupled antibodies were obtained from Invitrogen. In most experiments, a Leica DMIR2 inverted fluorescence microscope (camera DFC300FX, acquisition software Leica FW4000) was used. For adhesion assay, cells were trypsinized and seeded on Nunc culture dishes. Alternatively, cells were grown to confluence and the monolayer was wounded by scratching with a pipette tip. Photographs were taken with visible light at 10×. To calculate the cell surface area, cells cultured on polystyrene dishes were labeled for 40 min with 1 μM CellTracker green (Invitrogen) and washed, and then photographs were taken in 10 areas at 10× (LSM510 confocal microscope; Carl Zeiss MicroImaging, Inc; and ImageJ software; National Institutes of Health). Cells infected with CMV were examined with a LSM510, using LSM Image Browser (Carl Zeiss MicroImaging, Inc.. Transmission electron microscopy was performed as previously described (). ATP was measured with the bioluminescence assay kit HS-II (Roche). Polarographic studies performed on BJAB cells were performed as previously described (). The RC was calculated by dividing oxygen consumption before and after addition of ADP/ATP. The oligomycin-sensitive FFATP synthase activity was measured using the spectrophotometric-coupled assay with pyruvate kinase and lactic dehydrogenase (). To measure ANT activity in vitro, mitochondria were isolated from Neo and vMIA HeLa cells by differential centrifugation in 0.3 M sucrose buffer at 4°C. The ADP/ATP exchange rates of the two types of mitochondria were concomitantly evaluated on the equivalent of 75-μg mitochondrial proteins in a 96-multiwell plate. ATP efflux induced by externally added ADP was monitored by following NADP+ reduction, as previously described (). This assay was validated for microtiter plate format analysis and found to produce results similar to the radiolabeled-ATP–based assay (). For determination of phosphate carrier activity (), mitochondria from HeLa cells were isolated using a mitochondrial isolation kit (Pierce Chemical, Co.). 1 mM [P]phosphate was added to 60 μg mitochondria suspended in 250 mM sucrose, 1 mM EDTA, and 10 mM Tris-HCl, pH 7.4, in a final volume of 110 μl at 25°C. The reaction was terminated after 5 min by addition of 1 mM mersalyl (Sigma-Aldrich). D-Glucose and L-Lactic acid amounts in cell culture media were calculated by the UV method, with enzymatic kits (Boehringer Mannheim/R-Biopharm). To measure the mitochondrial ATP levels in vivo, HeLa cells seeded on glass coverslips were transfected with 3 μg mitochondrially targeted luciferase (). 36 h later, cells were transferred to a custom-built luminometer perfusion chamber (). The plasma membrane was selectively permeabilized with 25 μM digitonin for 1 min, and cells were suspended in an intracellular-like buffer (130 mM KCl, 10 mM NaCl, 1 mM MgSO, 5 mM succinate, 0.5 mM KHPO, 20 mM Hepes, pH 7.0, 1 mM EGTA, and 130 μM CaCl, giving a free [Ca] of 100 nM) at 37°C. Before addition of 20 μM luciferin, ATP, and ADP, cells were treated for 10 min with 5 μM oligomycin. Traces were normalized to the minimal and maximal luminescence levels obtained in the 500 μM ADP/no ATP and no ADP/500 μM ATP solutions, respectively. To measure phosphate dependency, cells were perfused consecutively with 0, 0.1, 0.5, and 1 mM HPO in conditions of maximal activity of the FF ATP synthase (obtained by addition of 1 mM ADP and 100 μM ATP). Cells were lysed in 0.1% SDS-PBS buffer and spun for 5 min at 15,000 . 25 μg total protein were separated by SDS-PAGE. The primary antibodies used were as follows: Myc 9E10 (Santa Cruz Biotechnology, Inc.), Bax N-ter (Millipore), GAPDH, actin (CHEMICON International, Inc.), and PiC (). All antibodies specific for OXPHOS subunits were obtained from Invitrogen. Antibodies for mTOR, AMPK, and p70S6K were purchased from Cell Signaling Technology. Secondary HRP conjugated antibodies were obtained from Southern Biotechnologies. Fig. S1 shows the cell cycle independence of cell size reduction induced by vMIA. Fig. S2 shows cell type–specific apoptosis inhibition by vMIA. Fig. S3 shows the effect of the cell culture substrates on cell size. Fig. S4 shows the effects of Rho and Rac GTPases (provided by C. Lamaze, Curie Institute, Paris, France), cell culture substrates on the actin cytoskeleton, and focal adhesion. Fig. S5 shows two-dimensional gel electrophoresis of respiratory chain complexes from control and vMIA-expressing mitochondria. Online supplemental material is available at .

Phagocytosis is the process by which cells internalize large particles of >0.5 μm in diameter. Most bacterial, fungal, and protozoan pathogens fall within this size range, and phagocytosis by macrophages, dendritic cells, and neutrophils is critical for the clearance of these organisms from infected mammals. Phagocytosis also plays a key role in the removal of dead cells and in the induction of tolerance or initiation of immune responses. Particle uptake is induced by the ligation of specific receptors on the surface of phagocytes, which trigger a membrane remodeling process that is largely dependent on actin polymerization. Depending on the receptors used, distinct signaling pathways are activated, and the engulfment pattern can vary from extensive pseudopod extension to an inward movement of the particle in a sinking fashion (; ). Early studies of phagocytosis in macrophages concluded that the phagosome was formed largely by invaginated plasma membrane (; ). Later studies found a transient increase in surface area during the early stages of particle uptake (; ) or after plating macrophages on immobilized IgG (), suggesting that membrane from an intracellular source was inserted by exocytosis. This intracellular source of membrane was first identified as early endosomes (; ; ), and, more recently, late endosomes were also implicated (). The functional inhibition of tetanus neurotoxin–insensitive vesicle-associated membrane protein (VAMP)/VAMP7, a late endosome/lysosome v-SNARE (, ), inhibited IgG or complement-mediated particle uptake (). Fusion of lysosomes with the plasma membrane was detected during the early stages of particle uptake (), which is consistent with earlier observations of lysosomal enzyme release by macrophages during phagocytosis (; ; ). The identification of synaptotagmin (Syt) VII as a regulator of lysosomal exocytosis (; ; ) prompted us to investigate its possible role in phagocytosis. Syts belong to a large family of membrane proteins characterized by the presence of two Ca-binding C2 domains on their cytoplasmic region (). Syts I and VII, two evolutionally conserved members of the family, are thought to promote membrane fusion by functioning as exocytic Ca sensors of low and high affinity, respectively (). Syt I controls synaptic vesicle exocytosis in neurons (; ), and Syt VII regulates the secretion of lysosomes () and of some nonsynaptic secretory granules of specialized cells (; ; ). In this study, by examining the phagocytic ability of macrophages from Syt VII mice (), we clarify a long-standing controversy about the role of intracellular free Ca ([Ca]) in phagocytosis. Previous studies, which were performed under different conditions, arrived at contradictory conclusions regarding the requirement for [Ca] transients in phagocytosis (; ; ; ; ). Our results indicate that [Ca] is preferentially required for a Syt VII–dependent component of the phagocytic process, which involves the delivery of lysosomal membrane to nascent phagosomes. Imaging analysis revealed that Syt VII–containing domains of lysosomal compartments are rapidly mobilized to phagosomes and to ruffling regions of the plasma membrane, facilitating the uptake of large particle loads. Dominant-negative constructs or RNAi silencing of the late endosomal/lysosomal v-SNARE VAMP7 inhibit particle uptake by RAW267.7 macrophages (), suggesting that lysosomal fusion is required for some forms of phagocytosis. To further understand this process, we investigated the role of Syt VII, the Ca-binding membrane protein that was previously shown to regulate the fusion of lysosomes with the plasma membrane (). Bone marrow macrophages (BMMs) from Syt VII or VII mice were exposed to IgG-opsonized fluorescent zymosan particles at different particle/cell ratios for increasing periods of time. Fluorescence microscopy examination suggested that uptake was reduced in SytVII BMMs, particularly at later time points (). This was confirmed by measuring the fluorescence associated with BMMs in a fluorimeter: as the particle/cell ratio increased, defective particle uptake became evident in Syt VII BMMs (). The phagocytosis defect increased with time, suggesting that Syt VII is required for the continuous uptake of large particle loads. A previous study revealed no difference in the uptake of , , , and by Syt VII or VII BMMs (). This suggested that the defective phagocytosis of zymosan might be related to their larger size (4–5 μm) when compared with bacteria (0.5–2 μm). To directly investigate this possibility, we exposed BMMs to 3- or 6-μm fluorescent latex beads and determined the number of phagocytosed particles after increasing periods of time. The uptake of 3-μm beads was nearly identical for both groups of BMMs except for a small inhibition detected after 1 h in Syt VII BMMs at the highest particle/cell ratio. In contrast, phagocytosis of 6-μm beads was reduced in Syt VII BMMs at all time points at both 10 and 25 particles/cell (). This phenotype was independent of the receptor-mediating particle uptake; impaired phagocytosis in Syt VII BMMs was observed when assays were performed with nonopsonized zymosan (not depicted) or with sheep RBCs opsonized with either IgG or complement (). The phagocytosis defect was not related to a reduced expression of surface receptors because flow cytometry detected similar levels of FcγR or CR3 (Cd11b) on Syt VII or VII BMMs (Fig. S1, available at ). The aforementioned results suggested that Syt VII might facilitate the uptake of large particle loads by mediating the Ca-dependent delivery of intracellular membranes to phagosomes. This hypothesis predicts that phagocytosis in Syt VII BMMs should be less sensitive to [Ca] chelation when compared with wild type. To investigate this possibility, BMMs from Syt VII or VII mice were pretreated or not treated with the membrane-permeant Ca chelator BAPTA-AM and exposed to IgG-opsonized zymosan particles for increasing periods of time. In Syt VII BMMs, BAPTA inhibited particle uptake by 66–84% (). In contrast, [Ca] chelation in Syt VII BMMs did not inhibit uptake at the early time points, and, after longer incubation periods, phagocytosis was only reduced by 30–37% (). Thus, [Ca] appeared to be preferentially required for a Syt VII–dependent component of the particle uptake process. To directly assess the importance of Syt VII Ca-binding activity in phagocytosis, we performed functional reconstitution experiments in Syt VII BMMs. The macrophages were retrovirally transduced with wild-type Syt VII–YFP, YFP alone, or a full-length Syt VII–YFP construct in which the four aspartic acid residues predicted to be involved in Ca binding (; ) were mutated to asparagines (Syt VII (D/N)–YFP). After exposure to zymosan particles for 1 h, Syt VII BMMs were fixed, and the number of internalized particles was quantified microscopically on transduced YFP-expressing cells. In several independent experiments, wild-type Syt VII–YFP markedly enhanced particle uptake when compared with BMMs expressing YFP alone. In contrast, when the putative Ca-binding sites of the C2A and C2B domains were mutated (D225N-D227N and D357N-D359N mutations), the ability of Syt VII to restore normal levels of phagocytosis was abolished (). Syt VII–YFP and Syt VII (D/N)–YFP were expressed at similar levels and were localized on tubulovesicular intracellular compartments of Syt VII BMMs (). However, only Syt VII–YFP was clearly detected on the membrane of most zymosan-containing phagosomes (, top). In contrast, Syt VII (D/N)–YFP did not appear to be efficiently delivered to the few phagosomes present in Syt VII BMMs expressing this construct (, bottom). Mutations in the predicted Ca-binding sites of the major neuronal Syt isoform Syt I (D230N-D232N and D363N-D365N, respectively) abolish its interaction with phosphatidylserine (PS; ) and disrupt Ca-regulated SNARE-mediated membrane fusion in vitro (; ). The effect of equivalent mutations on the Syt VII C2A (D225N-D227N) and C2B domains (D357N-D359N) had not yet been examined, so we performed binding assays using PS/phosphatidylcholine (PC) liposomes and immobilized Syts I and VII C2AB fragments. As expected (), Ca-dependent PS-binding activity was observed for both Syt isoforms (). However, when the putative Ca-binding ligands of C2AB VII were neutralized (D/N mutations), we observed marked Ca-independent binding to PS/PC liposomes. Similar constructs carrying equivalent mutations in the Syt I Ca ligands abolished Ca-triggered binding to PS/PC liposomes, which is consistent with previous experiments (). Thus, Ca ligand mutations in both Syts I and VII abolish Ca-regulated interactions with PS, but, in the case of Syt VII, the PS-binding activity is rendered constitutive. Because Syt VII (D/N) fails to rescue phagocytosis, this Ca-independent PS-binding activity does not appear to influence membrane traffic events leading to phagosome formation. PS is an essential effector of Syts I, IX, and VII during regulated membrane fusion (), and the Ca ligand mutations in Syt VII affect this activity. It is conceivable that the constitutive lipid-binding activity of Syt VII (D/N) causes the binding of the C2 domains to the lysosomal membrane (e.g., the cis membrane to which the protein is anchored via its transmembrane domain) rather than to the target membrane (i.e., interacting with the plasma membrane in response to a rise in intracellular Ca), preventing Syt from engaging the target membrane and, thus, failing to stimulate fusion. The D/N mutations may also impair Ca-dependent Syt VII–t-SNARE interactions, resulting in a failure to activate t-SNAREs for fusion (). In any case, our data demonstrate that the Ca ligand mutations disrupt the ability of Ca to regulate the interaction of Syt VII with an essential effector, PS, resulting in a failure to restore normal phagocytosis in Syt VII BMMs. In several cell types, Syt VII colocalizes with lysosomal markers and regulates lysosome fusion with the plasma membrane (; ; ; ). When expressed in Syt VII–deficient BMMs, Syt VII–YFP restored phagocytosis to wild-type levels and was detected on the membrane of phagosomes (). Collectively, these findings suggested that Syt VII might facilitate the uptake of large particle loads by regulating lysosomal membrane delivery to phagosomes. To test this hypothesis, BMMs from Syt VII or VII mice were exposed to zymosan for 10 min, fixed, and stained with antibodies against the major lysosomal glycoprotein Lamp1. These experiments were performed at a particle/BMM ratio of 10, a condition that allows the formation of a similar number of phagosomes in Syt VII and VII BMMs (). Equivalent numbers of phagosomes (>200) were scored in both groups of BMMs for the presence or absence of defined Lamp1 staining on phagosome membranes (). The results revealed a marked reduction in the number of Lamp1-positive phagosomes in Syt VII BMMs. After 10 min, 32% of recently formed phagosomes in Syt VII BMMs were Lamp1 positive, whereas in Syt VII BMMs, only 18% stained positive (). These results are consistent with a previous study describing the detection of the lysosomal v-SNARE VAMP7 on ∼40% of recently formed phagosomes in RAW264.7 macrophages (). The large majority of the phagosomes within Syt VII or VII BMMs (89.6 and 92.2%, respectively) were still heavily surrounded by polymerized actin after the 10-min incubation period (), indicating that the reduced Lamp1 staining in Syt VII BMMs was not a result of a slower rate of phagosome maturation (; ). Reinforcing this view, specific labeling for Lamp1 was detected by cryoimmuno-EM on the membrane of recently formed phagosomes that were still at the cell periphery, which is in close proximity to ruffling areas of the plasma membrane (Fig. S2, available at ). Retroviral expression of Syt VII–YFP restores phagocytosis in Syt VII BMMs to wild-type levels (), demonstrating that this construct is fully functional. Similarly tagged Syt VII constructs were previously shown to colocalize with endogenous lysosomal markers in the cell lines NRK (), CHO (), and PC12 (; ) as well as in primary keratinocytes (). Consistent with these earlier studies, Syt VII–YFP expressed in BMMs was detected in association with Lamp1-positive late endosomes/lysosomes. Interestingly, confocal z optical sectioning revealed higher concentrations of Syt VII on peripheral regions of Lamp1-positive compartments (). On sequential optical sections, Syt VII–enriched domains were consistently observed in close association with the Lamp1-containing late endosomes/lysosomes in a defined cap pattern oriented toward the plasma membrane (). A similar pattern was observed when the lysosomal compartments of Syt VII–YFP-transduced BMMs were loaded with rhodamine B–dextran by endocytosis followed by a 2-h chase period (). BMMs were also doubly transduced with Syt VII–YFP and CFP-CD63, a tetraspanin used extensively as a marker of late endosomes and lysosomes (; ). Syt VII–YFP was again detected in defined microdomains, which were closely associated with tubulovesicular compartments containing CFP-CD63 (). In contrast, the Syt VII–YFP-enriched membrane domains were not closely associated with transferrin-containing early endosomes (). To further investigate the apparent delivery of Syt VII to phagosomes (), we performed time-lapse confocal microscopy of live BMMs during the particle uptake process. We found that Syt VII–YFP was recruited to sites of phagosome formation during early steps of the process immediately after binding of the particle to the macrophage surface ( and Video 1, available at ). A few seconds after the contact of opsonized zymosan with the membrane of BMMs, the Syt VII–YFP fluorescent signal became very intense at the base of nascent phagosomes. This signal persisted during formation of the phagocytic cup and progressed to completely surround the particles ( and Video 1, frames from 0 min 18 s to 0 min 54 s). Interestingly, Syt VII delivery to the plasma membrane was also observed at the sites of intense membrane ruffling that follow phagocytic cup closure, leading to particle engulfment. Syt VII–YFP was also detected at sites of plasma membrane ruffling that were not associated with phagocytosis ( and Video 1, frames from 1 min 12 s to 10 min 43 s). These observations suggested that Syt VII was recruited to nascent phagocytic cups from the peripheral domains of lysosomal compartments, where the protein is localized in resting BMMs. Such a scenario would predict that additional lysosomal resident proteins such as Lamp1 should also be detected on nascent or recently formed phagosomes. To investigate this issue, wide-field time-lapse fluorescence microscopy was performed in live BMMs transduced with Syt VII–YFP and Lamp1-RFP. Lamp1-RFP was detected throughout the dense perinuclear portions of the lysosomal compartment and also on clearly defined tubules extending toward the cell periphery. These dynamic peripheral tubules also contained Syt VII–YFP (, insets of frame at 3 min 12 s). Shortly after Syt VII was visualized colocalizing with Lamp1 on the peripheral tubules, it appeared on the membrane of nascent phagocytic cups (, frames from 3 min 12 s to 10 min 24 s; and Videos 2–4, available at ). Lamp1-containing tubules were initially visualized as a scaffold at the site where Syt VII accumulated, before formation of the phagocytic cup. Immediately after this initial step, Lamp1-positive tubules were observed wrapping themselves around Syt VII–positive nascent phagocytic cups ( and Videos 2–4). A very similar process was observed during the formation of macropinosomes (Videos 2 and 3). Confocal microscopy confirmed the close association of Lamp1 with Syt VII on nascent phagosomes (, boxed area 1) and their full colocalization on recently formed phagosomes (, boxed area 2). The Lamp1 patches detected in association with Syt VII–positive peripheral phagosomes (, boxed area 1) are likely to correspond to the highly dynamic Lamp1 tubules observed by live microscopy (, insets in frame at 3 min 12 s; and initial frames of Videos 2 and 3). Immunogold EM also detected endogenous () or YFP-tagged () Syt VII on tubular structures closely associated with Lamp1-containing compartments and endogenous Syt VII and Lamp1 on the membrane surrounding recently ingested zymosan particles (). Although Syt VII and Lamp1were apparently present on contiguous membrane segments, Syt VII was predominantly detected in clusters that appeared to exclude Lamp1 (; arrows). Together with the rapid recruitment of Syt VII to sites of phagosome and macropinosome formation, these observations suggest that Syt VII–containing microdomains on late endosomal/lysosomal compartments of BMMs may have unique properties that favor their rapid mobilization to the cell surface. Quantitative image analysis reinforced the conclusion that Syt VII is mobilized to phagosomes from intracellular lysosomal compartments and not from the plasma membrane. The density of Syt VII–YFP on the phagosomal membrane relative to its density at the plasma membrane was measured on confocal images of BMMs fixed after 10 min of exposure to IgG-opsonized zymosan particles (). The density of CD11b, which was determined after staining with specific antibodies, was similar on the plasma membrane and on phagosomes. In 206 phagosomes examined, the mean CD11b phagosome/plasma membrane ratio was 1.19, confirming that phagosome maturation during this period was negligible. In contrast, Syt VII–YFP was significantly enriched in relation to the plasma membrane on 81% of the phagosomes (). Although the transient nature of plasma membrane ruffles did not allow a more extensive quantitative analysis, Syt VII–YFP also appeared to be enriched on areas of active plasma membrane ruffling (). In this study, we showed that primary macrophages lacking the lysosomal Ca sensor Syt VII are defective in the phagocytosis of large particle loads. This defect is independent of the receptor used for phagocytosis and can be rescued by the expression of wild-type Syt VII. Imaging of particle uptake revealed that Syt VII is localized on discrete, dynamic domains of the highly tubular lysosomal compartment of macrophages (). Syt VII is rapidly mobilized from these domains to nascent phagosomes, which shortly thereafter also acquire the abundant lysosomal glycoprotein Lamp1. Indicating a direct role for Syt VII in regulating lysosomal membrane delivery to phagocytic cups, Lamp1 detection on recently formed phagosomes was markedly reduced in Syt VII BMMs. Several lines of evidence indicate that the early lysosomal membrane delivery observed in our studies is not a consequence of phagosome maturation (; ). First, during the short particle uptake period used in these assays (10 min), the density of the plasma membrane protein CD11b remained constant on phagosomes and the plasma membrane (mean ratio of 1.19). Second, the large majority of phagosomes formed during this period were still surrounded by polymerized actin and were frequently observed close to the cell surface. Third, a detailed study of phagosome maturation was previously performed with cells from Syt VII mice and revealed no defects in the uptake, transport, and lysosomal degradation of EGF and in the uptake and intralysosomal killing of (). Mutations abolishing the Ca-dependent membrane association activity of Syt VII also abolished its ability to restore phagocytosis in Syt VII–null macrophages. The fact that Syt VII is a Ca-binding protein that regulates the fusion of lysosomes with the plasma membrane (; ; ) raised the possibility that the delivery of lysosomal membrane might correspond to a Ca-dependent supplemental source of membrane during phagocytosis. Consistent with this view, we found that the chelation of [Ca] inhibited phagocytosis in wild-type BMMs to a significantly larger extent when compared with the inhibition observed under the same conditions in Syt VII BMMs. It has long been known that the ligation of phagocytic receptors, particularly the macrophage FcγR, triggers transient increases in [Ca] (). In neutrophils, intracellular Ca stores seem to redistribute around sites of phagocytosis (), emphasizing the apparently important role played by [Ca] in particle uptake. However, the requirement of Ca signaling for phagocytosis in macrophages has remained controversial. Some studies concluded that the chelation of intracellular Ca had no effect on the uptake of IgG-opsonized particles (; Greenberg et al., 1991), whereas others reported a marked inhibition in phagocytosis (; ). Our present study clarifies this issue by demonstrating that [Ca] transients and Syt VII–mediated lysosome recruitment are required for phagosome formation, but only under conditions of high membrane demand. The experimental conditions of studies in which [Ca] transients were found not to play a role may have favored uptake pathways that were not dependent on Syt VII–mediated mobilization of lysosomal membrane (; Greenberg et al., 1991). Although nascent phagosomes are formed in large part by invagination of the plasma membrane (), it became evident in recent years that intracellular compartments also contribute to the process. Intracellular membrane is delivered to the cell surface during phagocytosis (; ), and functional inhibition of the exocytosis of VAMP3-containing early endosomes (; ; ) or VAMP7-containing late endosomes/lysosomes () inhibits particle uptake. The endoplasmic reticulum was also proposed as a source of phagosomal membrane (), but more recent studies failed to confirm that this is a widespread process, particularly during the uptake of IgG-opsonized particles (; ). Our present studies are fully consistent with recent evidence implicating lysosomes as an important source of membrane during phagocytosis (). Our results expand those findings by showing that the delivery of lysosomal membrane is dependent on the Ca sensor Syt VII and, thus, is likely to correspond to a Ca-sensitive component of the phagocytic process. Our findings are also consistent with additional observations from other laboratories, which have reported an early fusion of lysosomes with phagosomes containing () and the rapid wrapping of lysosomal tubules around nascent phagosomes in mouse BMMs (). These results suggest that the remarkable ability of macrophages to engulf daunting particle loads may be largely attributed to the Syt VII/Ca-dependent use of lysosomes as a supplemental source of membrane. Early endosome and lysosome membrane delivery appear to be tightly coordinated during phagosome formation, with the recruitment of VAMP7 immediately following the accumulation of the early endosome SNARE VAMP3 (). In our study, Syt VII was detected on late endosomal/lysosomal membrane domains proximal to the plasma membrane but was not associated with transferrin-containing early endosomes. Interestingly, in RAW264.7 macrophages, VAMP7 was also detected at more peripheral domains of Lamp1-positive lysosomes (). These observations suggest that the lysosomal compartments of macrophages may contain specialized microdomains containing both a Ca sensor (Syt VII) and a fusogenic SNARE (VAMP7) at a peripheral location that facilitates rapid fusion with the plasma membrane. Syt VII was also mobilized to sites of plasma membrane ruffling and to macropinosomes, which is consistent with the fact that the phagocytosis defect of Syt VII BMMs is receptor independent. This finding suggests an intriguing coupling between signaling events leading to rearrangements of the cortical cytoskeleton and the rapid fusion of Syt VII–containing lysosomal microdomains with the plasma membrane. PI 3-kinase–generated phosphoinositol phosphates have been extensively linked to membrane traffic events and to actin remodeling (). Thus, it is noteworthy that treatment of macrophages with the PI 3-kinase inhibitor wortmannin generates a phenotype that is very similar to the genetic ablation of Syt VII: phagocytosis is inhibited proportionally to an increase in the size of the particles (). Further analysis of the highly dynamic Syt VII–containing microdomains of macrophage lysosomes should provide useful insights into this process. Syt VII was previously shown to be involved in plasma membrane repair (; ) and in neurite extension by primary sympathetic neurons (), suggesting that intracellular signaling domains containing this Ca sensor may control several plasma membrane–remodeling events in addition to phagocytosis. Mouse BMMs were prepared from C57BL/6 wild-type (Syt VII) and Syt VII mice (backcrossed into the C57BL/6 background; ) as previously described (), seeded onto 96- (100 μl of a 5 × 10 BMM/ml suspension) or 24-well (0.5 ml of a 1.5 × 10 BMM/ml suspension) plates 24 h before experiments, and incubated in BMM media (RPMI, 10% FBS, 20% L cell–conditioned supernatant, and 1% penicillin/streptavidin) at 37°C and 5% CO. m o s a n r e d b i o p a r t i c l e s ( I n v i t r o g e n ) w e r e o p s o n i z e d b y i n c u b a t i o n w i t h p u r i f i e d r a b b i t p o l y c l o n a l I g G - o p s o n i z i n g r e a g e n t ( I n v i t r o g e n ) f o r 1 h a t 3 7 ° C a s r e c o m m e n d e d b y t h e m a n u f a c t u r e r . B M M s w e r e i n c u b a t e d w i t h o p s o n i z e d z y m o s a n r e d a t d i f f e r e n t p a r t i c l e / c e l l r a t i o s f o r d i f f e r e n t p e r i o d s o f t i m e , w a s h e d f o u r t i m e s w i t h P B S , a n d t h e f l u o r e s c e n c e w a s r e a d i n a f l u o r i m e t e r ( S p e c t r a M A X G E M I N I ; I n v i t r o g e n ) a t 5 9 5 / 6 2 0 n m . B a c k g r o u n d a b s o r b a n c e l e v e l s c o r r e s p o n d i n g t o w e l l s c o n t a i n i n g B M M s a l o n e w e r e s u b t r a c t e d f r o m a l l v a l u e s . T h e c e l l n u m b e r o n e a c h w e l l w a s s u b s e q u e n t l y d e t e r m i n e d b y l y s i n g t h e c e l l s i n 1 % T r i t o n X - 1 0 0 a n d a s s a y i n g f o r l a c t a t e d e h y d r o g e n a s e a c t i v i t y ( C y t o T o x 9 6 N o n - r a d i o a c t i v e C y t o t o x i c i t y A s s a y ; P r o m e g a ) . T h e s e a s s a y s i n d i c a t e d t h a t n o c e l l l o s s o c c u r r e d d u r i n g t h e e x p e r i m e n t s . F o r c a l c i u m c h e l a t i o n a s s a y s , B M M s w e r e i n c u b a t e d i n H B S S w i t h o u t c a l c i u m a n d m a g n e s i u m c h l o r i d e c o n t a i n i n g 1 0 0 μ M B A P T A - A M ( I n v i t r o g e n ) a n d 5 m M E G T A f o r 3 0 m i n a t r o o m t e m p e r a t u r e b e f o r e i n c u b a t i o n w i t h o p s o n i z e d z y m o s a n . #text 10 sheep RBCs (MP Biomedicals) were washed twice with PBS and opsonized for 30 min at room temperature with 15 μg/ml of rabbit anti–sheep RBC IgG (MP Biomedicals) or rabbit anti–sheep RBC IgM (Accurate) antibodies. IgM-opsonized sheep RBCs were further incubated with 10% C5-deficient human complement (Sigma-Aldrich) for 20 min at 37°C. After opsonization, sheep RBCs were washed twice with PBS and resuspended in prewarmed BMM media. Before incubation with complement-opsonized sheep RBCs, BMMs were preincubated with 150 ng/ml PMA (Sigma-Aldrich) for 15 min at 37°C. BMMs were then incubated for various periods of time with opsonized sheep RBCs at an RBC/macrophage ratio of 10:1. After incubation, cells were washed three times with PBS and fixed in 2% PFA overnight at 4°C. For microscopy-based assays, macrophages incubated with zymosan red were fixed with 2% PFA and stained with a 1:250 dilution of goat anti–rabbit IgG AlexaFluor488 (Invitrogen) without permeabilization to stain extracellular particles. For the detection of extracellular polystyrene particles, rabbit anti–goat IgG AlexaFluor546 (Invitrogen) was used at a 1:200 dilution. For the detection of extracellular IgG-coated sheep RBCs, goat anti–rabbit IgG AlexaFluor488 was used at a dilution of 1:250. To detect noninternalized C3bi-opsonized sheep RBCs, mouse anti–rabbit IgM-FITC (Sigma-Aldrich) was used at a dilution of 1:100. Approximately 300–350 BMMs were analyzed for each condition. Flow cytometry was performed in a flow cytometer (FacsCalibur; Becton Dickinson) to quantify the amount of surface-exposed FcγR or CR3 (CD11b) receptors in Syt VII and VII BMMs. 10 BMMs were kept in suspension on ice for 20 min in PBS and 2% BSA followed by staining with 5 μg/ml FITC-conjugated rat anti–mouse CD16/CD32 (Fcg III/II receptor) mAbs (BD Biosciences) or 5 μg/ml rat anti–mouse CD11b (integrin a chain Mac1) antibodies (BD Biosciences), secondary goat anti–rat AlexaFluor488-conjugated antibodies (Invitrogen) for 20 min on ice, washes, resuspension in 1 ml PBS, and FACS analysis. Images were acquired using a microscope (Axiovert; Carl Zeiss MicroImaging, Inc.) through a 100× objective using a cooled CCD camera (Orca II; Hamamatsu) controlled by MetaMorph software (Universal Imaging Corp.) or a laser scanning confocal microscope (z-stack images were acquired with optical sections of 0.5–0.8 μm at 1-μm intervals; LSM 510; Carl Zeiss MicroImaging, Inc.). For quantitative image analysis, images were imported into ImageJ (National Institutes of Health [NIH]; ), and the mean fluorescence per pixel was measured using the measurement tool. For the quantitation of lysosome fusion with phagosomes containing zymosan, for each condition, >200 phagosomes were counted in a blinded fashion by two independent investigators. For quantitation of phalloidin staining, 350 phagosomes were counted for each condition (Syt VII and VII BMMs). Bone marrow–derived progenitor cells were transduced using retroviral vectors encoding Lamp1-RFP (pLZRS-Lamp1-RFP), Syt VII–YFP (pLZRS-SytVII-YFP), or CFP-CD63 (pLZRS-CFP-CD63) with virus generation as previously described (). pLZRS-Lamp1-RFP was prepared by swapping the coding region GFP with RFP using BamHI–NotI sites in the retroviral construct pLZRS-EGFP-NL-lgp120 (a gift from J. Lippincott-Schwartz, NIH, Bethesda, MD; ). Syt VII–YFP and CFP-CD63 were constructed by insertion of the Syt VII or CD63 coding regions from Syt VII–GFP () or GFP-CD63 () into modified pLZRS-CFP or -YFP retroviral constructs using XhoI and HindIII sites. For retroviral transduction, 4 × 10 cells were incubated with viral supernatants 1 d after isolation at 4°C for 1–2 h before plating on 9-cm nontissue culture-treated dishes (Falcon). To improve transduction efficiency, additional aliquots of viral supernatant were added to the differentiating BMM cultures on days 2 and 3 after isolation. BMMs transduced or untransduced with Syt VII–YFP were incubated with 50 IgG-opsonized zymosan particles per cell for 10 min at 37°C, washed three times with PBS, and fixed with 4% PFA in 0.25 M Hepes, pH 7.4, for 1 h at room temperature followed by an overnight fixation in 8% PFA in the same buffer at 4°C. After washes in PBS, cells were scraped, embedded in 10% gelatin, infiltrated overnight in 2.3 M sucrose, and frozen in liquid nitrogen. Ultrathin frozen sections were double labeled with anti-GFP, anti–Syt VII (), and anti-Lamp1 antibodies using previously described methods (). Sections were examined in an electron microscope (Tecnai 12 Biotwin; FEI), and images were captured digitally using a CCD camera (Morada; Soft Imaging Systems). D225,227,357,359N (designated as D/N) mutations that neutralize the putative Ca-binding ligands in each C2 domain of Syt VII were introduced into the Syt VII–YFP plasmid using the QuikChange Site-Directed Mutagenesis Kit (Stratagene). After retroviral transduction with wild-type, mutated Syt VII–YFP, or vector alone, BMMs were incubated with 25 zymosan red particles/cell at 37°C for 1 h, and 400 BMMs were analyzed for each condition. For Syt–PS-binding assays, cDNA encoding the cytoplasmic domain of Syt I (C2AB I, residues 96–421) and Syt VII (C2AB VII, residues 134–403) were subcloned into pGEX-2T vector (GE Healthcare). D/N corresponds to mutations that neutralize the putative Ca-binding ligands in each C2 domain (D230,232,363,365N for C2AB I and D225,227, 357,359N for C2AB VII). Proteins were expressed as GST fusion proteins and purified using glutathione–Sepharose beads (GE Healthcare) as described previously (; ). Synthetic 1,2-dioleoyl--glycero-3-phospho--serine (PS), 1,2-dioleoyl--glycero-3-phosphocholine (PC), and -(lissamine rhodamine B sulfonyl)-1, 2-dipalmitoyl--glycero-3-phosphoethanolamine were purchased from Avanti Polar Lipids, Inc. Large (∼100 nm) unilamellar liposomes were prepared by extrusion as described previously (). Liposomes were composed of 1% -(lissamine rhodamine B sulfonyl)-1,2-dipalmitoyl--glycero-3-phosphoethanolamine, 25% PS, and 74% PC. Syt–PS interactions were measured by using rhodamine-labeled liposome pull-down assays as described previously (). Immobilized proteins and liposomes were coincubated in binding buffer plus 2 mM EGTA or 0.2 mM Ca in Micro Bio-Spin chromatography columns (Bio-Rad Laboratories) for 15 min at room temperature. Samples were then rapidly washed three times with their respective binding buffers. Bound liposomes were eluted with Hepes buffer containing 1% Triton X-100. The extent of binding was determined by measuring the fluorescence intensity of rhodamine in the eluate. Fig. S1 shows that Syt VII mouse BMMs do not have a reduced expression of surface receptors for IgG or complement. Fig. S2 shows immunogold EM localization of Lamp1 on a recently formed phagosome containing zymosan. The videos show that Syt VII–YFP expressed in a wild-type macrophage is delivered to nascent phagosomes during the uptake of zymosan particles. The macrophage shown in Videos 2–4 was also transduced with Lamp1-RFP. This marker is visualized on tubular lysosomes extending toward the site of zymosan uptake and surrounding recently formed phagosomes. Online supplemental material is available at .

Mammalian cells have evolved multiple mechanisms for internalizing fluids, surface-bound ligands, and plasma membrane components. Although some are well-characterized mechanisms such as endocytosis in clathrin-coated pits, many others including phagocytosis by nonprofessional phagocytes remain poorly defined. One way to generate new insight into cellular uptake processes is to make use of intracellular microorganisms that exploit endocytic mechanisms to infect mammalian cells. For example, receptor-mediated endocytosis is commonly used by intracellular pathogens including bacteria and viruses for infecting cells (; ; ; ; ). Apart from the convenient transport in vesicles, the acidic environment of endocytic vesicles is used by many viruses to efficiently penetrate into the host cytosol (). In the case of herpes simplex virus type-1 (HSV-1), endocytosis plays a dominant role in infection of many cell types (; ). This process, however, appears to be unique because it is likely not mediated by formation of clathrin-coated pits or caveolae and it may not always be pH dependent (; ; ). The other mode of HSV-1 entry is virion fusion at the plasma membrane, which is pH independent () and is facilitated by at least five viral glycoproteins: gB, gC, gD, gH, and gL (). Entry via this mode is initiated by interaction of viral gC and/or gB with heparan sulfate (HS) proteoglycan (). This interaction is followed by binding of viral gD to one of its receptors (). The gD receptors include HVEM, a member of the tumor necrosis factor receptor family; nectin-1 (CD111), a member of the Ig superfamily; and 3--sulfated heparan sulfate or 3-S HS (, ; ; ). The interaction between gD and its receptor mobilizes, in an unknown fashion, participation from gB, gH, and gL to trigger the process of membrane fusion (). Noticeably, the gD receptors are also essential for entry via endocytosis (; ). HSV causes conditions ranging from blisters on mucosal surfaces to deadly brain infections in immunocompromised individuals (). It also causes ocular diseases affecting not only the outer epithelial layer of the cornea but also the stroma and the tissues neighboring the cornea such as the trabecular meshwork () in the eye. Infection of the stroma or herpetic stromal keratitis (HSK) is rare, but is a significant cause of infectious blindness in the developed countries. HSK lasts longer and is more difficult to treat than epithelial keratitis (; ). Development of novel strategies to efficiently treat HSK is hampered by the relatively poor understanding of the infection of the corneal stroma. To better elucidate the infection of the stroma and to generate specific information on the endocytic mode of entry, the present study compared HSV-1 entry into primary human corneal fibroblasts (CF) cultured from excised tissue of the stroma with that of nectin-1 expressing Chinese hamster ovary (nectin-1-CHO) cells. Both cell types exhibited a pH-dependent mode of viral entry that mimics and retains many features of phagocytosis. Essentially, HSV-1 virions associate with cellular protrusions followed by an actin network–dependent uptake (internalization of the virions into the host cytosol) involving RhoA activation. This is a novel uptake mode that has implications for entry of virions into other professional and nonprofessional phagocytic cells. Monolayer cultures of CF and nectin-1-CHO or HVEM-expressing CHO cells (HVEM-CHO) were infected with a recombinant β-galactosidase (β-gal)–expressing reporter virus, HSV-1(KOS)gL86, abbreviated here as KOS-gL86 and interchangeably used in this manuscript for HSV-1. β-gal activity indicates that virus has entered the cell, released its genome into the nucleus, and activated the constitutive promoter for the enzyme production and activity (). Although no β-gal activity (, top) was observed in naturally resistant wild-type CHO-K1 cells (CHO-WT), dosages of input virus sufficient to infect 100% of nectin-1-CHO cells (, middle) also caused β-gal expression in 100% of the CF cells (, bottom). The viral dose-response curve for CF, performed in the linear range of 0–100 PFU/cell, was similar to that of nectin-1-CHO or HVEM-CHO cells and naturally susceptible HeLa cells (). CHO-WT cells transfected with an empty vector pcDNA3 were used as a negative control (). As a measure of productive infection, HSV-1 replication was verified by real-time PCR using HSV-1 gD probes and also by syncytial plaque formation. The viral DNA replications were indicated by threshold cycle (C) values: the lower the value, the higher the viral replication. In CF, the highest replication was seen at 36 h, which was also verified by PCR (). The HSV-1 virions were also able to form plaques and the numbers of plaques formed (unpublished data) were very similar for CF and nectin-1-CHO cells (Fig. S1, A–C, available at ). Ultrastructural analyses of HSV-1 entry into CF and nectin-1- CHO cells using EM suggested a role for plasma membrane protrusions and vesicles in entry. Transmission EM (TEM) micrographs demonstrated enveloped virions adjacent to and/or surrounded by protrusions of the plasma membrane in both nectin-1-CHO cells (, arrows) and CF (). The internalized virions were enveloped and located in uncoated and relatively large vesicles () not in clathrin-coated pits as seen in typical endocytosis. In contrast, HSV-1–infected primary cells cultured from excised human trabecular meshwork tissues (TM cells) had relatively few protrusions () and the virions were mainly attached to relatively smooth cell surfaces (, arrows). Intracellular vesicles containing virions were virtually absent in TM cells. The lower magnification views for all three cell types are shown in Fig. S2 (available at ). Scanning EM (SEM) results correlated TEM observations that many HSV-1 virions (, arrows) were attached to the protrusions seen on nectin-1-CHO cells (arrowheads). The protrusions were ∼5–30 μm long and rarely branched, mainly seen within 30 min of exposure to virus. Immunogold EM of HSV-1 envelope glycoproteins gD or gH–gL further confirmed that enveloped HSV-1 virions associated with the protrusions on nectin-1-CHO cells and localized in the intracellular vesicles (unpublished data). No coated pits or caveolae (“smooth coated” or “non-coated” omega- or flask-shaped invaginations) were observed. To completely rule out endocytosis of HSV-1 in clathrin-coated vesicles, dominant-negative EGFR pathway substrate clone 15 (Eps15) mutants were used (). Eps15, a constituent of plasma membrane clathrin-coated pits, is ubiquitously and constitutively associated with AP-2 (, ). AP-2, clathrin, and dynamin are three major coat proteins. The AP-2 complex plays a principal role in both the organization and function of plasma membrane coated pits; it drives clathrin assembly onto the plasma membrane and interacts with tyrosine based signals of membrane receptors. Clathrin provides an organizing framework to the pit. Dynamin is a GTPase that assembles into rings at the neck of invaginated coated pits and its GTPase activity is necessary for the scission of the vesicle from the plasma membrane (; ; ). Dominant-negative mutants of Eps15 inhibit clathrin-dependent endocytosis (; ). The mutants used were: DIIIΔ2 (control), a C-terminal domain construct of Eps15 lacking all the AP-2–binding sites that does not interfere with endocytosis; DIII, the C-terminal domain construct of Eps15 containing all the AP-2 binding sites (; ), which blocks endocytosis (), and EΔ95/295 or EH29, the construct lacking Eps15 homology (EH) domains at the N-terminal that are required for clathrin-coated pit targeting of Eps15 and thus an inhibitor of clathrin-coated pit assembly (). These controls and mutants, when transfected into nectin-1-CHO cells, all failed to block virus entry (). The transfection efficiency estimated by GFP expression () was at an appropriate level of 60%. In parallel experiments, the dominant-negative mutants DIII and EΔ95/295 expectedly blocked human herpes virus-8 (HHV-8) entry of 293T cells () as HHV-8 is known to use endocytosis via clathrin-coated pits for entry (). Absence of any clear evidence supporting endocytosis in coated pits and the development of protrusions in response to HSV-1 entry raised the possibility that virus entered into nectin-1-CHO and CF via a phagocytosis-like uptake. To test this possibility, we used an assay that allows the phagocytosis process to be observed and quantitated in cells. Essentially, the internalization of fluorescently labeled bacteria such as fluorescein-labeled K-12 ( bioparticles) was monitored. After quenching of the extracelluar fluorescence by trypan blue, the assay measures intracellular fluorescence emitted by the engulfed particles (; ). The phagocytosis assays () were performed using nectin-1-CHO, HVEM-CHO, and a CHO mutant (CHO-745 or pgsA-745) transfected with nectin-1 or HVEM. CHO-745 cells lack glycosaminoglycans including HS (). Because HS is required for HSV-1 attachment to cells (), the purpose of this experiment was to determine if the induction of phagocytosis was dependent on virus attachment to cells. As predicted, exposure of HSV-1 to nectin-1-CHO and HVEM-CHO elicited several hundred-fold induction of phagocytosis activity compared with the unexposed cells (). A similar but somewhat muted response was observed with CHO-745 cells transfected with the same receptors, which showed only about a 10-fold increase in the phagocytosis activity (). Thus, it appears that HS-mediated attachment of virions (normally mediated by HSPG) to cells is needed for maximal triggering of phagocytosis. Partial, HS-independent attachment of virions to cells does occur in the presence of HSV-1 gD receptors; therefore, complete loss of phagocytosis was not expected (). To gain further insight into the mechanism, we focused on the ubiquitously expressed dynamin-2 isoform of dynamin, which, as indicated earlier, is crucial for regulation of the actin dynamics at the plasma membrane and for detachment of intracellular vesicles including phagosomes from the plasma membrane (; ; ). We focused on dynamin also because it is considered to be essential for phagocytosis but not for macropinocytosis (), and could potentially be used to differentiate the two processes. Although nectin-1-CHO cells transfected with WT-dyn2 or its dominant-negative mutant, K44A-dyn2, exhibited significant enhancement of phagocytosis activity in the presence of HSV-1 virions (), there was a moderate yet significant (∼25%) decrease in the phagocytosis activity by K44A expression. Given a 50–60% transfection efficiency, the net decrease in the phagocytosis activity could be close to 50% if all the cells were expressing the mutant. Thus, dyn2-mediated membrane pinching might play a role in HSV-1 uptake via a phagocytosis-like mechanism. Next, to evaluate how this drop in phagocytosis could be translated into viral entry, we performed an entry assay. Unlike WT-dyn2, which showed the same effect on entry as transfection with an empty plasmid, the expression of K44A-dyn2 resulted in a significant (60% or more) decrease in HSV-1 entry (). Again, when extrapolating to 100% transfection efficiency, the data would indicate significant blocking of entry in the presence of the mutant and may implicate dynamin-mediated processes in HSV-1 internalization. HSV-1 induction of bioparticle internalization via phagocytosis does not necessarily imply that the virions themselves get internalized using the same route unless a colocalization is demonstrated. To determine whether the virus cointernalized with phagocytic tracers such as red fluorescent latex beads, a series of colocalization experiments was performed. CF or nectin-1-CHO cells were either treated with cytochalasin D (Cyto D), an inhibitor of actin polymerization and phagocytosis (; ; ), or mock-treated before coincubation with red fluorescent beads and a recombinant HSV-1 (K26GFP). The recombinant virus contains nucleocapsid tagged with GFP () that allows monitoring of the virus internalization into host cytosol (virus entry) using fluorescent confocal microscopy (CM). Non-internalized virions were removed by low pH treatment (citrate buffer pH 3) (Fig. S3, available at ). Using CM, z-stacks were generated and analyzed for internalization of both the virions (green) and the tracer (red). Indeed, in mock-treated cells, virions cointernalized with beads (, yellow), although a few clusters of virions were also noted to have entered into cells without the beads. It was not totally unexpected because the virus, in its capacity as an inducer, could potentially invoke its own internalization via phagocytosis-like uptake without the tracer. With Cyto D treatment, internalization of both virions and the tracer was significantly restricted, as very few virions and/or beads were seen (). Overall, the finding that the virions and the phagocytic tracers were cointernalized suggests that phagosomes may be the vesicles that transport HSV-1 virions during the entry. Given that Cyto D blocked cointernalization of virions and the beads, we next determined whether the actin inhibitors would also block entry of HSV-1 independent of the beads. Cyto D and another actin inhibitor, latrunculin B (Lat B) () were used. Although without showing any detectable effect on TM cells (), Cyto D blocked in a dose-dependent manner up to 80% of HSV-1 entry into nectin-1- CHO (or HVEM-CHO) and CF (). The latter expresses HVEM and 3-S HS, but not nectin-1 (unpublished data). Likewise, Lat B in a dose-dependent manner blocked up to 60% of HSV-1 entry into nectin-1-CHO and CF, but did not affect entry into TM cells (). To further demonstrate the significance of actin network in HSV-1 entry, cells were treated with Cyto D or Lat B either after (post-treatment) or before (pre-treatment) infection with HSV-1. It was postulated that pre-treatment would have more negative effect on entry than post-treatment, provided the actin-based protrusions (such as filopodia) played a role in the virus attachment to cells as well. After allowing for attachment and internalization of virions in both cases, cell surface virions that failed to infect were removed using low pH (citrate buffer pH 3.0) (Fig. S3). The internalization of the virus HSV-1 (K26GFP) was quantitated by measuring fluorescence intensity. There was ∼60–75% inhibition of internalization of HSV-1 in CF and nectin-1-CHO cells in post-treatment and up to 90% in pre-treatment cells (). TM cells did not exhibit significant inhibition in either case. Corresponding effects on phagocytosis-like uptake paralleled that of internalization (). It should, however, be noted that experiments using Cyto D and Lat B are suggestive, but not fully conclusive, of the role of actin network and phagocytosis-like uptake of HSV-1 because the drugs may have other effects not fully accounted for by the experiments shown here. We also examined the pH dependence of this mode of entry. have previously found that HSV-1 entry into nectin-1-CHO or HVEM-CHO is pH dependent. Thus, effects of lysosomotropic agents capable of interfering with vesicular acidification were tested. These included bafilomycin A1 (BFLA-1) (; ; ), chloroquine, and ammonium chloride (). Monolayer cultures of nectin-1-CHO or HVEM-CHO, CF, and TM cells were pretreated with BFLA-1 or chloroquine. In addition, ammonium chloride was applied to CF and TM cells. There was a strong dosage-dependent inhibition of HSV-1 entry by these drugs in nectin-1-CHO () and CF (); the inset shows the inhibitory effects of intermediate concentrations of chloroquine. Noticeably, entry into TM cells was inhibited by chloroquine, but not by BFLA-1 or ammonium chloride (). Collectively, the inhibition of acidification of vesicles severely retarded viral entry into CF and nectin-1-CHO, but not into TM cells. The gD receptor (nectin-1 or HVEM) is essential for HSV-1 entry (). To understand its role in phagocytosis-like uptake mode, we hypothesized that it should be present in the vesicles to facilitate viral capsid penetration into the cytosol, for which it is already known to be required (). To trace receptor trafficking we used a chimera of full-length nectin-1 and enhanced GFP (nectin-1-EGFP). This chimera is a functional receptor for HSV-1, which maintains full entry and cell-to-cell fusion activities (unpublished data). CHO cells transiently transfected with the chimera (nectin-1-EGFP-CHO) were briefly serum starved to eliminate endogenous transferrin and subsequently replaced with an exogenous Texas red conjugate (Texas red-transferrin), a molecule recycled via early endosomal vesicles including phagosomes (; ). Incorporation of nectin-1 into vesicles was traced by sequential CM to eliminate the “cross talk” or leakage between green and red, allowing for a factual colocalization. In mock-infected nectin-1-EGFP-CHO (), nectin-1-EGFP (green) and Texas red-transferrin–positive vesicles (red) were noted throughout the cells, but the nectin-1 and vesicles appeared at distinct diffuse areas with little colocalization. However, when nectin-1-EGFP-CHO cells were infected with purified HSV-1 (), large clusters of nectin-1 (green) and Texas red-transferrin–positive vesicles were observed (red). Their colocalization (yellow) was indicative of the association of nectin-1 with the vesicles. To verify the presence of the virions in the vesicles and to evaluate the stage at which virions transit from vesicles to release their capsid into the cytosol, antibodies to early endosome antigen 1 (EEA1) (marker for early stage endosomes and/or phagosomes) (, ), transferrin receptor (TfR) (marker for recycling endosomes), and LAMP-2 (marker for lysosomes) were independently used to stain infected (or mock-infected control) nectin-1-CHO or CHO-WT cells. Entry was with cold bound K26GFP incubated in media at 37°C over a time course ranging from 0 to 90 min. Significant numbers of virions (green) were found to localize in vesicles that mimic early stage endosomes (red) at time points from 5 to 60 min, with a peak seen at 30 min (). This time course corresponded well with the HSV-1 internalization observed while estimating optimal phagocytosis-like uptake (unpublished data). It appears that the virus stabilizes these early stage vesicles to facilitate its transit into the cytosol. Little colocalization of virus with both TfR () and LAMP-2 (unpublished data) was observed, suggesting that neither was significantly involved in virion trafficking. Plasma membrane protrusions that are formed in response to HSV-1 entry may be triggered by upstream signaling. Potential candidates may include tyrosine kinases () and lipid kinases. Pharmacological agents such as genistein (), a tyrosine kinase inhibitor, and wortmannin, a specific PI3-kinase inhibitor (), were used to assess the requirements for signaling. Both genistein and wortmannin adversely affected entry of HSV-1 into CF, nectin-1-CHO, and TM cells. However, the inhibitory effects were more pronounced in CF and nectin-1-CHO than in TM cells. Genistein at the higher dosage of 400 μM was more effective (Fig. S4 A, available at ). It might be that the less potent effect of wortmannin (∼50–60%) at the higher dosage of 100 nM was because it acted downstream (Fig. S4 A). Therefore, we performed dose-response entry experiments using nectin-1-CHO cells and found that genistein (Fig. S4 B) but not wortmannin (Fig. S4 C) induced a dose-dependent inhibition that significantly blocked HSV-1 entry. In further experiments, genistein blocked cointernalization of K26GFP and red beads but wortmannin failed to do so (unpublished data). Tyrosine kinase activation may play a role in HSV-1 uptake into CF and nectin-1- CHO cells. We next examined the role of Rho family GTPases that are known to regulate the actin cytoskeleton rearrangement that drives the process of phagocytosis. RhoA is required for complement-mediated phagocytosis and Rac and/or Cdc42 typically are implicated in triggered phagocytosis (). Rho alternates between the active GTP-bound and inactive GDP-bound state. Using the substrates GDP (negative control), GTPγS (positive control), and purified HSV-1 (potential activator of Rho) and pull-down assay we found that RhoA was activated in infected CF and nectin-1-CHO cells (). A time-course study using nectin-1-CHO cells indicated that the RhoA activation was sustained for 30 min (). Incidentally, this time frame matches well with the localization of the virions within early stage vesicles (). Cdc42 activation was not sustained. This activation dissipated in nectin-1-CHO cells within roughly 5 min of virus exposure (). Some of the plasma membrane protrusions were filamentous and rarely branched suggesting transitory filopodia formation, potentially to facilitate virus attachment. Similarly, no activation of Rac1 was detected (). These data prompted us to look more closely at changes in actin cytoskeleton upon viral exposure. Stress fibers became more prominent with entry of nectin-1-CHO cells, peaking at 30 min (Fig. S5, available at ). This time period, matching with that observed in RhoA activation and virion trafficking experiments, may be critical for virions transit via vesicles. The study presented herein attempts to shed light on many important but yet unclear aspects of a recently described endocytic route of HSV-1 entry (). Our work began with a goal to examine the ability of cultured human CF to support a productive HSV-1 entry, but it ended up with systematic exploration of the endocytic mode of entry. This cell type–dependent viral uptake was unexpectedly found to share many features of professional phagocytosis, hence the name “phagocytosis-like.” It likely requires actin rearrangement and dynamin assembly and is regulated by signaling pathways that apparently involve action of RhoA GTPase and may also involve tyrosine kinases. This pH-dependent and clathrin-independent viral uptake mechanism is marked by generation of cell surface protrusions in response to HSV-1 entry and clustering of gD receptors in large phagosome-like vesicles. Consistent with the notion that HSV-1 entry could occur by a phagocytosis-like uptake, it was found that the virus can significantly induce phagocytosis of bioparticles and HSV-1 virions cointernalized with phagocytic tracers. Induction of phagocytosis by HSV-1 might seem unexpected because it is generally believed that phagocytosis could not be induced by particles smaller than 0.5 μm in diameter (; ). The enveloped HSV-1 is only ∼0.2 μm. Yet evidence does exist to the contrary that phagocytosis could ensue for tracers as small as 0.13 μm (). Actually, it has been reported that granulocytes take up HSV via complement-mediated phagocytosis (). Also in the past, phagocytosis has been implicated in entry of viruses including many paramyxoviruses smaller in size than HSV-1 (). In addition, certain forms of vaccinia virus use an entry pathway that mimics the initial stages of phagocytosis (). Similarly, the induction of RhoA GTPase during HSV-1 entry () but not Cdc42 or Rac1 () would support the notion that the mechanism of entry into nectin-1-CHO and CF could be analogous to complement-activated phagocytosis (). It is perhaps logical to suggest that the molecular signaling mechanisms, not necessarily size, invokes phagocytosis. Our findings supported by some additional ultrastructural images allow us to propose a model (). In brief, the infection of nectin-1-CHO and CF cells begins with the virions making initial contact with the cell surface by associating with plasma membrane protrusions (may include filopodia) (, arrows, top left and right panels). As the virions surf to concentrate at the cell body, internalization may occur by phagocytosis-like uptake followed by transport in early stage vesicles that are likely phagosomes, as suggested by their relatively large size and apparent colocalization of HSV-1 with phagocytic tracers (). Presence of nectin-1 (or HVEM) in the vesicles () could provide the opportunity for the virions to fuse their envelopes with the vesicular membrane (, middle top and bottom panels). Consistent with a possibility that low pH environment found in vesicles can, in fact, enhance the fusion potential of HSV-1 virions, exposure of gB and gD to acidic pH has been shown to improve their membrane fusion potential () and similarly, we have also found that low pH shock enhances HSV-1 mediated cell-to-cell fusion (unpublished data). The final stage is the release of the naked nucleocapsid (arrowhead) from the vesicle into the cytosol (cy) proximal to the nucleus (nu) (, bottom) for replication in the nucleus. Our model raises an obvious question related to the initial molecular events that trigger phagocytosis-like uptake. Specific virus–cell interactions are likely to be important because HS-mediated attachment of virus to cells is needed for maximal induction of phagocytosis-like uptake. Association with HS, however, is likely not the trigger, as cells lacking HSPG can still induce some basal level of phagocytosis (). Similarly, gD receptors (nectin-1, HVEM) may not be directly involved because they appear to be essential for entry into all cells, irrespective of the mode of viral entry (). Based on the current theory that RhoA is usually activated in complement mediated phagocytosis (), it is possible that either an unknown complement receptor homologue, or a CF and CHO analogue of it, could trigger a similar process. The search for the trigger is further complicated by the existence of a dozen or so different HSV-1 envelope glycoproteins with little known functions. One or perhaps more of these HSV-1 envelope glycoproteins are likely to be needed for the triggering mechanism. Precedent exists, at least for some intracellular bacteria, that interactions with integrins (and potentially HS) could signal internalization by phagocytosis after binding of pathogens (; ; ). Interestingly, a glycoprotein (gH) essential for HSV-1 entry has recently been reported to interact in vitro with αvβ3 integrins, although the effect of this interaction on entry is still unclear (). Moreover, new findings including ours indicate that the mode of entry could vary with cell types (). Conceivably, the functions of various HSV-1 glycoproteins and their corresponding receptors could vary with cell types as well. For instance, gD, the most studied glycoprotein, has been reported to bind multiple cellular molecules including three very diverse cellular receptors (). Given the multi-protein binding ability of HSV-1 envelope proteins and the potential for cell type–dependent interactions, it is evident that extensive amount of work needs to be followed to identify the initial players and dissect their functions in the triggering mechanism(s). In any case, our current study provides new information on the existence of a novel pathway with implications for virus entry into both professional and nonprofessional phagocytes. It also provides new information on the poorly understood infection of the stroma and is likely to help develop new therapies for control of potentially blinding HSK. Plasmids used include gD receptors pBG38 (nectin-1) and pBEC10 (HVEM); control vector pcDNA3 (Invitrogen); pNec1-EGFP or pHVEM-EGFP expressing full-length nectin-1 or HVEM fused with EGFP (pEGFP-N1 vector; BD Biosciences) and rat dynamin2 (wild-type, WT-dyn2 and dominant-negative mutant, K44A-dyn2; from Mark McNiven, Mayo Clinic, Rochester, NY) fused with EGFP (CLONTECH Laboratories, Inc.). Eps15 mutants: DIIIΔ2 (control), DIII and EH29 corresponding to EΔ95/295 all subcloned in EGFP-C2 were kind gifts from Alexandre Benmerah (Universitė Paris, Paris, France). Patricia G. Spear (Northwestern University, Chicago, IL) provided wild-type CHO-WT and CHO-745 cells (). All CHO cell lines were grown in Ham's F12 (Invitrogen) supplemented with 10% fetal bovine serum (FBS). Stable nectin-1 (or HVEM)-CHO or control empty vector pcDNA3 cell lines were selected in medium containing geneticin (500 μg/ml) (Cellgro). For primary cultures of CF and TM cells, human eye tissues were obtained from the Illinos Eye Bank and cultured as described previously (). Recombinant viruses HSV-1(KOS) gL86 and K26GFP (gifts from P. Desai, The Johns Hopkins University, Baltimore, MD) were amplified and quantitated as described elsewhere (; ). HHV-8 virions were a kind gift from J. Vieira (University of Washington, Seattle, WA). A syn mutant strain, HSV-1(KOS) 804 was harvested from Vero cells. HSV-1(KOS) gL86 was purified by a modified sucrose gradient centrifugation method (). Experiments discussed in this manuscript were performed using HSV-1 titers within the linear range of 0–100 PFU/cell (). Entry assays were based on quantitation of β-gal expressed from HSV-1 genome by either insoluble 5-bromo-4-chloro-3-indoyl-β--galactoside (X-Gal; Invitrogen), 1 mg/ml or soluble O-nitro-phenyl β--galactopyranoside (ONPG; Pierce Biotechnology), 3 mg/ml. Alternatively, virus entry was determined by fluorescence readout assay. Stably transfected nectin-1 (or HVEM)-CHO cells, CF, and TM cells were treated with medium containing Cyto D (0.5 or 1.0 μg/ml), Lat B (0.25 or 2.5μM), BFLA-1 (0.01 or 0.1 μM), chloroquine (1.0 or 100 μM), genistein (400 μM) (Sigma-Aldrich), or wortmannin (100 nM; Calbiochem). CF and TM cells were in addition treated with ammonium chloride (10 or 50 mM; Fisher cientific). Infected (HSV-1 at 50 PFU/cell used for all experiments unless otherwise stated) and mock-infected cells were grown for 24 h at 37°C in 96-well culture dishes. Dosages of pharmacological agents for treatments were based on prior reports (; ; ; ). Real-time quantitative PCR was performed in 25-μl tubes with DNA prepared according to the manufacturer's instruction (QIAamp DNA Mini kit) using a Smart Cycler System (Cepheid). Primers, forward; AAGACCTTCCGGTCCTG and backward; TCCAACACGGCGTAGTA for the HSV-1 (KOS) glycoprotein D target were designed using Clone Manager 6 program (Sci Ed Central). The reaction was performed for 50 cycles under the following conditions: denaturation at 95°C for 30 s, annealing at 58°C for 10 s and extension at 72°C for 16 s. After amplification, one cycle of melting curve from 60 to 95°C by a transition rate of 0.2°C/s with continuous detection of fluorescence, was performed. The vibrant phagocytosis kit (Molecular Probes) developed with an adherent murine macrophage cell line (J774) has been adapted to evaluate phagocytosis in other adherent cell types (; ). Subcultured nectin-1-CHO, HVEM-CHO, and CHO-745 cells were adjusted to 10 cells/ml, aliquoted at 0 (negative control), 100 μl/microwell for positive and test into Microtest 96-well assay plates (BD Falcon) for settlement and adherence overnight at 37°C. The cells were washed and challenged with purified HSV-1 or purified K26GFP for test and PBS-G-CS for negative and positive controls. All cells were incubated at 4°C for 1 h, washed, and with warm medium further incubated at 37°C for 1 h. In experiments using HSV-1, all the microwells were in addition challenged for 2 h with reconstituted suspension of fluorescein-labeled K-12 bioparticles. For colocalization experiments, K26GFP and red fluorescent beads (FluoSpheres; Molecular Probes) were used. The effect of dynamin was evaluated by transfecting cells with WT-dyn2 or K44A-dyn2. For drug inhibition of internalization and phagocytosis, cells were first cold-bound with purified K26GFP or HSV-1 for 1 h before the 2-h incubation with virus in the presence of Cyto D (1 μg/ml) or Lat B (2.5 μM) at 37°C for post-treatment, or the cells were first incubated for 1 h with the drugs for pre-treatment. Virus that were bound to the cell surface were stripped with citrate buffer pH 3.0 for 1 min. Virus entry or phagocytic activity measured as relative fluorescence units (RFU) per treatment were determined by a fluorescence quenching assay with trypan blue using GENios Pro plate reader (TECAN) at 480-nm excitation and 520-nm emission spectrum. Measurements of five replicates of negative control, positive control, and test samples were performed. Data were expressed as mean ± SD. Nectin-1-CHO, CF, and TM cells cultured in Lab-Tek chamber slides (2.5 × 10 cell/tray) and in anopore wells (Nalge Nunc) (2.5 × 10 cells/well) were infected with purified HSV-1 at 10–100 PFU/cell for 10–90 min. EM was performed using standard methods (). For TEM (JEM-1220; JEOL USA Inc.), images were captured at 1,000–600,000×, point-0.36 nm (3.6A) and lattice-0.2 nm (2A) at temperature, ACC voltage 40–120 kV using a Gatan camera (Digital CCD; Gatan, Inc.) and using Gatan Digital Micrograph (DM) v2.5 acquisition software. For SEM, Field Emission SEM (JMS-6320F; JEOL Inc. USA) was used to capture images at 25–2,000× in LM and 500–650,000× in HR modes. Images were resolved at 15 kV 1.2 nm (12A) and 1 kV 2.5 nm (25A) at room temperature ACC 0.5–30 kV using camera system ARC64 and Arch software (JEOL Inc. USA). Alternatively, for immunogold the infected cells were processed (). Specimens mounted on 150 mesh nickel-formvar–coated grids were incubated in 1% BSA and further incubated at room temperature for 3 h with polyclonal rabbit anti-HSV-1 gD, R7 (1:200), or anti gH-gL, R137 (1:200). Sections were subsequently incubated at room temperature for 1 h with 12-nm colloidal gold-conjugated goat anti–rabbit IgG (H+L) (1:30; Jackson ImmunoResearch Laboratories). The grids were also counterstained with uranyl acetate and examined in TEM as previously described. Total cell lysates of equal amount of HSV-1 (50 PFU/cell for 15 min) or GDP (negative control) and GTPγS (positive control) as well as time courses for RhoA, Cdc42, and Rac) were spun in a microfuge at 14,000 rpm for 15 min and proteinase inhibitors were added to the supernatants. Rhotekin- RBD-GST (RhoA) and PAK-PBD-GST (Cdc42/Rac) beads (Cytoskeleton Inc.) were incubated with the supernatants for 1 h at 4°C on a rotator to bind the active proteins. The bound proteins were resolved on SDS-PAGE gels, transferred to nitrocellulose membranes, and immunoblotted with antibodies; RhoA, Cdc42, Rac 1:200 (Santa Cruz Biotechnology) and secondary at 1:10,000 HRP anti–mouse IgG (for RhoA) and HRP anti–rabbit IgG (for Cdc42 and Rac) (Jackson ImmunoResearch Laboratories). Immunoreactive bands were developed by West Pico substrate for enhanced chemiluminescence reactions (Pierce Biotechnology) and imaged on KODAK Biomax Mr film. Fig. S1 shows plaque formation on CF cells after infection with HSV-1(KOS-804). Fig. S2 shows large sections of CF, nectin-1-CHO, and TM cells after infection with HSV-1. Fig. S3 shows inactivation of uninternalized HSV-1 after treatment with citrate buffer (pH 3.0). Fig. S4 shows the effects of tyrosine kinase and PI3K inhibitors on HSV-1 entry. Fig. S5 shows phalloidin staining of actin of nectin-1-CHO and CHO-WT cells demonstrating HSV-1 induced cytoskeletal changes. Online supplemental material available at .